[go: up one dir, main page]

CN120705409A - Information recommendation method, electronic device, and vehicle - Google Patents

Information recommendation method, electronic device, and vehicle

Info

Publication number
CN120705409A
CN120705409A CN202510880136.4A CN202510880136A CN120705409A CN 120705409 A CN120705409 A CN 120705409A CN 202510880136 A CN202510880136 A CN 202510880136A CN 120705409 A CN120705409 A CN 120705409A
Authority
CN
China
Prior art keywords
user
target
content
intention
contents
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202510880136.4A
Other languages
Chinese (zh)
Inventor
严飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chengdu Great Wall Motors R&d Co ltd
Original Assignee
Chengdu Great Wall Motors R&d Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Chengdu Great Wall Motors R&d Co ltd filed Critical Chengdu Great Wall Motors R&d Co ltd
Priority to CN202510880136.4A priority Critical patent/CN120705409A/en
Publication of CN120705409A publication Critical patent/CN120705409A/en
Pending legal-status Critical Current

Links

Landscapes

  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The application relates to the field of data processing, and particularly provides an information recommendation method, electronic equipment and a vehicle. The method comprises the steps of receiving user demand information, sending the user demand information to a pre-trained intention complexity classification model to obtain a complexity classification result, determining a corresponding target intention recognition model according to the complexity classification result, analyzing based on user intention information by using the target intention recognition model to determine target intention, selecting a target recommendation model corresponding to the target intention, determining target recommendation content by using the target recommendation model, and feeding back the target recommendation content to a user. According to the complexity of the user demand information, the user demand can be accurately understood, the accurate target intention is further determined, the target recommendation model corresponding to the target intention is utilized to capture the user demand and conduct intelligent analysis, and the most accurate target recommendation content is determined.

Description

Information recommendation method, electronic equipment and vehicle
Technical Field
The present application relates to the field of data processing technologies, and in particular, to an information recommendation method, an electronic device, and a vehicle.
Background
Currently, some information recommendation software directly recommends information according to requirements (such as voice requirements or text requirements) expressed by users.
However, the current information recommendation software determines corresponding keywords according to the requirements expressed by the user, and then intelligently makes corresponding information recommendation according to the keywords. Such information recommendation made according to the keywords often cannot meet the needs of the user, so that the experience effect of the user is poor.
Disclosure of Invention
Accordingly, the present application is directed to an information recommendation method, an electronic device, and a vehicle, which are used for solving the problem that the current method of directly recommending information according to determined keywords cannot meet the user requirements.
Based on the above object, the present application provides an information recommendation method, comprising:
receiving user demand information, and sending the user demand information to an intention complexity classification model obtained by training in advance to obtain a complexity classification result;
Determining a corresponding target intention recognition model according to the complexity classification result, and analyzing the target intention recognition model based on user intention information to determine target intention, wherein the target intention recognition model is a model trained in advance;
Selecting a target recommendation model corresponding to the target intention, determining target recommendation content by using the target recommendation model, and feeding back the target recommendation content to a user, wherein the recommendation model corresponding to each intention is stored.
Based on the same inventive concept, the application also provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable by the processor, the processor implementing the method as described above when executing the computer program.
Based on the same inventive concept, the application also provides a vehicle comprising the electronic device as described above.
According to the information recommendation method, the electronic equipment and the vehicle, the complexity classification model which is obtained through training in advance can be utilized to conduct identification processing on the user demand information to determine the complexity classification result corresponding to the user demand information, the complexity classification result can represent the demand complexity degree of the user demand information, then the target intention recognition model corresponding to the complexity classification result can be determined based on the complexity classification result, the target intention recognition model can accurately conduct intention recognition analysis on the user intention information corresponding to the complexity classification result to obtain accurate target intention, then the target recommendation model corresponding to the target intention is selected, the target recommendation model can process content recommendation in the field corresponding to the target intention, and therefore the target recommendation content can be determined according to user demands from the field corresponding to the target intention by utilizing the target recommendation model, the obtained target recommendation content is accurate, the actual requirement of a user can be met, and the user experience is improved.
Drawings
FIG. 1 is a flow chart of an information recommendation method according to an embodiment of the present application;
FIG. 2 is a flowchart of an information recommendation method according to another embodiment of the present application;
FIG. 3 is a block diagram illustrating an information recommendation apparatus according to an embodiment of the present application;
Fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
When a control system (for example, a car machine system) recommends information according to various demands of a user, the following problems generally exist:
first, the demand understanding is imprecise
For some complex needs of users, it is difficult to accurately understand the true intention. For example, a user expresses "want to listen to some songs suitable for long distance driving and refreshing", and a traditional control system may only recommend according to the keywords of "songs", and cannot understand the deep requirements such as "long distance driving" and "refreshing", so that the recommendation result is inconsistent with the user's expectations.
Secondly, poor scene adaptation capability
The driving scene where the user is can not be effectively perceived, and the targeted recommendation is performed by combining the scenes. For example, when the user feels tired during long distance driving, the system still recommends a large amount of driving-related content without considering the scene that the user needs to rest and relax, or fails to timely provide local travel strategies, food recommends and other services after the user arrives at a travel destination.
Thirdly, the interactive experience is poor
The general recommendation function has single interaction mode, is mostly key operation or simple voice instruction response, does not support natural and smooth multi-round dialogue, and is complex in operation when expressing complex demands, and driving safety and experience are affected.
Embodiments of the present application are described in detail below with reference to the accompanying drawings.
The information recommendation method provided by the embodiment of the application is applied to a control system (for example, a vehicle control system) as shown in fig. 1, and comprises steps 101-103.
Step 101, receiving user demand information, and sending the user demand information to an intention complexity classification model obtained through training in advance to obtain a complexity classification result.
In specific implementation, the user demand information is voice information, text information, gesture information or eye movement information. The method comprises the steps of converting voice information into characters if the voice information is, converting the meaning expressed by gesture information into characters if the gesture information is, and converting the meaning expressed by eye movement information into characters if the eye movement information is.
Therefore, characters corresponding to the user demand information can be input into the intention complexity classification model to classify the intention complexity, and a complexity classification result is obtained. The complexity classification result can be a plurality of categories which are classified according to the complexity, and the number of the specific classification categories can be set according to actual needs. Preferably, the two classes are divided into a simple class or a complex class, respectively.
The corresponding intention complexity classification model is a classification model, a classification model is built in advance, user requirement information of a simple class (for example, the user requirement information is that ' I want to eat Sichuan dish in particular at present ', the intention is clear, the user requirement information is easy to identify, the sign is ' 0 '), user requirement information of a complex class (for example, the user requirement information is that ' we finally arrive soon to Nanjing ' and the intention is fuzzy, the user requirement information is not easy to identify, the sign is1 ') is collected, the classification model is used as a first training sample, the pre-built classification model is subjected to supervised training, a loss function is determined according to the difference between a training result output during training and the complexity classification of the sign (for example, the simple class or the complex class), and parameter adjustment is carried out on the classification model according to the loss function. And determining that all the first training samples are completely trained, and taking the final classification model as an intention complexity classification model. The intention complexity classification model can classify the complexity of various user demand information. The corresponding simple and complex classes have different marking signs, and can be specifically set according to actual needs.
Step 102, determining a corresponding target intention recognition model according to the complexity classification result, and analyzing based on user intention information by utilizing the target intention recognition model to determine target intention, wherein the target intention recognition model is a model obtained by training in advance.
In the specific implementation, intention processing models corresponding to various complexity classifications are stored in the control system in advance, and the number of the stored intention processing models is matched with the number of the complexity classifications. Therefore, the target intention recognition model corresponding to the complexity classification result can be utilized to process the user intention information corresponding to the complexity, and more targeted intention recognition can be performed, so that the determined target intention is more accurate.
Each intention processing model is a second training sample formed by classifying user intention information according to corresponding complexity and marking corresponding intention for each user intention information, a pre-constructed network model (such as LLM, large Language Model, a large language model, or a two-class model) is subjected to supervised training, a loss function is determined according to the difference between a training result output during training and the marked intention, and parameter adjustment is performed on the network model according to the loss function. And determining that all second training samples are completely trained, and taking the final network model as a model for intention recognition. The intent recognition model is capable of performing an intent recognition model on the complexity-classified user intent information. For example, the intents include music intents, movie intents, dining intents, scenic spot intents, and the like.
And step 103, selecting a target recommendation model corresponding to the target intention, determining target recommendation contents by using the target recommendation model, and feeding back the target recommendation contents to a user.
In specific implementation, various recommendation models corresponding to intentions, such as a music recommendation model corresponding to music intentions, a movie recommendation model corresponding to movie intentions, a catering recommendation model corresponding to catering intentions and a scenic spot recommendation model corresponding to scenic spot intentions, are stored in advance. Different recommendation models can recommend recommended contents of the category corresponding to the intention for the user according to the user requirements.
The recommendation models may be function modules capable of recommending corresponding content data according to interest preferences of the user.
The recommendation model is obtained by training a neural network model. Specifically, for each intention, specific information (such as an interaction problem corresponding to the intention) of the intention is collected first, and corresponding recommended content is marked for each specific information, so as to form a third training sample. And performing supervised training on the pre-constructed neural network model by using a third training sample, determining a loss function according to the difference between the training result output during training and the recommended content of the mark, and performing parameter adjustment on the neural network model according to the loss function. Determining that all third training samples are completely trained, and taking the final neural network model as a recommended model of the intention. The intent is associated with the corresponding recommendation model and stored in a database of the control system.
After the target intention recognition model outputs the target intention, the target recommendation model corresponding to the target intention is called, the target recommendation model is utilized to determine corresponding target recommendation content according to the requirement characteristics and the preference characteristics of the user, and the target recommendation content is fed back to the user. The specific feedback mode can be voice broadcasting and/or screen display, and preferably, the voice broadcasting and the screen display are fed back simultaneously. The corresponding obtained target recommended content can be one or more, so that the user can determine whether the at least one target recommended content has the required content of the user, and if so, the user can trigger the required content and execute the corresponding function.
The number of the target recommended contents cannot be too large (for example, cannot exceed 10), so that the target recommended contents are prevented from being too large, and the user cannot quickly select the required contents due to too long time for confirming the target recommended contents.
According to the scheme, the complexity classification result corresponding to the user demand information can be determined by utilizing the intention complexity classification model obtained through training in advance, the complexity classification result can represent the demand complexity degree of the user demand information, then the target intention recognition model corresponding to the complexity classification result can be determined based on the complexity classification result, the intention recognition model can accurately recognize and analyze the user intention information corresponding to the complexity classification result to obtain accurate target intention, then the target recommendation model corresponding to the target intention is selected, the target recommendation model can process content recommendation in the field corresponding to the target intention, and therefore target recommendation content can be determined from the field corresponding to the target intention by utilizing the target recommendation model, the obtained target recommendation content is accurate according to the user demand, the actual requirement of a user can be met, and the user experience is improved.
In some embodiments, the complexity classification result is a simple class or a complex class, and step 102 includes:
And step A, in response to determining that the complexity classification result is a complex class, determining that a corresponding target intention recognition model is a first intention recognition model, receiving a multi-round dialogue result with a user, recognizing the multi-round dialogue result by using the first intention recognition model, and determining the target intention.
In particular, for a user demand with complex intention, a first intention recognition model obtained by training a large language model (LLM, large Language Model) is required to determine the target intention of the user demand in a multi-turn dialogue manner.
The method comprises the steps of collecting various dialogue contents in advance, marking corresponding intentions for the dialogue contents to obtain a fourth training sample, training a large language model by using the fourth training sample, outputting training intention results by using the large language model, comparing the training intention results with marked intentions, if the training intention results are the same, continuing to train by using the next fourth training sample, and if the training intention results are different, carrying out parameter adjustment (for example, carrying out parameter adjustment by back propagation) on the large language model, so that after the large language model is subjected to parameter adjustment, the training intention results which are the same as the marked intentions can be output, and then continuing to train by using the next fourth training sample. After the fourth training sample is determined to be completely trained, the first intention recognition model can be obtained.
For example, the target intent is determined via the following dialog:
The user is immediately away from the hotel.
A first intention recognition model, do a respected owner hear you get to the hotel immediately, need you recommend nearby places to play?
The user is that me is now starved, recommending me a nearby restaurant bar.
The first intention recognition model is good, and recommends some restaurants to you immediately.
The first intent recognition model analyzes the dialog results to determine that the target intent is a dining intent.
Or step B, in response to determining that the complexity classification result is a simple class, determining that a corresponding target intention recognition model is a second intention recognition model, and recognizing the user intention information by using the second intention recognition model to determine a target intention.
In specific implementation, aiming at the user requirement with simple intention, the second intention recognition model obtained by training the relatively simple classification model is used for classifying and recognizing the user intention information to determine the target intention.
Wherein, a large amount of user intention information (e.g. query) is collected in advance, and the actual intention is used to label the user intention information to obtain a fifth training sample. And then training the two classification models by using a fifth training sample to obtain a second intention recognition model. The classification level of the classification model is determined in particular according to the number of corresponding intent classifications.
For example, the number of intention classifications is 4, namely, music intention, movie intention, catering intention and scenic spot intention, and the classification levels in the corresponding classification models are 4 layers, and each classification level is used for identifying the corresponding intention class. The 4-layer classification hierarchy specifically comprises a music classification hierarchy for identifying music intents, a film and television classification hierarchy for identifying film and television intents, a catering classification hierarchy for identifying catering intents and a scenic spot classification hierarchy for identifying scenic spot intents.
According to the scheme, a more targeted target intention recognition model can be determined according to the complexity classification result, so that the complex class can use a first intention recognition model capable of processing complex intention, a multi-round dialogue result is obtained in a multi-round dialogue mode, the multi-round dialogue result is further recognized, the target intention is accurately determined, if the complex class is simple, a second intention recognition model capable of processing simple intention can be used, the target intention is determined in a multi-round two-classification mode, and the accuracy of the determined target intention is ensured.
In some embodiments, in step a, the receiving the multi-round dialog result with the user, identifying the multi-round dialog result using the first intention identification model, and determining the target intention includes:
And step A1, analyzing the user intention information by using the first intention recognition model to obtain a first analysis result and outputting the first analysis result so that a user can feed back according to the output first analysis result.
In specific implementation, the first intention recognition model is utilized to perform analysis processing according to the user intention information, so that a first analysis result of rough intention classification is determined and output and fed back to the user (particularly, the user can be fed back in a voice and/or screen display mode). And the user can make answer feedback according to the first analysis result.
And step A2, receiving user feedback information, analyzing the user feedback information by using the first intention recognition model to obtain the type of the undetermined intention, and outputting the type of the undetermined intention so as to ensure that a user can determine whether the type of the undetermined intention which is output is correct or not.
In specific implementation, the first intention recognition model analyzes the type of the pending intention of the user according to the feedback information of the user, and the first intention recognition model sends the type of the pending intention to the output module so as to feed back the type of the pending intention to the user (particularly, the type of the pending intention can be fed back to the user in a voice and/or screen display mode). The user determines whether the pending intent type is correct.
And step A3, in response to receiving an undetermined intention type error fed back by a user, iteratively executing a correction process of the undetermined intention type by utilizing the first intention recognition model until the corrected intention type is correct, and taking the corrected intention type as the target intention.
In specific implementation, if the user feedback received by the first intention recognition model is an error of the pending intention type, a process of correcting the pending intention type is executed, specifically:
And C, re-determining the corrected analysis result, outputting feedback to the user, receiving new feedback information of the user, repeating the process of the step A2, determining the corrected intention type, and outputting feedback to the user.
And continuously and iteratively executing the correction process until the user feedback received by the first intention recognition model is correct in corrected intention type, stopping iteration, and taking the corrected intention type as a target intention.
Or step A4, responding to the received user feedback that the type of the undetermined intention is correct, and taking the type of the undetermined intention as the target intention.
In the implementation, if the user feedback received by the first intention recognition model is that the type of the pending intention is correct, the modification process is not needed to be executed, and the type of the pending intention is directly taken as the target intention.
For example, the first intent recognition model determines that the first analysis result is "the owner wants to eat" based on the user intent information of "end-to-end" and issues the first analysis result of "whether the owner wants to determine the restaurant of the destination.
The user makes feedback that 'I do not want to go to the restaurant, I want to go to nearby scenic spots to browse' according to the query.
And the first intention recognition model is used for determining that the type of the undetermined intention is 'scenic spot intention' according to 'I do not want to go to a restaurant, I want to visit nearby scenic spots' fed back by the user, and feeding the undetermined intention type back to the user.
And determining that the pending intention type is correct.
And the first intention recognition model takes the undetermined intention type as a target intention, and sends the target intention to a target recommendation model corresponding to the target intention.
Or alternatively
Determining the pending intent type error.
The first intention recognition model sends out the corrected analysis result of "whether the owner wants to listen to music.
I want to listen to the nostalgic music.
And a first intention recognition model for determining that the undetermined intention type is music intention and feeding the music intention back to the user.
And determining that the pending intention type is correct.
And the first intention recognition model takes the undetermined intention type as a target intention, and sends the target intention to a target recommendation model corresponding to the target intention.
According to the scheme, even if the intention of the user is complex, the target intention which is most in line with the user requirement can be determined by using the first intention recognition model through the polling interaction mode of the user, the obtained target intention is more accurate, and therefore, the target recommended content determined according to the target intention is more in line with the user requirement, and the user experience is improved.
In some embodiments, determining the target recommended content required by the user using the target recommendation model in step 103 includes:
Step 1031, obtaining behavior information of a user by using a target recommendation model, and determining a corresponding first preset number of contents to be recommended according to the behavior information of the user.
In implementation, the preference characteristics of the user can be determined according to the behavior information of the user, so as to determine a first predetermined number (for example, 100) of to-be-recommended contents which accord with the preference characteristics of the group user. And integrating the list of the first preset number of contents to be recommended according to the sequence to obtain a recall list.
The first predetermined number may be a fixed number or an unfixed number.
The method for determining the content to be recommended can be (1) determining the content to be recommended according to the cooperation of the user and the similar user, (2) determining the content to be recommended similar to the content according to the content corresponding to the behavior of the user, and (3) determining the interest tag according to the behavior of the user so as to find the content to be recommended.
And step 1032, utilizing the target recommendation model to sequence and adjust the first preset number of contents to be recommended according to the matching degree with the user requirement, and obtaining a sequencing result.
In specific implementation, the obtained first preset number of to-be-recommended contents are not accurately ordered, and cannot be determined to be more in accordance with the needs of the user, so that the matching degree of each to-be-recommended content and the needs of the user in the first preset number of to-be-recommended contents can be determined by using the target recommendation model, and then the first preset number of to-be-recommended contents are reordered according to the order of the matching degree from high to low, so that an ordering result is obtained.
And 1033, selecting a second preset number of contents to be recommended with highest matching degree from the sorting results by using a target recommendation model as the target recommended contents, wherein the second preset number is smaller than the first preset number.
In practice, a second predetermined number of values is preset, and generally the second predetermined number is smaller than the first predetermined number. The target recommendation model selects the first second preset number of contents to be recommended from the sorting result as target recommendation contents, and feeds the target recommendation contents back to the user for the user to select.
According to the scheme, the first preset number of the contents to be recommended can be initially selected according to the behavior information of the user, and the first preset number of the contents to be recommended are not accurately ordered, so that in order to conveniently screen out the target recommended contents which meet the requirements of the user, the target recommended contents are reordered according to the matching program with the requirements of the user, further, a more accurate ordering result is obtained, and therefore the target recommended contents selected from the first preset number of the contents to be recommended are more accurate.
In some embodiments, in step 1031, the determining a corresponding first predetermined number of contents to be recommended according to the behavior information of the user includes:
(1) And determining the content to be recommended according to the cooperation of the user and the similar user.
And step C1, determining similar users based on the behavior information of the users, wherein the similar users corresponding to each user are prestored, and the similar users are other users matched with the behavior information of the users.
In the implementation, the similarity between the user and other users can be calculated according to the behavior information of the user, the other users are ranked according to the similarity, and a third preset number of other users with high similarity are selected as similar users. Or selecting other users with similarity exceeding the similarity threshold as similar users. These similar users are then stored in association with the user's identity information (e.g., ID address) in a database. Thus, the corresponding similar users can be directly called according to the identity information of the users.
And step C2, acquiring the operation behaviors of the similar users in a preset time period, and determining a plurality of trigger contents according to the operation behaviors.
In particular, the operational behavior includes clicking, collecting or praying. It is then determined that the user has occurred the trigger content corresponding to the operational behavior within a predetermined period of time (e.g., one month or 10 days). For example, it is determined that the user clicks on favorite or praise music content for the last month.
And step C3, filtering the plurality of trigger contents, and screening a first preset number of contents to be recommended from the filtered plurality of trigger contents.
In the implementation, the above-mentioned obtained trigger contents may be repeated, so that the trigger contents are subjected to de-duplication filtering processing, and a plurality of filtered trigger contents are obtained. And sequencing the plurality of trigger contents according to the operation frequency of the operation behaviors, and selecting a first preset number of contents to be recommended.
According to the scheme, the trigger content corresponding to the operation behaviors of the similar users matched with the user can be determined, and the corresponding favorites are similar because the trigger content is similar to the user, so that the obtained trigger contents are the content which accords with the favorites of the user, and the first preset number of contents to be recommended are screened from the trigger contents and are the contents which accord with the requirements of the user. So that the target recommended content determined later therefrom is more accurate.
And/or, (2) determining the content to be recommended similar to the content according to the content corresponding to the behavior of the user.
And D1, determining triggering content according to the behavior information of the user.
In implementation, a behavior list corresponding to behavior information (such as clicking, collecting and praying) of the user is obtained, and corresponding trigger content is determined according to the behavior list of the user in a preset time period.
And D2, acquiring a plurality of similar contents corresponding to the trigger contents, wherein the similar contents corresponding to each trigger content are stored in advance.
In particular implementations, the similarity between the content and the content is pre-calculated, and a fourth predetermined number (e.g., the top 100 or the top 50 or the top 10, ordered from high to low according to the similarity) of similar content corresponding to each content is determined and stored in the database. Thus, the corresponding similar content can be searched from the database directly according to the trigger content.
If the trigger content obtained by the D1 is multiple, determining multiple similar content corresponding to each trigger content, and integrating the similar content.
And D3, filtering the similar contents, and screening a first preset number of contents to be recommended from the filtered similar contents.
In implementation, the above obtained similar contents may be repeated, so that the duplicate removal filtering process is performed on the similar contents, thereby obtaining filtered similar contents. And sequencing the plurality of similar contents according to the operation frequency of the operation behaviors, and selecting a first preset number of contents to be recommended.
According to the scheme, the trigger content in the preset time period can be determined according to the behavior of the user, and a plurality of similar contents corresponding to the trigger content can be selected from the similar contents according to the similar contents corresponding to the pre-stored contents. So that the target recommended content determined later therefrom is more accurate.
And/or (3) interest tags determined according to the behavior of the user, and then found content to be recommended.
And E1, determining interest tags according to the behavior information of the user.
And E2, searching a plurality of matching contents corresponding to the interest tag, filtering the plurality of matching contents, and screening a first preset number of contents to be recommended from the plurality of filtered matching contents.
In specific implementation, the interest tag of the user can be determined according to the behavior information (such as clicking, collecting or praying) of the user, so that corresponding matching contents can be searched directly according to the interest tag, and the matching contents can be repeated, so that duplicate removal filtering processing is performed on the matching contents to obtain filtered matching contents, and then the filtered matching contents are ranked according to heat, so that a first preset number of contents to be recommended are selected.
Through the scheme, the interest tags are directly determined according to the behavior information of the user, and the first preset number of contents to be recommended are screened out based on the interest tags, so that the operation is simple and quick, and the found contents to be recommended more accord with the interests of the user.
In a preferred embodiment, the content to be recommended obtained in the above 3 may be integrated, and the recall list of the final content to be recommended may be obtained after performing the deduplication process. This further ensures the accuracy of determining the content to be recommended.
In some embodiments, step 1032 comprises:
step 10321, obtaining feature data of the user, inputting the feature data of the user and a first predetermined number of contents to be recommended into a click rate prediction model, determining a predicted click rate of each content to be recommended, and taking the predicted click rate as a matching degree of the contents to be recommended and the user requirement.
In particular implementations, the characteristic data of the user includes the user's gender, age, health, interest preferences, and the like. And determining the content characteristics corresponding to each content to be recommended. And inputting the characteristic data of the user and the content characteristics of each content to be recommended into a click rate prediction model, analyzing the characteristic data of the user by using the click rate prediction model to determine the predicted click rate of the content to be recommended, and taking the predicted click rate as the matching degree of the content to be recommended and the user requirement. The predicted click rate is the probability that the user clicks the content to be recommended.
And step 10322, performing sorting adjustment on the first predetermined number of contents to be recommended according to the matching degree to obtain a sorting result.
In the specific implementation, the content to be recommended is ranked and adjusted according to the matching program, and the ranking is performed according to the sequence from high to low of the matching degree, so that a ranking result is obtained, and the second preset number of the content to be recommended before can be selected as target recommended content according to the ranking result.
According to the scheme, since the ordering of the first preset number of to-be-recommended contents is not accurate enough, in order to accurately order the to-be-recommended contents, a click rate prediction model obtained through training in advance is used, the predicted click rate of each to-be-recommended content is accurately determined according to the combination analysis of the characteristic data of the user and the content characteristics corresponding to each to-be-recommended content, so that the ordering result obtained by accurately ordering based on the predicted click rate as the matching degree meets the requirements of the user, and the second preset number of to-be-recommended contents before the second preset number of to-be-recommended contents are selected as target recommended contents.
In some embodiments, the training process of the click rate prediction model includes:
And F1, acquiring characteristic data, historical content and historical click data corresponding to the historical content of a user, combining the characteristic data and the historical content, and marking by utilizing the historical click data to obtain a training sample.
In specific implementation, when the historical content corresponds to clicking, praying or collecting, the historical clicking data is a first label (for example, 1), when the historical content corresponds to no clicking, praying or collecting, the historical clicking data is a second label (for example, 0), and thus, each combination of the historical content and the characteristic data is marked by utilizing the historical clicking data, and the obtained training sample is accurate.
And F2, training the built initial model by using a training sample, outputting a click rate training result corresponding to the training sample, and adjusting parameters of the initial model according to the difference degree of the click rate training result and marked historical click data.
In specific implementation, the initial model is a pre-built initial CTR (Click-Through-Rate) model. The characteristic data of the user in the training sample is converted into a user characteristic vector, the content characteristic of the corresponding historical content is converted into a content characteristic vector, the user characteristic vector and the content characteristic vector are analyzed, the crossing degree of the user characteristic vector and the content characteristic vector is determined, and the click rate training result is determined according to the crossing degree.
And determining a loss function according to the difference between the click rate training result and the marked historical click data, and adjusting the parameters of the initial model according to the loss value calculated by the loss function to complete the training of a training sample.
And F3, after the training of all training samples is determined, taking the final adjusted initial model as a click rate prediction model.
When the method is implemented, after training of all training samples is determined, a group of test samples are redetermined according to the collection mode of the training samples, the final adjusted initial model is tested by using the test samples, and the prediction accuracy of the final adjusted initial model is determined. And if the prediction accuracy is greater than or equal to the accuracy threshold, taking the initial model after final adjustment as a click rate prediction model. If the prediction accuracy is smaller than the accuracy threshold, the test sample is used as a training sample to train the initial model after final adjustment until the prediction accuracy of the initial model after final adjustment is larger than or equal to the accuracy threshold, and the initial model after final adjustment is used as a click rate prediction model.
Through the scheme, the learning training can be performed by utilizing the relatively accurate training sample based on the initial model, and then the parameters of the initial model are continuously adjusted, so that the finally obtained click rate prediction model can accurately identify the requirements of the user on each content, further accurately determine the predicted click rate of the user on each content, and ensure the prediction accuracy of the click rate prediction model.
In some embodiments, after step 103, further comprising:
And 104, in response to receiving unsatisfactory feedback of the user on the target recommended content, iteratively executing a dialogue process with the user by using the target recommended model, and determining new target recommended content according to dialogue content until the user makes a selection based on the new target recommended content to stop iteration.
In specific implementation, a target recommendation model of a type corresponding to a target intention is prestored in the target recommendation model, and the target recommendation model is obtained through learning training by using a Large Language Model (LLM). If the user is not satisfied with the target recommended content, an interactive dialogue is performed between the user and the target recommended content by using the target recommended model in the target recommended model, and new target recommended content is redetermined according to the interactive dialogue result until the user can select one of the new target recommended content which is needed most.
And step 105, responding to receiving the execution content selected by the user from the target recommended content or the new target recommended content, and executing the corresponding function according to the execution content.
In the specific implementation, the user selects one of the most needed contents from the target recommended contents (or the new target recommended contents) as the execution content according to the own needs, and then performs the function execution according to the execution content.
Through the scheme, even if the obtained target recommended content does not meet the requirement of the user, the new target recommended content can be redetermined through interactive dialogue with the user, so that the user can select the most needed execution content to perform function execution, a more accurate requirement function is provided for the user, and the user can use the target recommended content conveniently.
In the following, a procedure specifically executed in the vehicle control system is shown in fig. 2 in a specific embodiment of the information recommendation method of the present application.
1. User demand information (e.g., user voice request) is received.
2. The user demand information is pre-processed (e.g., the voice request is converted to text content by a voice-to-text model).
3. And inputting the user demand information into the intention complexity classification model, and outputting an intention complexity classification result.
The training process of the intention complexity classification model is as follows:
Training samples are obtained according to various collected user demand information (e.g., query) and labeled with corresponding complexity types. For example, user demand information, i say, "i now want to eat special Sichuan dishes", the intention is clear, easy to identify, the label is simple, the label is 0, the intention is fuzzy, the label is complex, and the label is 1.
The pre-built classification model (e.g., BERT, bidirectional Encoder Representations from Transformers, pre-trained language model) is then trained with training samples to obtain the intent complexity classification model.
4. If the complexity classification result is a complex class, the complex class is input into a first intention recognition model which is trained based on a large language model LLM, and the target intention is determined through a multi-round dialogue result with the user.
A) And obtaining a first analysis result according to the user demand information by using the first intention recognition model, and outputting and feeding back to the user.
B) And the user feeds back according to the first analysis result, and the first intention recognition model receives user feedback information, analyzes the user feedback information to obtain the undetermined intention type and outputs the undetermined intention type to the user.
C) If the user accepts, the user feeds back that the type of the pending intention is correct, and the type of the pending intention is taken as a target intention;
if the user feeds back the undetermined intention type errors, the undetermined intention type is corrected by using the first intention recognition model until the user feeds back the undetermined intention type errors, and the undetermined intention type is taken as a target intention.
For example, the user is immediately away from the hotel.
A first intention recognition model, do a respected owner hear you get to the hotel immediately, need you recommend nearby places to play?
The user is that me is now starved, recommending me a nearby restaurant bar.
The first intention recognition model is good, and recommends some restaurants to you immediately.
The first intent recognition model analyzes the dialog results to determine that the target intent is a dining intent.
5. If the complexity classification result is a simple class, the complexity classification result is input into a second intention recognition model trained based on the classification model to determine the target intention.
The training process of the second intention recognition model is as follows:
a) A large amount of user intention information (e.g., query) is collected in advance, and the user intention information is labeled with actual intention to obtain a training sample.
B) The pre-built bi-classification model (e.g., BERT, bidirectional Encoder Representations from Transformers, pre-trained language model) is trained to obtain a second intent recognition model.
6. And selecting a corresponding target recommendation model according to the target intention.
A) If the target intention is a music intention, selecting a music recommendation model;
b) If the target intention is a movie intention, selecting a movie recommendation model;
c) If the target intention is a catering intention, selecting a catering recommendation model;
d) If the target intent is a sight intent, a sight recommendation model is selected.
7. And inputting the user demand information into a target recommendation model, determining a first preset number of contents to be recommended through a recall process, finely ranking to obtain a ranking result, and selecting a second preset number (for example, top5 in the ranking result) of target recommended contents.
The I recall process is specifically classified as the 3-case.
(1) And forming a recall list according to the content to be recommended determined by the cooperation (Usercf) of the user and the similar user.
A) And calculating the similarity between the user and other users according to the behavior information of the user, sorting the other users according to the similarity, selecting a third preset number of other users with high similarity as similar users, and storing the identity information (such as ID addresses) of the similar users and the users in a database in an associated manner.
B) And obtaining similar users from the database according to the identity information of the users, determining the operation behaviors (clicking, collecting or praying) of the similar users in a preset time period, determining a plurality of trigger contents according to the operation behaviors, removing the duplication, screening out a first preset number of contents to be recommended, and forming a recall list.
(2) And forming a recall list according to the content corresponding to the behavior of the user and the determined content to be recommended which is similar to the content (Itemc).
A) The similarity between the contents is calculated, and a fourth predetermined number (for example, the top 100 or the top 50 or the top 10 ordered from the top to the bottom according to the similarity) of the similar contents corresponding to each of the contents is determined and stored in the database.
B) The method comprises the steps of obtaining a behavior list corresponding to behavior information (such as clicking, collecting and praying) of a user, determining corresponding trigger content according to the behavior list of the user in a preset time period, calling similar content corresponding to the trigger content from a database, filtering out a first preset number of to-be-recommended content from the similar content after repeated use, and forming a recall list.
(3) And determining interest tags according to the behaviors of the user, and further finding the content to be recommended.
A) And determining the interest tag according to the behavior information of the user.
B) Searching a plurality of contents corresponding to the interest tags, performing de-duplication processing on the plurality of contents, and screening a first preset number of contents to be recommended with highest heat from the contents to form a recall list.
The fine discharge process II is as follows:
a) Training a Click Rate prediction model, namely acquiring characteristic data (comprising at least one of basic characteristics, historical behavior characteristics and preference characteristics) of a user, historical content and historical Click data corresponding to the historical content, determining content characteristics (comprising at least one of content release time, content title, content label and other characteristics of the content) according to the historical content, combining the characteristic data with the content characteristics, marking by utilizing the historical Click data to obtain a training sample, and training a constructed initial CTR (Click-Through-Rate) model by utilizing the training sample to obtain the Click Rate prediction model.
B) And inputting the characteristic data of the user and the first preset number of to-be-recommended contents in the recall list into a click rate prediction model, outputting predicted click rates (namely, the matching degree of the to-be-recommended contents and the user requirements) corresponding to the to-be-recommended contents, and accurately sequencing the recall list according to the predicted click rates to obtain a sequencing result.
8. And acquiring detailed information of the target recommended content.
A) If the target recommended content is music, artist information, genre information, album information, emotion information, language information, and the like are acquired.
B) If the target recommended content is a movie, actor information, director information, genre information, material information, language information, movie comment information, and the like are acquired.
C) If the target recommended content is a restaurant, restaurant scoring information, cuisine information, average price information, good scoring information, featured dish information, address information, and the like are obtained.
D) And if the target recommended content is a scenic spot, acquiring the card punching point information, the playing route information, the characteristic information, the route information and the like.
9. And sending the target recommended content and the detailed information to the user.
10. If the user is satisfied, the execution content is selected, and the corresponding execution system is started to execute the corresponding function according to the execution content.
A) If the execution content is music, starting the audio system, and playing the song corresponding to the execution content.
B) If the execution content is a film and television, starting a video system, and playing the video corresponding to the execution content.
C) If the execution content is a restaurant or a scenic spot, the navigation system is turned on to determine a destination and navigate to the destination.
11. If the user is not satisfied, a dialogue process is performed with the user and new target recommended content is determined based on the dialogue content until the user selects the execution content from the new target recommended content.
And iteratively executing a dialogue process interacted with a user by using the target recommendation model, and determining new target recommendation contents according to dialogue contents until the user makes a selection to stop iteration based on the new target recommendation contents.
For example, the user is immediately away from the hotel.
Target recommendation model is the owner who gets to the hotel immediately to recommend nearby scenic spots for you?
The user is starved and recommends some restaurant bars.
And (5) an intention processing model, namely re-determining the intention of the target and determining a new target recommendation model.
And (3) a new target recommendation model, namely determining restaurants (namely, target recommendation contents) according to the destination hotels and feeding back the restaurants (namely, target recommendation contents) to the user.
The user dislikes the restaurants and reselects Hangzhou cuisine restaurants.
And determining a restaurant (namely new target recommended content) of the Hangzhou cuisine and feeding back the new target recommended model to the user.
The user selects the first restaurant.
And (3) the new target recommendation model is that the first restaurant and the corresponding detailed information are sent to a navigation system to navigate to the first restaurant.
In summary, the user demand can be accurately understood by combining various models according to the complexity of the user demand information and combining multiple rounds of dialogue interaction, so that the accurate target intention is determined, the target recommendation model corresponding to the target intention is utilized to capture the user demand, and the most accurate target recommendation content is determined by intelligent analysis through multiple rounds of interaction feedback with the user.
It should be noted that, the method of the embodiment of the present application may be performed by a single device, for example, a computer or a server. The method of the embodiment can also be applied to a distributed scene, and is completed by mutually matching a plurality of devices. In the case of such a distributed scenario, one of the devices may perform only one or more steps of the method of an embodiment of the present application, the devices interacting with each other to accomplish the method.
It should be noted that the foregoing describes some embodiments of the present application. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments described above and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
Based on the same inventive concept, the application also provides an information recommendation device corresponding to the method of any embodiment.
Referring to fig. 3, the apparatus includes:
The intention complexity classification module 201 is configured to receive user demand information, send the user demand information to a intention complexity classification model trained in advance, and obtain a complexity classification result;
An intent analysis module 202 configured to determine a corresponding target intent recognition model according to the complexity classification result, analyze based on user intent information using the target intent recognition model, and determine a target intent, wherein the target intent recognition model is a model trained in advance;
the recommended content determining module 203 is configured to select a target recommended model corresponding to the target intention, determine a target recommended content using the target recommended model, and feed back the target recommended content to the user, wherein the recommended model corresponding to each intention is stored.
In some embodiments, the complexity classification result is a simple class or a complex class;
The intent analysis module 202 is specifically configured to:
In response to determining that the complexity classification result is a complex class, determining a corresponding target intention recognition model as a first intention recognition model, receiving a multi-round dialogue result with a user, recognizing the multi-round dialogue result by using the first intention recognition model, and determining a target intention;
Or alternatively
And in response to determining that the complexity classification result is a simple class, determining that a corresponding target intention recognition model is a second intention recognition model, and recognizing the user intention information by using the second intention recognition model to determine a target intention.
In some embodiments, the intent analysis module 202 is specifically further configured to:
analyzing the user intention information by using the first intention recognition model to obtain a first analysis result and outputting the first analysis result so that a user can make feedback according to the output first analysis result;
Receiving user feedback information, analyzing the user feedback information by using the first intention recognition model to obtain a pending intention type, and outputting the pending intention type so as to enable a user to determine whether the outputted pending intention type is correct or not;
iteratively performing a correction process of the pending intent type using the first intent recognition model in response to receiving a user feedback pending intent type error until the corrected intent type is correct, taking the corrected intent type as the target intent, or
And responding to the received user feedback that the type of the undetermined intention is correct, and taking the type of the undetermined intention as the target intention.
In some embodiments, the recommended content determination module 203 is specifically configured to:
acquiring behavior information of a user by utilizing a target recommendation model, and determining corresponding first preset number of contents to be recommended according to the behavior information of the user;
sorting and adjusting the first preset number of contents to be recommended according to the matching degree with the user requirement by utilizing a target recommendation model to obtain a sorting result;
And selecting a second preset number of contents to be recommended with highest matching degree from the sorting results by using a target recommendation model as the target recommended contents, wherein the second preset number is smaller than the first preset number.
In some embodiments, the recommended content determination module 203 is specifically further configured to:
determining similar users based on the behavior information of the users, wherein the similar users corresponding to each user are prestored, and the similar users are other users matched with the behavior information of the users;
acquiring operation behaviors of the similar users in a preset time period, and determining a plurality of trigger contents according to the operation behaviors;
Filtering the plurality of trigger contents, and screening a first preset number of contents to be recommended from the filtered plurality of trigger contents;
and/or the number of the groups of groups,
Determining triggering content according to the behavior information of the user;
acquiring a plurality of similar contents corresponding to the trigger contents, wherein the similar contents corresponding to each trigger content are stored in advance;
filtering the similar contents, and screening a first preset number of contents to be recommended from the filtered similar contents;
and/or the number of the groups of groups,
Determining interest tags according to the behavior information of the user;
searching a plurality of matching contents corresponding to the interest tags, filtering the plurality of matching contents, and screening a first preset number of contents to be recommended from the plurality of filtered matching contents.
In some embodiments, the recommended content determination module 203 is specifically further configured to:
Acquiring characteristic data of a user, inputting the characteristic data of the user and a first preset number of contents to be recommended into a click rate prediction model, determining a predicted click rate of each content to be recommended, and taking the predicted click rate as the matching degree of the content to be recommended and the user requirement;
and sorting and adjusting the first preset number of contents to be recommended according to the matching degree to obtain a sorting result.
In some embodiments, the apparatus further comprises a model training module configured to:
Acquiring characteristic data, historical content and historical click data corresponding to the historical content of a user, combining the characteristic data with the historical content, and marking by utilizing the historical click data to obtain a training sample;
Training the built initial model by using a training sample, outputting a click rate training result corresponding to the training sample, and adjusting parameters of the initial model according to the difference degree of the click rate training result and marked historical click data;
And after the training of all training samples is determined, taking the initial model after final adjustment as a click rate prediction model.
In some embodiments, the apparatus further comprises a modification adjustment module configured to:
After the target recommended content is fed back to the user, responding to receiving unsatisfactory feedback of the user on the target recommended content, iteratively executing a dialogue process with the user by utilizing the target recommended model, and determining new target recommended content according to dialogue content until the user makes a selection based on the new target recommended content to stop iteration;
after the target recommended content is fed back to the user, responding to receiving the execution content selected by the user from the target recommended content or the new target recommended content, and executing a corresponding function according to the execution content.
For convenience of description, the above devices are described as being functionally divided into various modules, respectively. Of course, the functions of each module may be implemented in the same piece or pieces of software and/or hardware when implementing the present application.
The device of the foregoing embodiment is configured to implement the corresponding method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which is not described herein.
Based on the same inventive concept, the application also provides an electronic device corresponding to the method of any embodiment, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor executes the computer program to realize the method of any embodiment.
Fig. 4 shows a more specific hardware architecture of an electronic device provided by the present embodiment, which may include a processor 1010, a memory 1020, an input/output interface 1030, a communication interface 1040, and a bus 1050. Wherein processor 1010, memory 1020, input/output interface 1030, and communication interface 1040 implement communication connections therebetween within the device via a bus 1050.
The processor 1010 may be implemented by a general-purpose CPU (Central Processing Unit ), a microprocessor, an Application SPECIFIC INTEGRATED Circuit (ASIC), or one or more integrated circuits, etc. for executing related programs to implement the technical solutions provided in the embodiments of the present disclosure.
The Memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory ), static storage, dynamic storage, etc. Memory 1020 may store an operating system and other application programs, and when the embodiments of the present specification are implemented in software or firmware, the associated program code is stored in memory 1020 and executed by processor 1010.
The input/output interface 1030 is used to connect with an input/output module for inputting and outputting information. The input/output module may be configured as a component in a device (not shown) or may be external to the device to provide corresponding functionality. Wherein the input devices may include a keyboard, mouse, touch screen, microphone, various types of sensors, etc., and the output devices may include a display, speaker, vibrator, indicator lights, etc.
Communication interface 1040 is used to connect communication modules (not shown) to enable communication interactions of the present device with other devices. The communication module may implement communication through a wired manner (such as USB, network cable, etc.), or may implement communication through a wireless manner (such as mobile network, WIFI, bluetooth, etc.).
Bus 1050 includes a path for transferring information between components of the device (e.g., processor 1010, memory 1020, input/output interface 1030, and communication interface 1040).
It should be noted that although the above-described device only shows processor 1010, memory 1020, input/output interface 1030, communication interface 1040, and bus 1050, in an implementation, the device may include other components necessary to achieve proper operation. Furthermore, it will be understood by those skilled in the art that the above-described apparatus may include only the components necessary to implement the embodiments of the present description, and not all the components shown in the drawings.
The electronic device of the foregoing embodiment is configured to implement the corresponding method in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiment, which is not described herein.
Based on the same inventive concept, the present application also provides a non-transitory computer readable storage medium storing computer instructions for causing the computer to perform the method according to any of the embodiments above, corresponding to the method according to any of the embodiments above.
The computer readable media of the present embodiments, including both permanent and non-permanent, removable and non-removable media, may be used to implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change Memory (PRAM, PARAMETER RANDOM ACCESS MEMORY, parameter Random Access Memory), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM, dynamic Random Access Memory), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically erasable programmable read Only Memory (EEPROM, ELECTRICALLY ERASABLE PROGRAMMABLE READ ONLY MEMORY), flash Memory or other Memory technology, read Only optical disk read Only Memory (CD-ROM, compact Disc Read-Only Memory), digital versatile disks (DVD, digital Video Disc) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by the computing device.
The storage medium of the foregoing embodiments stores computer instructions for causing the computer to perform the method of any of the foregoing embodiments, and has the advantages of the corresponding method embodiments, which are not described herein.
Based on the same conception, the application also provides a computer program product corresponding to the method of any embodiment, comprising computer program instructions, which when run on a computer, cause the computer to execute the method of any embodiment, and the method has the beneficial effects of the corresponding method embodiment, which are not repeated herein.
Based on the same inventive concept, the application also provides a vehicle, which comprises the vehicle-mounted charging device or the electronic equipment. The vehicle-mounted charging device has the beneficial effects of the corresponding embodiments of the vehicle-mounted charging device or the electronic equipment, and are not repeated herein.
It will be appreciated that before using the technical solutions of the embodiments of the present application, the user is informed of the type, the range of use, the use scenario, etc. of the related personal information in an appropriate manner, and the authorization of the user is obtained.
For example, in response to receiving an active request from a user, a prompt is sent to the user to explicitly prompt the user that the operation it is requesting to perform will require personal information to be obtained and used with the user. Therefore, the user can select whether to provide personal information to the software or hardware such as the electronic equipment, the application program, the server or the storage medium for executing the operation of the technical scheme according to the prompt information.
As an alternative but non-limiting implementation, in response to receiving an active request from a user, the manner in which the prompt information is sent to the user may be, for example, a popup, in which the prompt information may be presented in a text manner. In addition, a selection control for the user to select to provide personal information to the electronic device in a 'consent' or 'disagreement' manner can be carried in the popup window.
It will be appreciated that the above-described notification and user authorization acquisition process is merely illustrative, and not limiting of the implementation of the present application, and that other ways of satisfying relevant legal regulations may be applied to the implementation of the present application.
It will be appreciated by persons skilled in the art that the foregoing discussion of any embodiment is merely exemplary and is not intended to imply that the scope of the application (including the claims) is limited to these examples, that combinations of technical features in the foregoing embodiments or in different embodiments may be implemented in any order and that many other variations of the different aspects of the embodiments described above exist within the spirit of the application, which are not provided in detail for clarity.
Additionally, well-known power/ground connections to Integrated Circuit (IC) chips and other components may or may not be shown within the provided figures, in order to simplify the illustration and discussion, and so as not to obscure the embodiments of the present application. Furthermore, the devices may be shown in block diagram form in order to avoid obscuring the embodiments of the present application, and also in view of the fact that specifics with respect to implementation of such block diagram devices are highly dependent upon the platform within which the embodiments of the present application are to be implemented (i.e., such specifics should be well within purview of one skilled in the art). Where specific details (e.g., circuits) are set forth in order to describe example embodiments of the application, it should be apparent to one skilled in the art that embodiments of the application can be practiced without, or with variation of, these specific details. Accordingly, the description is to be regarded as illustrative in nature and not as restrictive.
While the application has been described in conjunction with specific embodiments thereof, many alternatives, modifications, and variations of those embodiments will be apparent to those skilled in the art in light of the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.
The present embodiments are intended to embrace all such alternatives, modifications and variances which fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, and the like, which are within the spirit and principles of the embodiments of the application, are intended to be included within the scope of the application.

Claims (10)

1.一种信息推荐方法,其特征在于,包括:1. An information recommendation method, comprising: 接收用户需求信息,将所述用户需求信息发送至预先训练得到的意图复杂度分类模型,得到复杂度分类结果;Receiving user demand information, and sending the user demand information to a pre-trained intent complexity classification model to obtain a complexity classification result; 根据所述复杂度分类结果,确定对应的目标意图识别模型,利用所述目标意图识别模型基于用户意图信息进行分析,确定目标意图,其中,所述目标意图识别模型是预先训练得到的模型;Determine a corresponding target intent recognition model based on the complexity classification result, and use the target intent recognition model to analyze based on the user intent information to determine the target intent, wherein the target intent recognition model is a pre-trained model; 选择与所述目标意图对应的目标推荐模型,利用目标推荐模型确定目标推荐内容,并将所述目标推荐内容反馈给用户。A target recommendation model corresponding to the target intention is selected, target recommendation content is determined using the target recommendation model, and the target recommendation content is fed back to the user. 2.根据权利要求1所述的方法,其特征在于,所述复杂度分类结果为简单类或复杂类;2. The method according to claim 1, wherein the complexity classification result is a simple class or a complex class; 所述根据所述复杂度分类结果,确定对应的目标意图识别模型,利用所述目标意图识别模型基于用户意图信息进行分析,确定目标意图,包括:Determining a corresponding target intent recognition model according to the complexity classification result, and analyzing the target intent based on the user intent information using the target intent recognition model to determine the target intent includes: 响应于确定所述复杂度分类结果为复杂类,确定对应的目标意图识别模型为第一意图识别模型,接收与用户的多轮对话结果,利用所述第一意图识别模型对多轮对话结果进行识别,确定目标意图;In response to determining that the complexity classification result is a complex class, determining that the corresponding target intent recognition model is a first intent recognition model, receiving multiple rounds of conversation results with the user, and using the first intent recognition model to recognize the multiple rounds of conversation results to determine the target intent; 或者,or, 响应于确定所述复杂度分类结果为简单类,确定对应的目标意图识别模型为第二意图识别模型,利用所述第二意图识别模型对所述用户意图信息进行识别确定目标意图。In response to determining that the complexity classification result is a simple class, determining that the corresponding target intent recognition model is a second intent recognition model, and using the second intent recognition model to identify the user intent information to determine the target intent. 3.根据权利要求2所述的方法,其特征在于,所述接收与用户的多轮对话结果,利用所述第一意图识别模型对多轮对话结果进行识别,确定目标意图,包括:3. The method according to claim 2, wherein receiving the results of multiple rounds of conversations with the user and using the first intent recognition model to recognize the results of the multiple rounds of conversations and determine the target intent comprises: 利用所述第一意图识别模型对所述用户意图信息进行分析,得到第一分析结果并输出,以供用户根据输出的第一分析结果做出反馈;Analyzing the user intention information using the first intention recognition model to obtain a first analysis result and output it, so that the user can provide feedback based on the output first analysis result; 接收用户反馈信息,利用所述第一意图识别模型对所述用户反馈信息进行分析得到待定意图类型并输出,以供用户确定输出的待定意图类型是否正确;receiving user feedback information, analyzing the user feedback information using the first intent recognition model to obtain a pending intent type and outputting the result, so that the user can determine whether the outputted pending intent type is correct; 响应于接收到用户反馈的待定意图类型错误,利用所述第一意图识别模型迭代执行对待定意图类型的修正过程,直至修正后的意图类型正确,将所述修正后的意图类型作为所述目标意图;或者,In response to receiving user feedback that the pending intent type is incorrect, iteratively performing a correction process on the pending intent type using the first intent recognition model until the corrected intent type is correct, and using the corrected intent type as the target intent; or 响应于接收到用户反馈的待定意图类型正确,将所述待定意图类型作为所述目标意图。In response to receiving user feedback that the pending intent type is correct, the pending intent type is used as the target intent. 4.根据权利要求1所述的方法,其特征在于,所述利用目标推荐模型确定用户需要的目标推荐内容,包括:4. The method according to claim 1, wherein determining the target recommendation content required by the user using the target recommendation model comprises: 利用目标推荐模型执行:Using the target recommendation model to perform: 获取用户的行为信息,根据所述用户的行为信息确定对应的第一预定数量的待推荐内容;Acquiring user behavior information, and determining a first predetermined number of corresponding content to be recommended based on the user behavior information; 对所述第一预定数量的待推荐内容,按照与用户需求的匹配程度进行排序调整,得到排序结果;Sorting and adjusting the first predetermined number of contents to be recommended according to their matching degree with the user's needs to obtain a ranking result; 从所述排序结果中,选出匹配程度最高的第二预定数量的待推荐内容作为所述目标推荐内容,其中所述第二预定数量小于所述第一预定数量。A second predetermined number of to-be-recommended contents with the highest matching degree is selected from the sorting results as the target recommended contents, wherein the second predetermined number is smaller than the first predetermined number. 5.根据权利要求4所述的方法,其特征在于,所述根据所述用户的行为信息确定对应的第一预定数量的待推荐内容,包括:5. The method according to claim 4, wherein determining the first predetermined number of content to be recommended according to the user's behavior information comprises: 基于所述用户的行为信息确定相似用户,其中,预先存储每个用户对应的相似用户,所述相似用户为与用户的行为信息匹配的其他用户;Determining similar users based on the user's behavior information, wherein similar users corresponding to each user are pre-stored, and the similar users are other users matching the user's behavior information; 获取所述相似用户在预定时间段内的操作行为,根据所述操作行为确定多个触发内容;Obtaining operation behaviors of the similar users within a predetermined time period, and determining multiple trigger contents according to the operation behaviors; 对多个所述触发内容进行过滤处理,从过滤后的多个所述触发内容中筛选出第一预定数量的待推荐内容;Filtering the plurality of triggering contents, and selecting a first predetermined number of contents to be recommended from the filtered plurality of triggering contents; 和/或,and/or, 根据所述用户的行为信息确定触发内容;Determining trigger content based on the user's behavior information; 获取与所述触发内容对应的多个相似内容,其中,预先存储与每个触发内容对应的相似内容;Acquire a plurality of similar contents corresponding to the triggering content, wherein similar contents corresponding to each triggering content are pre-stored; 对多个所述相似内容进行过滤处理,从过滤后的多个相似内容中筛选出第一预定数量的待推荐内容;Filtering the plurality of similar contents, and selecting a first predetermined number of contents to be recommended from the filtered plurality of similar contents; 和/或,and/or, 根据所述用户的行为信息确定兴趣标签;Determining interest tags based on the user's behavior information; 查找与所述兴趣标签对应的多个匹配内容,对多个所述匹配内容进行过滤处理,从过滤后的多个匹配内容中筛选出第一预定数量的待推荐内容。A plurality of matching contents corresponding to the interest tag is searched, the plurality of matching contents are filtered, and a first predetermined number of to-be-recommended contents are screened out from the filtered plurality of matching contents. 6.根据权利要求4所述的方法,其特征在于,所述对所述第一预定数量的待推荐内容,按照与用户需求的匹配程度进行排序调整,得到排序结果,包括:6. The method according to claim 4, wherein the step of sorting the first predetermined number of recommended contents according to their matching degree with user needs to obtain a sorting result comprises: 获取用户的特征数据,将所述用户的特征数据与第一预定数量的待推荐内容输入至点击率预测模型中,确定每个待推荐内容的预测点击率,将所述预测点击率作为所述待推荐内容与用户需求的匹配程度;Obtaining user feature data, inputting the user feature data and a first predetermined number of to-be-recommended content into a click-through rate prediction model, determining a predicted click-through rate for each to-be-recommended content, and using the predicted click-through rate as a degree of match between the to-be-recommended content and the user's needs; 对所述第一预定数量的待推荐内容按照所述匹配程度进行排序调整,得到排序结果。The first predetermined number of contents to be recommended are sorted and adjusted according to the matching degree to obtain a sorting result. 7.根据权利要求6所述的方法,其特征在于,所述点击率预测模型的训练过程包括:7. The method according to claim 6, wherein the training process of the click-through rate prediction model comprises: 获取用户的特征数据、历史内容以及与历史内容对应的历史点击数据,将特征数据与历史内容组合,并利用所述历史点击数据进行标记,得到训练样本;Acquire user feature data, historical content, and historical click data corresponding to the historical content, combine the feature data with the historical content, and use the historical click data for labeling to obtain training samples; 利用训练样本对构建的初始模型进行训练,输出训练样本对应的点击率训练结果,根据所述点击率训练结果与标记的历史点击数据的差异程度,对初始模型的参数进行调整;Using the training samples to train the constructed initial model, outputting click-through rate training results corresponding to the training samples, and adjusting the parameters of the initial model according to the degree of difference between the click-through rate training results and the marked historical click data; 确定所有训练样本训练完成后,将最终调整后的初始模型作为点击率预测模型。After all training samples are trained, the final adjusted initial model is used as the click-through rate prediction model. 8.根据权利要求1所述的方法,其特征在于,所述将所述目标推荐内容反馈给用户之后,还包括:8. The method according to claim 1, characterized in that after feeding back the target recommended content to the user, the method further comprises: 响应于接收到用户对所述目标推荐内容的不满意反馈,利用所述目标推荐模型迭代执行与用户交互对话过程,并根据对话内容确定新的目标推荐内容,直到用户基于所述新的目标推荐内容做出选择停止迭代;In response to receiving user feedback indicating dissatisfaction with the target recommended content, iteratively executing an interactive dialogue process with the user using the target recommendation model, and determining new target recommended content based on the dialogue content, until the user makes a selection based on the new target recommended content and the iteration stops; 响应于接收到用户从所述目标推荐内容中或从所述新的目标推荐内容中选择的执行内容,按照所述执行内容执行对应的功能。In response to receiving the execution content selected by the user from the target recommended content or from the new target recommended content, a corresponding function is executed according to the execution content. 9.一种电子设备,包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序,其特征在于,所述处理器执行所述计算机程序时实现如权利要求1至8任意一项所述的方法。9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 8 when executing the computer program. 10.一种车辆,其特征在于,包括权利要求9所述的电子设备。10. A vehicle, comprising the electronic device according to claim 9.
CN202510880136.4A 2025-06-27 2025-06-27 Information recommendation method, electronic device, and vehicle Pending CN120705409A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202510880136.4A CN120705409A (en) 2025-06-27 2025-06-27 Information recommendation method, electronic device, and vehicle

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202510880136.4A CN120705409A (en) 2025-06-27 2025-06-27 Information recommendation method, electronic device, and vehicle

Publications (1)

Publication Number Publication Date
CN120705409A true CN120705409A (en) 2025-09-26

Family

ID=97121600

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202510880136.4A Pending CN120705409A (en) 2025-06-27 2025-06-27 Information recommendation method, electronic device, and vehicle

Country Status (1)

Country Link
CN (1) CN120705409A (en)

Similar Documents

Publication Publication Date Title
US8612866B2 (en) Information processing apparatus, information processing method, and information processing program
CN109033229B (en) Question and answer processing method and device
CN109255053B (en) Resource searching method, device, terminal, server and computer readable storage medium
CN113424175B (en) Intuitive voice search
US9507326B2 (en) System and method for using biometrics to predict and select music preferences
US20200226133A1 (en) Knowledge map building system and method
US9003435B2 (en) Method of recommending media content and media playing system thereof
US20190340201A1 (en) Computerized domain expert
TWI449410B (en) Personalized Sorting Method of Internet Audio and Video Data
CN110019943B (en) Video recommendation method and device, electronic equipment and storage medium
CN106227815A (en) The personalized application program function of a kind of multi-modal clue recommends method and system thereof
CN107729578B (en) Method and device for recommending music
JP2013517563A (en) User communication analysis system and method
WO2014176241A2 (en) Explanations for personalized recommendations
CN110991476A (en) Training of decision classifier, audio and video recommendation method, device and storage medium
CN112765400B (en) Weight updating method, content recommending method, device and equipment for interest labels
CN109740072B (en) Hotel ordering method and system based on POI under OTA platform
KR20130055748A (en) System and method for recommending of contents
CN109275047A (en) Video information processing method and device, electronic device, and storage medium
KR20200133976A (en) Contents Curation Method and Apparatus thereof
CN117809651A (en) Voice interaction device and voice interaction method
CN114691990A (en) Recommendation method, device, server, storage medium and product of query options
CN116521979A (en) Target object display method and device, electronic equipment and storage medium
JP7649874B2 (en) Method and apparatus for recommending music content - Patents.com
CN111831892A (en) Information recommendation method, information recommendation device, server and storage medium

Legal Events

Date Code Title Description
PB01 Publication
SE01 Entry into force of request for substantive examination