CN115775052A - Complaint prediction method and device, electronic equipment and storage medium - Google Patents

Abstract

The embodiment of the application provides a complaint prediction method, a complaint prediction device, electronic equipment and a storage medium, which are applied to a man-machine conversation scene, wherein the method comprises the following steps: acquiring preset static characteristics used for representing a user and preset dynamic characteristics used for representing the current speech of the user; determining user characteristics according to the static characteristics and the dynamic characteristics; and inputting the user characteristics into the trained complaint prediction model, and acquiring the complaint probability of the user under the current speech. Static features are inherent attributes of a user. The dynamic feature is associated with the current utterance. Because the probability that a certain user generates complaints in one man-machine conversation is related to the user and each speaking of the user in the conversation, in the prediction process, the inherent attribute of the user and the current speaking condition of the user are taken into consideration, complaint prediction is carried out from multiple dimensions and multiple characteristics, various factors influencing the complaints can be taken into consideration, and the prediction accuracy of the complaint probability is improved.

Description

Complaint prediction method, device, electronic equipment and storage medium

technical field

The present application relates to the technical field of human-computer interaction, and in particular, to a complaint prediction method, device, electronic equipment, and storage medium.

Background technique

In recent years, with the rapid development of artificial intelligence (AI) technology, man-machine dialogue is widely used in various scenarios, such as telephone sales scenarios, intelligent customer service scenarios, and so on. In the human-machine dialogue, due to the deviation of the understanding ability of the robot, it is easy to cause poor user experience and cause complaints. How to improve the robot's understanding and prediction of the user's complaint intention is an urgent technical problem in this field.

Contents of the invention

The purpose of the embodiment of the present application is to provide a complaint prediction method, device, electronic equipment and storage medium, so as to achieve the technical effect of predicting the user's complaint intention in the man-machine dialogue.

The first aspect of the embodiment of the present application provides a complaint prediction method, which is applied to a man-machine dialogue scenario, and the method includes:

Obtain preset static features used to characterize the user and dynamic features used to characterize the user's current speech;

determining user characteristics according to the static characteristics and the dynamic characteristics;

Inputting the user characteristics into the trained complaint prediction model to obtain the complaint probability of the user under the current speech.

In the above implementation process, static features are used to characterize users and are inherent attributes of users. Dynamic features are used to characterize the current utterance and are related to the current utterance. Since the probability of a user complaining in a man-machine dialogue is related to the user himself and each speech of the user in the conversation, in the prediction process, the inherent attributes of the user itself and the current situation of the user’s speech are combined And taking it into account, predicting complaints from multiple dimensions and features can take into account various factors that affect complaints, thereby improving the accuracy of the prediction of complaint probability.

Further, the static features include one or more of the following features: user age, gender, geographic location;

The dynamic feature includes one or more of the following features: the intention of the current speech, the speech rate, the total duration of the dialogue, and the dialogue interval.

In the above implementation process, considering the user's age, gender, geographical location, and the current speaking intention, speech rate, total duration of the dialogue, and dialogue interval are related to the probability of a user complaining in a man-machine dialogue, so in During the forecasting process, the above characteristics are taken into consideration, and complaint prediction is made from multiple dimensions and features, and various factors that affect complaints can be taken into account, thereby improving the prediction accuracy of complaint probability .

Further, the complaint prediction model includes a decision tree layer and a linear regression layer; the input of the user characteristics into the trained complaint prediction model to obtain the complaint probability of the user under the current speech includes:

inputting said user characteristics into said decision tree layer;

The features extracted by the leaf nodes of the decision tree layer are input into the linear regression layer;

Obtain the complaint probability output by the linear regression layer.

In the above implementation process, the complaint prediction model including the decision tree layer and the linear regression layer is used to predict the probability of complaints, and the technical effect of predicting the user's complaint intention in the man-machine dialogue is realized.

Further, the method also includes:

Obtaining the feature importance of each feature in the user features from the decision tree layer;

For a target feature whose feature importance is greater than an importance threshold, when the user feature of the user matches the target feature, a target measure corresponding to the target feature is taken.

In the above implementation process, the cause of complaints is analyzed by obtaining the feature importance of each feature from the decision tree layer, and when the user features match the target features, corresponding target measures are taken to avoid complaints and increase man-machine The intelligence of dialogue avoids complaints.

Further, the method also includes:

If the complaint probability is greater than a preset probability threshold, a preset early warning measure is executed.

In the above implementation process, the complaint probability is used to judge whether to take early warning measures. In the case of a high probability of complaints, the occurrence of complaints can be avoided by taking early warning measures before complaints are generated, thereby increasing the intelligence of man-machine dialogue.

Further, the method also includes:

Acquiring the text sequence of the current speech;

performing keyword matching and/or sentiment analysis processing on the text sequence;

If it is determined according to the processing result that the user shows an intention to complain under the current speech, the early warning measures are executed.

In the above implementation process, through keyword matching and/or sentiment analysis technology, it is further analyzed whether the user has the intention to express complaints, and early warning measures are implemented, thereby improving the accuracy of early warning.

Further, before inputting the user characteristics into the trained complaint prediction model, performing the processing of keyword matching and/or sentiment analysis;

If it is determined according to the processing result that the user has no complaint intention under the current utterance, the step of inputting the user characteristics into the trained complaint prediction model is performed.

In the above implementation process, when using the complaint prediction model to predict complaints, firstly, the text sequence of the current speech is processed by keyword matching and/or sentiment analysis, and early warning measures are executed according to the processing results. Compared with predicting the probability of complaints, keyword matching results and sentiment analysis results can more accurately and quickly reflect users' complaint intentions. Therefore, performing keyword matching and/or sentiment analysis first can quickly determine whether to implement early warning measures and avoid complaints. produce.

The second aspect of the embodiment of the present application provides a complaint prediction device, which is applied to a man-machine dialogue scenario, and the device includes:

An acquisition module, configured to acquire preset static features used to characterize the user, and dynamic features used to characterize the user's current speech;

A determination module, configured to determine user characteristics according to the static characteristics and the dynamic characteristics;

A prediction module, configured to input the user characteristics into a trained complaint prediction model to obtain the complaint probability of the user under the current utterance.

The third aspect of the embodiment of the present application provides an electronic device, and the electronic device includes:

processor;

memory for storing processor-executable instructions;

Wherein, when the processor invokes the executable instruction, the operation of any one of the methods in the first aspect is implemented.

The fourth aspect of the embodiments of the present application provides a computer-readable storage medium, on which computer instructions are stored, and when the computer instructions are executed by a processor, the steps of any one of the methods described in the first aspect are implemented.

Description of drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the accompanying drawings that need to be used in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application, so It should not be regarded as a limitation on the scope, and those skilled in the art can also obtain other related drawings according to these drawings without creative work.

FIG. 1 is a schematic flowchart of a complaint prediction method provided in an embodiment of the present application;

FIG. 2 is a schematic flowchart of another complaint prediction method provided in the embodiment of the present application;

FIG. 3 is a schematic flowchart of another complaint prediction method provided in the embodiment of the present application;

FIG. 4 is a schematic flowchart of another complaint prediction method provided in the embodiment of the present application;

FIG. 5 is a schematic flowchart of another complaint prediction method provided in the embodiment of the present application;

FIG. 6 is a structural block diagram of a complaint prediction device provided in an embodiment of the present application;

FIG. 7 is a hardware structural diagram of an electronic device provided by an embodiment of the present application.

Detailed ways

The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.

It should be noted that like numerals and letters denote similar items in the following figures, therefore, once an item is defined in one figure, it does not require further definition and explanation in subsequent figures. Meanwhile, in the description of the present application, the terms "first", "second" and the like are only used to distinguish descriptions, and cannot be understood as indicating or implying relative importance.

In recent years, with the rapid development of artificial intelligence (AI) technology, man-machine dialogue is widely used in various scenarios, such as telemarketing scenarios, logistics, banking and other intelligent customer service scenarios. In the human-machine dialogue, due to the deviation of the understanding ability of the robot, it is easy to cause poor user experience and cause complaints. Especially in the telemarketing scenario, if a customer complaint occurs, it will have a serious negative impact on business development. Therefore, how to improve the robot's understanding and prediction of the user's complaint intention is a technical problem to be solved urgently in this field.

For this reason, this application proposes a complaint prediction method, which is applied to the scene of man-machine dialogue. As shown in Figure 1, the method comprises the steps of:

Step 110: Obtain preset static features used to characterize the user and dynamic features used to characterize the user's current speech;

Step 120: Determine user characteristics according to the static characteristics and the dynamic characteristics;

Step 130: Input the user characteristics into the trained complaint prediction model, and obtain the complaint probability of the user under the current utterance.

In the human-machine dialogue scenario, the user and the robot have a dialogue in the form of a question and an answer. Each speech of the user can be a question to the robot or a response to the robot. A man-machine dialogue includes one or more user speeches. That is, during a man-machine dialogue, the user will make at least one speech, generally speaking, several speeches. For each utterance of the user, step 110 to step 130 are performed to predict the user's complaint probability in the current utterance.

User characteristics are used to predict complaint probability. User features include static features and dynamic features. Among them, the static feature is used to represent the user and is an inherent attribute of the user; the dynamic feature is used to represent the user's current speech. In each utterance of the same user, the static features are the same, but the dynamic features are different. That is to say, the static feature is only related to the user himself, and has nothing to do with the user's speech; while the dynamic feature is related to the user's speech. Therefore, in a man-machine dialogue, only one static feature needs to be obtained from the database. In each speech, it is necessary to reacquire the dynamic features of the current speech. User characteristics are then determined based on the static and dynamic characteristics.

For example, static features and dynamic features may be combined in a preset order to obtain user features. For another example, static features and dynamic features may be added and summed to obtain user features.

Then, input the user characteristics into the trained complaint prediction model, and the complaint prediction model can output the user's complaint probability under the current speech by using the user characteristics.

In this embodiment, the user features used for complaint prediction include features of two dimensions: static features and dynamic features. Static features are used to characterize users and are inherent attributes of users. Dynamic features are used to characterize the current utterance and are related to the current utterance. Since the probability of a user complaining in a man-machine dialogue is related to the user himself and each speech of the user in the conversation, in the prediction process, the inherent attributes of the user itself and the current situation of the user’s speech are combined And taking it into account, predicting complaints from multiple dimensions and features can take into account various factors that affect complaints, thereby improving the accuracy of the prediction of complaint probability.

With regard to the static features characterizing the user, as mentioned above, the static features are the inherent attributes of the user, which are only related to the user himself and have nothing to do with the user's speech. In some embodiments, static features may include one or more of user age, gender, and geographic location. Wherein, the geographic location may be the geographical area where the user is located, such as East China, South China, North China, etc.; it may also be the province, urban area, etc. where the user is located.

Since the probability of a user generating a complaint may be related to age, gender, and geographic location, the static feature includes one or more of the user's age, gender, and geographic location in this embodiment.

Among them, the above-mentioned various user data, including age, gender, geographical location, etc., are obtained and processed on the basis of legality. For example, to obtain the user's consent, or to perform a contract, etc. And the above-mentioned various user data will only be processed within the prescribed or agreed scope.

Regarding the dynamic feature characterizing the user's current utterance, as described above, the dynamic feature is related to the user's utterance. In different utterances of the user, the corresponding dynamic features are not the same. In some embodiments, the dynamic feature may include one or more of the current speaking intention, speech rate, total duration of the dialogue, and dialogue interval.

Wherein, the intention of the current speech is predicted according to the speech content of the current speech. Exemplarily, an intention library or an intention set may be pre-established, which contains all possible intentions expressed by the user. For example, intents may include, but are not limited to, end-conversation intents, interest intents, inquiry intents, decline intents, and the like. Subsequently, as an example, the intent recognition model can be used to perform semantic analysis on the content of the current speech, and the intent recognition model can output the type of intent expressed by the current speech, so as to obtain the intent of the current speech. As another example, different intents correspond to different keywords. For example, the keywords corresponding to the intention of ending the dialogue include but not limited to "busy", "no time", "go to work" and so on; the keywords corresponding to the intention of interest include but not limited to "introduction", "specific" and so on. In this way, keyword matching can be performed on the current utterance to predict its intent. If the current utterance matches the keyword, it is determined that the intent of the current utterance is the intent corresponding to the matched keyword.

The speech rate of the current speech refers to the speed at which the user speaks in the current speech. If the speaking speed is fast and the user is currently in a state of anxiety or anger, the probability of complaints will be relatively high.

The total conversation duration refers to the total conversation duration of the user in this man-machine conversation. In the telemarketing scenario, if the total duration of the conversation is long, it means that the user may be interested in selling the product, so the probability of complaints is relatively small. However, in smart customer service scenarios, such as express delivery smart customer service, if the total conversation time is long, it means that it took a long time to solve the problems raised by the user and meet the needs of the user, then the user experience may be poor, so the probability of complaints is relatively low. larger.

The dialogue interval refers to the time interval from the last man-machine dialogue to the current man-machine dialogue of the same user, that is, the time from the last call to this call. If the dialogue interval is short, users may feel bored, so the probability of complaints is relatively high.

In this embodiment, considering that the user's age, gender, geographic location, and current speaking intention, speech rate, total duration of the dialogue, and dialogue interval are related to the probability that a certain user generates a complaint in a man-machine dialogue, so in During the forecasting process, the above characteristics are taken into consideration, and complaint prediction is made from multiple dimensions and features, and various factors that affect complaints can be taken into account, thereby improving the prediction accuracy of complaint probability .

In some embodiments, the complaint prediction model includes a decision tree layer and a linear regression layer. Exemplarily, the decision tree layer may be a Gradient Boost Decision Tree (Gradient Boost Decision Tree, GBDT). The linear regression layer is a linear regression model (Logistic Regression. LR). In the decision tree layer, the output of each leaf node is input into the linear regression layer with one-hot encoding for classification. In this way, the above step 130 may include steps as shown in FIG. 2:

Step 131: Input the user characteristics into the decision tree layer;

Step 132: Input the features extracted by the leaf nodes of the decision tree layer into the linear regression layer;

Step 133: Obtain the complaint probability output by the linear regression layer.

User features include static features and dynamic features. The static feature and the dynamic feature respectively include different features, so the user feature includes multiple features. In the decision tree layer, the path from the root node to the leaf node is equivalent to a feature combination of different features, and the leaf node can uniquely represent this path. In this way, when the user features are input into the decision tree layer, the final user features will fall on a certain leaf node. According to the fall of user features in the leaf nodes, the features extracted by the leaf nodes can be obtained and input into the linear regression layer in the form of one-hot encoding. Finally, a linear regression layer can output complaint probabilities.

Among them, the complaint prediction model is obtained through supervised training. The sample data includes historical complaint data and historical non-complaint data. Among them, the historical complaint data carries the first label, and the historical non-complaint data carries the second label.

Of course, the complaint prediction model in this application is not limited to the above-mentioned GBDT-LR model, and other models capable of outputting complaint probability in related technologies can be used as the complaint prediction model in this application.

Furthermore, in the man-machine dialogue scenario, in addition to predicting the user’s complaint probability, when it is found that the user has a tendency to complain, if the corresponding appeasement or remedial measures can be taken in time before the complaint occurs, the complaint can be avoided.

As mentioned above, the probability of complaints is related to various characteristics, and there are many reasons for complaints. It can be understood that the contribution of different features to the complaint probability is not the same, that is to say, the feature importance of each feature is not the same. According to the importance of features, it can assist in analyzing the cause of complaints. And the feature importance can be extracted from the decision tree layer. In this way, the above method also includes the steps shown in Figure 3:

Step 141: Obtain the feature importance of each feature in the user features from the decision tree layer;

Step 142: For the target feature whose feature importance is greater than the importance threshold, when the user feature of the user matches the target feature, take the target measure corresponding to the target feature.

Taking user characteristics including user age, gender, geographical location, current speaking intention, speech rate, total conversation duration, and conversation interval as an example, user characteristics include multiple characteristics. The feature importance of each of the above features can be obtained from the decision tree layer. According to the value of the importance of the feature, the cause of the complaint can be analyzed.

As an example, if the feature of user age is the most important, or is greater than a preset importance threshold, it means that the age factor is one of the reasons for the complaint. Users who meet certain age criteria tend to make complaints. For example, in some scenarios, older users may not be able to keep up with the speaking speed of the smart customer service, and complain because they cannot hear the content of the conversation of the smart customer service clearly. In this case, user age is the target feature. When the user feature of a certain user matches the target feature, for example, the age of a certain user is greater than a preset age threshold, a target measure corresponding to the user's age is taken. The target measures corresponding to the user's age include multiple types, for example, including adopting a dialogue mode with a slow speaking speed to have a dialogue with the user, and for example, including transferring the user to a human service.

In this way, the user's static characteristics, including the user's age, are acquired when the man-machine dialogue is started. If the user's age is greater than the preset age threshold, a dialogue mode with a slower speech rate will be used to have a dialogue with the user, or the user will be directly transferred to a manual service.

As an example, if the feature of the intention of the current speech has the greatest importance, or is greater than a preset importance threshold, it indicates that the intention factor is one of the reasons for the complaint. Users who show some intent or intentions tend to make complaints. For example, in a telemarketing scenario, for a user who shows the intention to end the conversation, if the intelligent customer service still insists on selling the product, the user may complain. In this case, intent is the target feature. When the intent feature in the user profile matches the target feature, the target measure corresponding to the intent is taken, such as ending the conversation, indicating the next visit, and so on.

In this way, during the man-machine dialogue, the user's dynamic characteristics, including the current speaking intention, are acquired. If the current speaking intention is a preset intention type, such as the intention to end the dialogue, actively end the dialogue with the user. It is worth noting that in this case, even if the complaint probability output by the complaint prediction model is small, in order to ensure that complaints are avoided, the target measures corresponding to the target characteristics are implemented.

For example, if the characteristic of the total duration of the conversation is the most important, or is greater than a preset importance threshold, it indicates that the total duration of the conversation is one of the reasons for the complaint. For example, in the smart customer service scenario of express delivery, when the total conversation time is longer than the preset time threshold, it means that it takes a long time to solve the problems raised by users, which leads to complaints from users. In this case, the total duration of the dialogue is the target feature. When the user characteristics of a certain user match the target characteristics, for example, when the total conversation duration of a certain user is greater than the time threshold, the target measure corresponding to the total conversation duration is taken. For example, it includes automatically transferring artificial agents to serve users and so on.

In this embodiment, the cause of the complaint is analyzed by obtaining the feature importance of each feature from the decision tree layer. In this case, regardless of the predicted complaint probability value, when the user feature matches the target feature, then Take corresponding targeted measures to avoid complaints, increase the intelligence of man-machine dialogue, and avoid complaints.

In some embodiments, the above method also includes:

Step 150: If the complaint probability is greater than the preset probability threshold, execute preset early warning measures.

In step 130, after the complaint prediction model outputs the complaint probability of the user under the current speech, the complaint probability may be further compared with the probability threshold. If the complaint probability is greater than the preset probability threshold, it indicates that the user has a high tendency to complain, and thus the preset early warning measures can be implemented.

For example, the early warning measures may include outputting complaint warning information to the artificial agent, so as to prompt the artificial agent to actively intervene in the dialogue and provide services for the user. For another example, the early warning measures may include proactively asking the user whether a human agent service is required, and the like.

In this embodiment, it is judged whether to take early warning measures based on the complaint probability. In the case of a high probability of complaints, the occurrence of complaints can be avoided by taking early warning measures before complaints are generated, thereby increasing the intelligence of man-machine dialogue.

Further, in order to improve the accuracy of early warning, in some embodiments, the above method also includes the steps shown in Figure 4:

Step 161: Obtain the text sequence of the current speech;

Step 162: Perform keyword matching and/or sentiment analysis on the text sequence;

Step 163: If it is determined according to the processing result that the user shows an intention to complain under the current utterance, execute the early warning measures.

In the man-machine dialogue, the voice data of the user's current speech can be obtained. The voice data can be converted into text through the ASR (Automatic Speech Recognition, automatic speech recognition) technology, and the text sequence of the current speech can be obtained.

Then, keyword matching and/or sentiment analysis can be performed on the text sequence, and whether to implement early warning measures can be judged according to the processing results.

Among them, in the keyword matching process, keywords that generate complaints, such as keywords expressing dissatisfaction and attacks, can be collected according to historical complaint data, and a keyword library can be established in advance. Then match the text sequence with the keyword database. If the match is successful, it is determined that the user expresses the intention to complain under the current speech, and any of the above-mentioned early warning measures are executed.

In sentiment analysis processing, a trained sentiment analysis model can be used to analyze the sentiment information carried in the text sequence. For example, sentiment information may include positive sentiment, negative sentiment, and no sentiment. Optionally, the sentiment analysis model may include but not limited to a deep learning model, such as TextCNN (Convolutional Neural Networks, Convolutional Neural Networks). If the analyzed text sequence carries negative emotions, it is determined that the user expresses the intention to complain under the current speech, and any of the above-mentioned early warning measures are implemented.

In this embodiment, through keyword matching and/or sentiment analysis technology, it is further analyzed whether the user expresses a complaint intention, and early warning measures are implemented.

Further, in some embodiments, the complaint probability prediction, keyword matching, and sentiment analysis are not executed sequentially, and may also be executed synchronously. As long as one or more of the conditions that the complaint probability is greater than the probability threshold, the keyword matching is successful, and the sentiment analysis result is negative, the early warning measures are implemented.

In some other embodiments, keyword matching and/or sentiment analysis may be performed before complaint probability prediction. That is, firstly, keyword matching and/or sentiment analysis are performed on the user's current utterance, and after determining that the user has no intention of complaining under the current utterance according to the processing results, the user's complaint probability under the current utterance is predicted.

Exemplarily, as shown in FIG. 5 , in the man-machine dialogue, the text sequence of the user's current speech is obtained (step S1 ). Subsequently, use the pre-established keyword library to perform keyword matching on the text sequence (step S2), and judge whether the matching is successful (step S3). If the matching is successful, an early warning measure is executed (step S4); if the matching is not successful, sentiment analysis is performed on the text sequence (step S5). According to the processing result of sentiment analysis, it is judged whether the text sequence carries negative emotion (step S6). If so, execute early warning measures (step S4); otherwise, obtain the user's static and dynamic features to obtain user characteristics, and input the user characteristics into the complaint prediction model to obtain the complaint probability of the current speech (step S7). Subsequently, it is judged whether the complaint probability is greater than a preset probability threshold. If yes, execute early warning measures (step S4); if not, acquire the user's next utterance as the current utterance, and return to step S1 until the man-machine dialogue ends.

In this embodiment, when using the complaint prediction model for complaint prediction, first perform keyword matching and/or sentiment analysis processing on the text sequence of the current speech, as long as the keyword matching is successful or the sentiment analysis result is a negative emotion, that is, execute Early warning measures. Compared with predicting the probability of complaints, keyword matching results and sentiment analysis results can more accurately and quickly reflect users' complaint intentions. Therefore, performing keyword matching and/or sentiment analysis first can quickly determine whether to implement early warning measures and avoid complaints. produce.

Based on the complaint prediction method provided in any of the above embodiments, the present application also provides a complaint prediction device, as shown in FIG. 6 , the device 600 includes:

An acquisition module 610, configured to acquire preset static features used to characterize the user and dynamic features used to characterize the user's current speech;

A determining module 620, configured to determine user characteristics according to the static characteristics and the dynamic characteristics;

The prediction module 630 is configured to input the user characteristics into the trained complaint prediction model, and obtain the complaint probability of the user under the current utterance.

In some embodiments, the static features include one or more of the following features: user age, gender, geographic location;

The dynamic feature includes one or more of the following features: the intention of the current speech, the speech rate, the total duration of the dialogue, and the dialogue interval.

In some embodiments, the complaint prediction model includes a decision tree layer and a linear regression layer; the prediction module 630 is specifically used for:

inputting said user characteristics into said decision tree layer;

The features extracted by the leaf nodes of the decision tree layer are input into the linear regression layer;

Obtain the complaint probability output by the linear regression layer.

In some embodiments, prediction module 630 is also used to:

Obtaining the feature importance of each feature in the user features from the decision tree layer;

For a target feature whose feature importance is greater than an importance threshold, when the user feature of the user matches the target feature, a target measure corresponding to the target feature is taken.

In some embodiments, device 600 also includes:

An early warning module, configured to execute preset early warning measures if the complaint probability is greater than a preset probability threshold.

In some embodiments, the device 600 also includes a processing module for:

Acquiring the text sequence of the current speech;

performing keyword matching and/or sentiment analysis processing on the text sequence;

If it is determined according to the processing result that the user shows an intention to complain under the current speech, the early warning measures are executed.

In some embodiments, before said inputting said user features into the trained complaint prediction model, performing said keyword matching and/or sentiment analysis processing;

If it is determined according to the processing result that the user has no complaint intention under the current utterance, the step of inputting the user characteristics into the trained complaint prediction model is performed.

For the implementation process of the functions and effects of each module in the above-mentioned device, please refer to the implementation process of the corresponding steps in the above-mentioned method for details, and details will not be repeated here.

Based on the complaint prediction method described in any of the above embodiments, the present application also provides a schematic structural diagram of an electronic device as shown in FIG. 7 . As shown in Figure 7, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and of course may also include hardware required by other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, so as to realize a complaint prediction method described in any of the above embodiments.

The present application also provides a computer storage medium, where a computer program is stored in the storage medium, and when the computer program is executed by a processor, it can be used to implement a complaint prediction method described in any of the above embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed devices and methods may also be implemented in other ways. The device embodiments described above are only illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functions and possible implementations of devices, methods and computer program products according to multiple embodiments of the present application. operate. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or part of code that includes one or more Executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks in succession may, in fact, be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by a dedicated hardware-based system that performs the specified function or action , or may be implemented by a combination of dedicated hardware and computer instructions.

In addition, each functional module in each embodiment of the present application may be integrated to form an independent part, each module may exist independently, or two or more modules may be integrated to form an independent part.

If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes. .

The above descriptions are only examples of the present application, and are not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and changes may be made to the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall be included within the protection scope of this application. It should be noted that like numerals and letters denote similar items in the following figures, therefore, once an item is defined in one figure, it does not require further definition and explanation in subsequent figures.

The above is only a specific implementation of the application, but the scope of protection of the application is not limited thereto. Anyone familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the application. Should be covered within the protection scope of this application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is a relationship between these entities or operations. There is no such actual relationship or order between them. Furthermore, the term "comprises", "comprises" or any other variation thereof is intended to cover a non-exclusive inclusion such that a process, method, article, or apparatus comprising a set of elements includes not only those elements, but also includes elements not expressly listed. other elements of or also include elements inherent in such a process, method, article, or device. Without further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

Claims (10)

1. A complaint prediction method applied to a human-computer conversation scenario, the method comprising:

acquiring preset static characteristics used for representing a user and preset dynamic characteristics used for representing the current speech of the user;

determining user characteristics according to the static characteristics and the dynamic characteristics;

and inputting the user characteristics into the trained complaint prediction model, and acquiring the complaint probability of the user under the current speech.

2. The method of claim 1, wherein the static features comprise one or more of the following: user age, gender, geographic location;

the dynamic characteristics include one or more of the following: the current speaking intention, the speaking speed, the total conversation duration and the conversation interval.

3. The method of claim 1, wherein the complaint prediction model comprises a decision tree layer and a linear regression layer; the inputting the user characteristics into the trained complaint prediction model to obtain the complaint probability of the user under the current speech includes:

inputting the user features into the decision tree layer;

inputting the features extracted from the leaf nodes of the decision tree layer into the linear regression layer;

and obtaining the complaint probability output by the linear regression layer.

4. The method of claim 3, further comprising:

obtaining feature importance of each of the user features from the decision tree layer;

and aiming at the target features with feature importance larger than an importance threshold, when the user features of the user are matched with the target features, taking target measures corresponding to the target features.

5. The method of claim 1, further comprising:

and if the complaint probability is greater than a preset probability threshold, executing a preset early warning measure.

6. The method of claim 5, further comprising:

acquiring a text sequence of the current speech;

processing keyword matching and/or emotion analysis on the text sequence;

and if the user shows the complaint intention under the current speech according to the processing result, executing the early warning measure.

7. The method of claim 6, wherein said keyword matching and/or emotion analysis processing is performed prior to said inputting said user features into said trained complaint prediction model;

and if the user does not express the complaint intention under the current speech according to the processing result, executing the step of inputting the user characteristics into the trained complaint prediction model.

8. A complaint prediction apparatus, applied to a human-machine conversation scenario, the apparatus comprising:

the device comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring preset static characteristics used for representing a user and preset dynamic characteristics used for representing the current speech of the user;

a determining module for determining a user characteristic according to the static characteristic and the dynamic characteristic;

and the prediction module is used for inputting the user characteristics into the trained complaint prediction model and acquiring the complaint probability of the user under the current speech.

9. An electronic device, characterized in that the electronic device comprises:

a processor;

a memory for storing processor-executable instructions;

wherein the processor, when invoking the executable instructions, implements the operations of any of the methods of claims 1-7.

10. A computer readable storage medium having stored thereon computer instructions which, when executed by a processor, carry out the steps of the method of any one of claims 1 to 7.

Images