CN110769286B - A channel-based recommendation method and device, and storage medium - Google Patents

Abstract

The embodiment of the present application discloses a channel-based recommendation method, device, and storage medium. The method includes: obtaining a period for at least one channel based on acquired playback history data for a channel within at least one preset statistical period based on the acquired viewing history data for the target user within at least one preset statistical period, obtain the periodic viewing video feature model for the target user; the periodic playback video feature model for at least one channel Match the video feature distribution data of the periodic viewing video feature model with the video feature distribution data of the periodic viewing video feature model, and obtain recommendation data for the target user based on the matching result. In this way, on the basis of fully considering the periodic characteristics of the channel playing video programs and the user watching the video programs, the personalized recommendation for the user is realized.

Description

A channel-based recommendation method and device, and storage medium

technical field

The present invention relates to the field of big data processing, and more particularly, to a channel-based recommendation method and device, and a storage medium.

Background technique

Channel recommendation refers to recommending TV channels that users are interested in; however, TV channels will play different video programs in different time periods, and the program content is more widely distributed, so channel recommendation is more difficult than program content recommendation, and the recommendation for TV programs is more difficult. The method is also not suitable for recommendation scenarios for TV channels.

SUMMARY OF THE INVENTION

In view of the above problems, the present invention proposes a channel-based recommendation method, device, and storage medium. In this way, on the basis of fully considering the periodic characteristics of channel playing video programs and users watching video programs, personalized recommendation for users is realized. .

The technical solutions of the embodiments of the present application are implemented as follows:

In a first aspect, an embodiment of the present application provides a channel-based recommendation method, including:

Based on the acquired playback history data for a channel within at least one preset statistical period, a periodic playback video feature model for at least one channel is obtained, where the periodic playback video feature model can characterize the periodicity of the channel's regular video feature distribution data;

Based on the obtained viewing history data for the target user in at least one preset statistical period, a periodic video viewing feature model for the target user is obtained, and the periodic video viewing feature model can represent the periodic regular video features of the target user distribution data;

Matching the video feature distribution data of the periodically playing video feature model for at least one channel with the video feature distribution data of the periodically viewing video feature model, and obtaining recommendation data for the target user based on the matching result.

In a specific example, the obtaining recommendation data for the target user based on the matching result includes:

Obtaining viewing period recommendation data of the target user on the target channel in at least one channel based on the matching result; and/or,

Channel recommendation data of the target user in the selected viewing period is obtained based on the matching result.

In a specific example, the method further includes:

The playback characteristics of the video programs played by the channel in at least one preset statistical period and the video characteristics of the video programs are acquired as the playback history data for the channel.

In a specific example, the step of obtaining the periodic playback video feature model for the channel includes:

Based on the playback history data for the channel, determine the video feature distribution data of the video programs played by the channel in the statistical period, and the probability distribution data of the video programs played by the channel in the statistical period;

Based on the video feature distribution data of the video programs played by the channel in the statistical period, and the probability distribution data of the video programs played by the channel in the statistical period, a periodically played video feature model for the channel is obtained.

In a specific example, the method further includes:

Acquiring viewing characteristics and video characteristics of video programs watched by the target user in at least one preset statistical period, as viewing history data for the target user;

or,

Acquiring viewing characteristics and video characteristics of video programs watched by the target user in at least one preset statistical period, as the first sub-viewing history data for the target user;

Obtain the viewing characteristics and video characteristics of the video programs watched by the target user within the determined neighborhood time as the second sub-viewing history data for the target user; wherein the neighborhood time is related to the preset statistical period time of connection;

The first sub-viewing history data and the second sub-viewing history data are used as viewing history data for the target user.

In a specific example, the step of obtaining the feature model of periodically watching videos for the target user includes:

Based on the viewing history data for the target user, determine the video feature distribution data of the video programs watched by the target user at least in the statistical period, and the probability distribution data of the video programs watched by the target user at least in the statistical period;

Based on the video feature distribution data of the video programs watched by the target user at least in the statistical period, and the probability distribution data of the video programs watched by the target user at least in the statistical period, a periodic viewing video feature model for the target user is obtained.

In a specific example, the video feature distribution data of the periodic playing video feature model for at least one channel is matched with the video feature distribution data of the periodic viewing video feature model, and based on the matching result, the target user is obtained. Recommended data, including:

determining the feature difference between the video feature distribution data representing the periodicity of the channel and the video feature distribution data representing the periodicity of the target user;

The feature difference value is sorted, and the data whose sorting result satisfies the preset rule is used as the recommended data.

In a second aspect, an embodiment of the present application provides a channel-based recommendation device, including:

A model processing unit, configured to obtain a periodic play video feature model for at least one channel based on the acquired play history data for the channel within at least one preset statistical period, where the periodic play video feature model can characterize the channel period Regular video feature distribution data; based on the acquired viewing history data for the target user within at least one preset statistical period, obtain a periodic viewing video feature model for the target user, and the periodic viewing video feature model can represent Periodic and regular video feature distribution data of target users;

The recommending unit is configured to match the video feature distribution data of the periodically playing video feature model for at least one channel with the video feature distribution data of the periodically viewing video feature model, and obtain recommendation data for the target user based on the matching result.

In a specific example, the recommending unit is further configured to obtain, based on the matching result, the viewing period recommendation data of the target user on the target channel in at least one channel; and/or, obtain the channel of the target user in the selected viewing period based on the matching result Recommended data.

In a specific example, the model processing unit is further configured to acquire playback characteristics of video programs played by the channel in at least one preset statistical period, and video characteristics of the video programs, as playback history data for the channel.

In a specific example, the model processing unit is also used to:

Based on the playback history data for the channel, determine the video feature distribution data of the video programs played by the channel in the statistical period, and the probability distribution data of the video programs played by the channel in the statistical period;

Based on the video feature distribution data of the video programs played by the channel in the statistical period, and the probability distribution data of the video programs played by the channel in the statistical period, a periodically played video feature model for the channel is obtained.

In a specific example, the model processing unit is also used to:

Acquiring viewing characteristics and video characteristics of video programs watched by the target user in at least one preset statistical period, as viewing history data for the target user;

or,

Acquiring viewing characteristics and video characteristics of video programs watched by the target user in at least one preset statistical period, as the first sub-viewing history data for the target user;

Obtain the viewing characteristics and video characteristics of the video programs watched by the target user within the determined neighborhood time as the second sub-viewing history data for the target user; wherein the neighborhood time is related to the preset statistical period time of connection;

The first sub-viewing history data and the second sub-viewing history data are used as viewing history data for the target user.

In a specific example, the model processing unit is also used to:

Based on the viewing history data for the target user, determine the video feature distribution data of the video programs watched by the target user at least in the statistical period, and the probability distribution data of the video programs watched by the target user at least in the statistical period;

Based on the video feature distribution data of the video programs watched by the target user at least in the statistical period, and the probability distribution data of the video programs watched by the target user at least in the statistical period, a periodic viewing video feature model for the target user is obtained.

In a specific example, the recommending unit is further configured to determine the feature difference between the video feature distribution data representing the periodicity of the channel and the video feature distribution data representing the periodicity of the target user; perform sorting processing on the feature difference, The data whose sorting result satisfies the preset rules is used as the recommended data.

In a third aspect, an embodiment of the present application further provides a channel-based recommendation device, including:

one or more processors;

a memory communicatively coupled to the one or more processors;

one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs are configured to perform the above the method described.

In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and when the program is executed by a processor, implements the above-mentioned method.

The embodiment of the present application fully considers the periodicity of the video programs played by the channel and the video programs watched by the user, and makes personalized recommendations based on the periodicity. Therefore, the embodiment of the present application is more in line with the law of the video program played by the channel and the viewing by the user. The characteristics of video programs, and on the basis of realizing personalized recommendation, the user experience is improved.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present application more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained from these drawings without creative effort.

FIG. 1 shows a schematic flowchart of a channel-based recommendation method according to an embodiment of the present application;

2 shows a schematic diagram of a preset statistical period in a specific application scenario according to an embodiment of the present application;

FIG. 3 shows a schematic structural diagram of a logical unit of a channel-based recommendation apparatus according to an embodiment of the present application.

FIG. 4 shows a schematic diagram of a hardware structure of a channel-based recommendation apparatus according to an embodiment of the present application.

Detailed ways

In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application.

In some of the processes described in the description and claims of the present invention and the above-mentioned drawings, various operations are included in a specific order, but it should be clearly understood that these operations may not be in accordance with the order in which they appear herein. For execution or parallel execution, the sequence numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the sequence numbers themselves do not represent any execution order. Additionally, these flows may include more or fewer operations, and these operations may be performed sequentially or in parallel. It should be noted that the descriptions such as "first" and "second" in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequence, nor do they limit "first" and "second" are different types.

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts shall fall within the protection scope of the present invention.

The video programs of TV channels have strong periodic characteristics, which are mainly reflected in two aspects. On the one hand, TV channels are cyclical in programming programs. Arrange to broadcast news programs; on the other hand, users' viewing interests are characterized by periodic evolution. For example, they mainly watch news programs on weekdays, and tend to watch more entertaining programs on weekends. Therefore, based on the above periodic characteristics, the embodiments of the present application provide a channel-based recommendation method and device, and a storage medium, which deeply digs the evolution characteristics of the video programs played on the channel and the video programs watched by the user during the period, and then The mining results (such as periodic video features) are used as the basis for TV channel recommendation. In this way, channel recommendation for users is completed, which satisfies the viewing needs of different users and lays a foundation for improving user experience.

Specifically, FIG. 1 shows a schematic flowchart of a channel-based recommendation method according to an embodiment of the present application; as shown in FIG. 1 , the method includes:

Step 101: Based on the acquired playback history data for channels within at least one preset statistical period, obtain a periodic playback video feature model for at least one channel, where the periodic playback video feature model can characterize the periodicity of the channel. Video feature distribution data.

In practical applications, the preset statistical period described in the embodiments of the present application may be set according to actual scenarios. For example, the preset statistical period is one month. Two months, or the playback history data of the last several months, in this way, it is convenient to dig out the periodicity rule based on the obtained playback history data.

Here, the preset statistical period may also include multiple time dimensions, for example, three time dimensions, namely month, week, and day, so that even if only the data in one preset statistical period is counted, the The periodicity of video programs played by the channel lays the foundation for targeted channel recommendation based on the periodicity.

It should be noted that, in the embodiment of the present application, the periodic playing video feature model is a model corresponding to a channel. In other words, after the preset statistical period is determined, each channel corresponds to a periodic playing video feature model. Therefore, in an actual scenario, in order to improve the accuracy of channel recommendation, it is necessary to determine the periodic play video feature models of multiple different channels.

In a specific example, the playback history data for the channel at least includes: playback characteristics and video characteristics of the video programs played by the channel in at least one preset statistical period; that is, before step 101, the method described in this embodiment of the present application further includes: : Acquire the playback characteristics of the video programs played by the channel in at least one preset statistical period, and the video characteristics of the video programs, as the playback history data for the channel.

In a specific example, the step of obtaining the periodic playback video feature model for the channel specifically includes: based on the playback history data for the channel, determining the video feature distribution data of the video programs played by the channel in the statistical period, and The probability distribution data of the video programs played in the statistical period; based on the video feature distribution data of the video programs played by the channel in the statistical period, and the probability distribution data of the video programs played by the channel in the statistical period, get the channel-specific data. Play the video feature model periodically. In practical applications, when there are multiple channels, the periodically playing video feature model of each channel can be based on this example to determine the periodically playing video feature model according to the respective playing history data of each channel.

In actual scenarios, the video features described in the embodiments of the present application may specifically include topic tags. For example, when the video program is a news broadcast, the topic tags may be "news" and "current affairs". At this time, the video features specifically include: " News", "Current Affairs", in this way, the content characteristics presented by the video program can be obtained by using the subject tag. Of course, in an actual scenario, a video program may correspond to one or at least two topic tags, and the topic tags of different video programs may be the same or different.

Step 102: Based on the acquired viewing history data for the target user within at least one preset statistical period, obtain a periodic video viewing feature model for the target user, where the periodic video viewing feature model can represent the periodicity of the target user video feature distribution data.

In a specific example, the viewing history data for the target user includes: viewing characteristics and video characteristics of the video programs watched by the target user in at least one preset statistical period, that is, before step 102, the method described in this embodiment of the present application further: The method includes acquiring the viewing characteristics and video characteristics of the video programs watched by the target user in at least one preset statistical period, as viewing history data for the target user.

In another specific example, considering that the viewing behavior of the target user may have strong sparseness, it is impossible to obtain a periodic video viewing feature model for the target user. In this example, the probability smoothing technology is used. The basic idea is to use the target user. The viewing records in the neighborhood time of the preset statistical period are used to supplement the viewing records of the target users in the preset statistical period. In this way, it is avoided that there is a sparsity problem in the acquired viewing records, which makes it impossible to determine the periodic viewing video feature model. Based on this, the viewing history data for the target user includes: the viewing characteristics and video characteristics of the video programs watched by the target user in at least one preset statistical period, and the viewing characteristics of the video programs watched by the target user within the neighborhood time. and video features. That is, before step 102, the method described in this embodiment of the present application further includes:

Obtain the viewing characteristics and video characteristics of the video programs watched by the target user in at least one preset statistical period, as the first sub-viewing historical data for the target user; obtain the video programs watched by the target user within the determined neighborhood time The viewing characteristics and video characteristics of the target user are used as the second sub-viewing historical data for the target user; wherein, the neighborhood time is the time associated with the preset statistical period; the first sub-viewing historical data and the first sub-viewing historical data are Erzi viewing history data is used as viewing history data for target users.

Here, it should be noted that the neighborhood time may be a specific time period; the specific time period is determined based on the preset statistical period, for example, when the preset statistical period corresponds to the first time dimension, the first time dimension One or at least two first dimensions are included. In this case, the second time dimension corresponding to the neighborhood time is smaller than the first time dimension, and the second dimension included in the second time dimension is the first time dimension included in the first time dimension. In this way, the number of viewing records can be enriched by using the neighborhood time, thereby laying a foundation for avoiding the problem of data sparsity.

Or, the neighborhood time is all time periods except the preset statistical period, and at this time, the acquired second sub-viewing history data is the viewing characteristics and videos of all video programs watched by the target user except for the preset statistical period. feature.

For example, the first time dimension corresponding to the preset statistical period is 3, that is, it includes three first dimensions of month, week, and day. In this case, the second time dimension corresponding to the neighborhood time may be 2, 1, or less than 0. In the first time dimension; further, if the second time dimension is 2, it may specifically include two second dimensions of month and week, or two second dimensions of month and day, or two second dimensions of week and day ; If the second time dimension is 1, the second dimension is month or week or day. Or, the second time dimension is 0. At this time, the neighborhood time is all time periods except the preset statistical period. Correspondingly, the second sub-viewing historical data obtained based on the neighborhood time statistics is the target user Viewing characteristics and video characteristics of all video programs watched outside the preset statistical period.

It should be noted that the above-mentioned month, week, and day may correspond to a natural month, a natural week, and a natural day.

In a specific example, the step of obtaining the periodic viewing video feature model for the target user includes: based on the viewing history data for the target user, determining the video feature distribution data of the video programs watched by the target user at least in the statistical period, And the probability distribution data of the video programs watched by the target user at least in the statistical period; based on the video feature distribution data of the video programs watched by the target user at least in the statistical period, and the video programs watched by the target user at least in the statistical period The probability distribution data of , obtain the periodic watching video feature model for target users.

In another specific example, when the viewing history data includes the first sub-viewing history data and the second sub-viewing history data, at this time, the step of obtaining the periodic viewing video feature model for the target user includes: based on the viewing of the target user Historical data, determine the video feature distribution data of the video programs watched by the target users in the statistical period and the neighborhood time, and the probability distribution data of the video programs watched by the target users in the statistical period and the neighborhood time; based on The video feature distribution data of the video programs watched by the target users in the statistical period and the neighborhood time, as well as the probability distribution data of the video programs watched by the target users in the statistical period and the neighborhood time, to obtain the period for the target user Sexual viewing video feature model.

Step 103 : Match the video feature distribution data of the periodically playing video feature model for at least one channel with the video feature distribution data of the periodically viewing video feature model, and obtain recommendation data for the target user based on the matching result.

In a specific example, the recommendation data obtained for the target user based on the matching result includes two types, respectively:

Type 1: Based on the matching result, the target user's viewing period recommendation data on the target channel in at least one channel is obtained;

Type 2: Obtain the channel recommendation data of the target user during the selected viewing period based on the matching result.

That is to say, the channel recommendation for the target user includes two forms, that is, the viewing time recommendation on the target channel, or the channel recommendation at a specific viewing time, thus enriching the application scenarios. In practical applications, one can be selected and executed based on the actual scenario.

In another specific example, the following matching methods may be used to determine the recommended data, specifically, determining the feature difference between the video feature distribution data representing the periodicity of the channel and the video feature distribution data representing the periodicity of the target user; The feature difference value is sorted, and the data whose sorting result satisfies the preset rule is used as the recommended data. For example, determine the relative entropy between two video feature distribution data, and then measure the matching degree between the models based on the relative entropy, and then obtain recommendation data based on the matching degree.

The embodiment of the present application fully considers the periodicity of the video programs played by the channel and the video programs watched by the user, and performs personalized recommendation based on the periodicity. Therefore, the channel-based recommendation method in the embodiment of the present application is more in line with the video program played by the channel. On the basis of personalized recommendation, the user experience is improved.

The embodiments of the present application will be described in further detail below in conjunction with specific application scenarios; specifically,

In the embodiment of the present application, multiple dimensions are used to represent the preset statistical period. In this way, the historical data (playing historical data and viewing historical data) obtained by statistics in the preset statistical period can be used to obtain the periodicity of the arrangement of video programs played by TV channels. and the periodicity of user viewing. For example, Figure 2 shows a schematic diagram of using the weekly dimension and the daily dimension to represent the preset statistical period. As shown in Figure 2, the weekly dimension is divided into Monday to Sunday, and the daily dimension is divided by time period. It is divided into morning, noon, afternoon, evening, gold, and late night, etc., and then any moment can be represented as an element in a two-dimensional matrix. For example, if the target user watched a specific channel at 19:14 on October 12, 2019, then The periodic feature of the target user's viewing behavior is represented as (Saturday, evening). Similarly, the viewing history data or playback history data obtained within the preset statistical period can be represented by the periodic feature structure representation method.

It should be noted that, it can be seen from the above description that the periodic feature structure is determined based on the dimensional features of the preset statistical period. Further, FIG. 2 is only an example, and in practical applications, more dimensions (such as month, week, day, etc.) can be used to represent the preset statistical period and the periodic feature structure. There may be more flexible division methods, for example, the weekly dimension may be divided into weekdays and weekends, and the daily dimension may be divided into 24 time periods according to hours, etc., which is not limited in this embodiment of the present application.

Based on the above-mentioned periodic feature structure representation method, the method of the embodiment of the present application will be further described. First of all, it should be noted that this example uses topic tags to represent video features; further, the following Pr(·|·) represents the conditional probability, Pr( ) represents the probability, c represents the channel (such as a digital TV channel), l represents the video feature of the video program (such as a hashtag), p represents the video program, u represents the user, and t represents the use of the aforementioned periodic features The time represented by the structure. Specifically, the method includes:

Step 1: Build a periodic video feature model for the channel.

Here, it should be noted that, in practical applications, each channel corresponds to a periodic playing video feature model, and each periodic playing video feature model can be implemented in the following manner.

First, a TV channel is selected, and the playing history data of the video programs played by the TV channel within a certain period of time (for example, one or at least two preset statistical periods) are collected. Here, when the preset statistical period is represented by the above periodic feature structure, the collected time period is an integer multiple of the maximum granularity dimension representing the preset statistical period, that is, the statistical duration is the maximum granularity dimension of the preset statistical period For example, three dimensions of "month, week, and day" are used to represent the preset statistical period. At this time, the duration of data collection (that is, the statistical duration of the preset statistical period) is required to be the maximum granularity "monthly" ”, such as N months, where N is a positive integer.

Here, the play history data records the start and end time of each video program played by the TV channel, and the subject tag of each video program. For example, CCTV 1 broadcasts the news broadcast from 19:00 to 19:30 on October 12, 2019, and the topic tags of the news broadcast are "news", "current affairs", etc. In practical applications, each video program corresponds to at least one topic tag. Of course, in order to accurately represent the characteristics of the video program, multiple topic tags may also be used to mark the video program, that is, one video program may also correspond to multiple topic tags.

Secondly, based on the playback history data, using the probability distribution feature of the channel's hashtags, a periodic playback video feature model for the channel is constructed, and the model is characterized as follows:

Here, Equation (1) is obtained by Bayesian transformation of Pr(l|c; t); Equation (2) is obtained by using the TV program set Prog(c, t) broadcast by channel c at time t It is obtained by characterizing the channel and expanding it using the probability addition formula; Equation (3) is obtained by Bayesian transformation of Pr(p|l) in Equation (2); Equation (4) is obtained from Equation (3) ) is obtained by reducing the same items in the numerator and denominator, and adding the normalization factor Z _c . Here, after increasing the normalization factor Z _c , the final calculated value can meet the requirements of the probability distribution, Z _c The specific calculation method is: Z _c =∑ _l′ ∈L ∑ _{p∈Prog(c,t)} Pr(l|p)·Pr(p), where L is the complete set of topic tags.

It can be seen from the above-mentioned periodic playing video feature model that the distribution of topic tags of a TV channel in a specific time period depends on the distribution of topic tags Pr(l|p) of the video programs played by the TV channel in the specific time period, And the probability distribution of each video program. The former can be directly obtained from the broadcast history data of the TV channel; the latter can be calculated according to the periodic characteristics of the TV channel in the preset statistical period, for example, according to the proportion of the duration of the video program played in the preset statistical period. Assuming that the statistical duration of the preset statistical period is 120 minutes, and the video program is played for 60 minutes, then Pr(p)=60/120=0.5.

Step 2: Build a periodic video feature model for target users.

First, select the target user, and collect the TV viewing history data watched by the target user within a certain period of time. The specific collection strategy is similar to collecting the broadcast history data of TV channels, that is, the collection time period is an integer multiple of the maximum granularity dimension representing the preset statistical period. For example, two dimensions of "week and day" are used to represent the preset statistics Period, at this time, the duration of data collection (that is, the statistical duration of the preset statistical period) is required to be an integer multiple of the maximum granularity "week", such as M weeks, where M is a positive integer.

Here, the viewing history data records the start and end time of the target user's viewing of the video program and the subject tag of each video program, so that the viewing history data can be used to determine the target user's interest in watching TV.

Secondly, based on the viewing history data, using the probability distribution feature of the target user's hashtags, a feature model of periodically watching videos is constructed. Here, a derivation method similar to the periodic video feature model is used to obtain the periodic video feature model, which is characterized as follows:

Here, Prog(u, t) in Equation (5) represents the list of video programs watched by the target user u within time t. Pr(l|p), Pr(p) and Z _u are calculated in the same way as the periodic playback video feature model.

It can be seen from the above-mentioned periodic watching video feature model that the distribution of hashtags of the target user in a specific time period depends on the distribution of hashtags Pr(l|p) of the video programs watched by the target user in the specific time period, And the probability distribution of each video program. The former can be obtained directly in the viewing history data; the latter can be calculated according to the periodic characteristics of the video program watched by the target user in the preset statistical period, for example, calculated according to the time-length ratio of the video program watched within the preset statistical period, Specifically, assuming that the statistical duration of the preset statistical period is 120 minutes, and the watched video program is played for 60 minutes, then Pr(p)=60/120=0.5.

In actual scenarios, user viewing behavior usually has strong sparseness, that is, users have no viewing history in multiple preset statistical periods. At this time, Prog(u,t) will be an empty set, which will lead to Pr(l|u;t) cannot be computed efficiently. Therefore, to solve this problem, the embodiment of the present application adopts a probabilistic smoothing technique, and the basic idea is to use the user's viewing records in the neighborhood time of the preset statistical period to supplement the user's viewing records in the preset statistical period.

For example, in a specific example, the smoothed periodic viewing video feature model is:

Prog(u, t') in formula (6) represents the list of video programs watched by the target user u in the neighborhood time t' of the time t in the preset statistical period, and Prog(u) represents all the videos watched by the target user u. video program. Neig(t) is the set of all neighborhood times at time t of the preset statistical period. w ₁ , w ₂ and w ₃ are weights, which can be empirical values, satisfying the conditions:

0<w ₁ <1, 0<w ₂ <1, 0<w ₃ <1, w ₁ +w ₂ +w ₃ =1.

Here, the neighborhood time is set as:

Using the dimension of the preset statistical period represented by d dimensions, at this time, the preset statistical period can be expressed as: t=(t ₁ , t ₂ . . . , t _d-1 , t _d ); then the neighborhood time set is:

Neig(t)={(t ₁ ,...,t _d-1 ),(t ₁ ,...,t _d-2 ,t _d ),...,(t ₂ ,...,t _d )};

That is, the dimension of the domain time is any d-1 dimension of t. In other words, the dimension of the neighborhood time is lower than the dimension of the preset statistical period; here, d is a positive integer. For example, for a 2-dimensional preset statistical period t=(Wednesday, evening), its neighborhood time set Neig(t)={(Wednesday), (evening)}.

Of course, in practical applications, the dimension of the neighborhood time can also be d-1, d-2, ..., 1, or even the dimension of the neighborhood time is 0. In this case, the neighborhood time statistics are calculated by the user except the preset statistical period. The viewing history data of all video programs watched outside; it should be noted that the embodiment of the present application does not impose too many restrictions on the dimension of the neighborhood time, as long as the dimension of the neighborhood time is less than the dimension of the preset statistical period, supplementary viewing can be realized Historical data can solve the sparsity problem.

Here, because the scope of the domain time is larger, it usually has more viewing history data than the original preset statistical period, that is, the sparsity is lower. Therefore, the sparsity problem of the user's viewing behavior can be overcome to a certain extent.

Moreover, this example further provides the case where the neighborhood time is 0, that is, when the dimension of the neighborhood time is 0, all the viewing history data Prog(u) of the user can be obtained by statistics. For users of , Prog(u) is a non-empty set, which lays a foundation for effectively calculating the smoothed periodic viewing video feature model Pr′(l|u;t).

Step 3: Matching the periodic playing video feature model and the periodic viewing video feature model to obtain recommendation data for the target user.

Specifically, in this embodiment of the present application, the relative entropy between the hashtag probability distribution of the TV channel and the target user's hashtag probability distribution is used to measure the matching degree between the periodic playing video feature model and the periodic viewing video feature model, The specific calculation method is:

In this way, based on formula (7), the Dist(c, u; t) of all candidate channels c for the target user can be calculated, and finally the top M channels with the largest Dist(c, u; t) values are taken and recommended to the user .

Here, first, the embodiment of the present application establishes a periodic feature representation method, by which any time can be represented as a multi-dimensional periodic time, thereby laying a foundation for channel recommendation. Moreover, the periodic feature representation method has good scalability, has no explicit constraints on dimensions, and has no explicit constraints on the specific division of time in each dimension. It can reflect the arbitrary periodic characteristics of time, laying a solid foundation for engineering applications Base.

Secondly, the embodiment of the present application constructs a periodic playing video feature model for the channel based on the video feature probability distribution data of the channel, and constructs a periodic viewing video feature model for the user based on the video feature probability distribution data of the user. The derivation gives a specific calculation method, which can guarantee the correctness of the calculation method in mathematics, and the whole derivation process is simple and interpretable.

Finally, the embodiment of the present application adopts a probability smoothing technology to ensure that any user with a TV viewing history can effectively calculate its distribution, thereby obtaining channel recommendation data, thus improving user experience.

The embodiment of the present application also provides a channel-based recommendation device, as shown in FIG. 3 , the device includes:

The model processing unit 31 is configured to obtain a periodic play video feature model for at least one channel based on the acquired play history data for the channel within at least one preset statistical period, where the periodic play video feature model can characterize the channel Periodic and regular video feature distribution data; based on the obtained viewing history data for the target user within at least one preset statistical period, obtain a periodic viewing video feature model for the target user, where the periodic viewing video feature model can Video feature distribution data representing the periodicity of target users;

The recommending unit 32 is configured to match the video feature distribution data of the periodic playback video feature model for at least one channel with the video feature distribution data of the periodic viewing video feature model, and obtain recommendation data for the target user based on the matching result .

In a specific example, the recommending unit 32 is further configured to obtain, based on the matching result, the viewing period recommendation data of the target user on the target channel in at least one channel; and/or, based on the matching result, obtain the target user's viewing period of the selected viewing period. Channel recommendation data.

In a specific example, the model processing unit 31 is further configured to acquire the playback characteristics of the video programs played by the channel in at least one preset statistical period, and the video characteristics of the video programs, as the playback history data for the channel.

In a specific example, the model processing unit 31 is further configured to:

Based on the playback history data for the channel, determine the video feature distribution data of the video programs played by the channel in the statistical period, and the probability distribution data of the video programs played by the channel in the statistical period;

Based on the video feature distribution data of the video programs played by the channel in the statistical period, and the probability distribution data of the video programs played by the channel in the statistical period, a periodically played video feature model for the channel is obtained.

In a specific example, the model processing unit 31 is further configured to:

Acquiring viewing characteristics and video characteristics of video programs watched by the target user in at least one preset statistical period, as viewing history data for the target user;

or,

Acquiring viewing characteristics and video characteristics of video programs watched by the target user in at least one preset statistical period, as the first sub-viewing history data for the target user;

Obtain the viewing characteristics and video characteristics of the video programs watched by the target user within the determined neighborhood time as the second sub-viewing history data for the target user; wherein the neighborhood time is related to the preset statistical period time of connection;

The first sub-viewing history data and the second sub-viewing history data are used as viewing history data for the target user.

In a specific example, the model processing unit 31 is further configured to:

Based on the viewing history data for the target user, determine the video feature distribution data of the video programs watched by the target user at least in the statistical period, and the probability distribution data of the video programs watched by the target user at least in the statistical period;

Based on the video feature distribution data of the video programs watched by the target user at least in the statistical period, and the probability distribution data of the video programs watched by the target user at least in the statistical period, a periodic viewing video feature model for the target user is obtained.

In a specific example, the recommending unit 32 is further configured to determine the feature difference between the video feature distribution data representing the periodicity of the channel and the video feature distribution data representing the periodicity of the target user; perform sorting processing on the feature difference. , and the data whose sorting result satisfies the preset rules is used as the recommended data.

It should be pointed out here that the descriptions of the above device embodiment items are similar to the above method descriptions, and have the same beneficial effects as the method embodiments, so they will not be repeated. For the technical details not disclosed in the device embodiments of the present invention, those skilled in the art should refer to the descriptions of the method embodiments of the present invention to understand them, and to save space, they will not be repeated here.

In a third aspect, an embodiment of the present application provides a channel-based recommendation device, including:

one or more processors;

a memory communicatively coupled to the one or more processors;

one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs are configured to perform the above the method described.

In a specific example, the channel-based recommendation device described in this embodiment of the present application may be specifically structured as shown in FIG. 4 , the device includes a processor 41, a storage medium 42, and at least one external communication interface 43; the The processor 41 , the storage medium 42 and the external communication interface 43 are all connected through the bus 44 . The processor 41 may be an electronic component with processing functions, such as a microprocessor, a central processing unit, a digital signal processor, or a programmable logic array. Computer-executable codes are stored in the storage medium, and the computer-executable codes can execute the image processing method described in any one of the above embodiments. In practical applications, the obtaining unit 41 , the preliminary screening unit 42 and the text matching unit 43 can all be implemented by the processor 41 .

It should be pointed out here that the descriptions of the above device embodiment items are similar to the above method descriptions, and have the same beneficial effects as the method embodiments, so they will not be repeated. For the technical details not disclosed in the device embodiments of the present invention, those skilled in the art should refer to the descriptions of the method embodiments of the present invention to understand them, and to save space, they will not be repeated here.

Embodiments of the present application provide a computer-readable storage medium, which stores a computer program, and when the program is executed by a processor, implements the following steps:

Based on the acquired playback history data for a channel within at least one preset statistical period, a periodic playback video feature model for at least one channel is obtained, where the periodic playback video feature model can characterize the periodicity of the channel's regular video feature distribution data;

Based on the obtained viewing history data for the target user in at least one preset statistical period, a periodic video viewing feature model for the target user is obtained, and the periodic video viewing feature model can represent the periodic regular video features of the target user distribution data;

Matching the video feature distribution data of the periodically playing video feature model for at least one channel with the video feature distribution data of the periodically viewing video feature model, and obtaining recommendation data for the target user based on the matching result.

It should be pointed out here that the descriptions of the above storage medium embodiment items are similar to the descriptions of the above-mentioned methods, and have the same beneficial effects as the method embodiments, so they will not be repeated. For technical details not disclosed in the embodiments of the storage medium of the present invention, those skilled in the art should refer to the descriptions of the method embodiments of the present invention to understand them. To save space, they will not be repeated here.

In the several embodiments provided in this application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are only illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented. On the other hand, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit. The above-mentioned integrated units may be implemented in the form of hardware, or may be implemented in the form of software functional units.

Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium can include: Read Only Memory (ROM, Read Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk, etc.

Those of ordinary skill in the art can understand that all or part of the steps in the method of the above-mentioned embodiments can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer-readable storage medium. The medium can be read-only memory, magnetic disk or optical disk, etc.

A channel-based recommendation method, device, and storage medium provided by the present invention have been described in detail above. For those skilled in the art, based on the ideas of the embodiments of the present application, there will be specific implementation methods and application scopes. For changes, in summary, the contents of this specification should not be construed as limiting the present invention.

Claims (8)

1. A channel-based recommendation method, the method comprising:

the method comprises the steps of obtaining a periodic playing video feature model for at least one channel based on obtained playing history data for the channel in at least one preset statistical period, wherein the periodic playing video feature model can represent video feature distribution data of the periodic regularity of the channel, and is a model corresponding to the channel, and the method for determining the playing history data of the channel comprises the following steps: acquiring playing characteristics of a video program played by a channel in at least one preset statistical period and video characteristics of the video program as playing history data aiming at the channel;

obtaining a periodic watching video feature model for the target user based on the obtained watching history data for the target user in at least one preset statistical period, wherein the periodic watching video feature model can represent video feature distribution data of the periodic regularity of the target user, and the watching history data for the target user comprises: the method comprises the following steps of obtaining watching characteristics and video characteristics of a video program watched by a target user in at least one preset statistical period, and watching characteristics and video characteristics of a video program watched by the target user in neighborhood time, wherein the method further comprises the following steps: acquiring the watching characteristics and video characteristics of a video program watched by a target user in at least one preset statistical period as first sub-watching history data aiming at the target user; acquiring the watching characteristics and the video characteristics of the video program watched by the target user in the determined neighborhood time as second sub-watching history data aiming at the target user; wherein the neighborhood time is a time associated with the preset statistical period; taking the first sub-viewing history data and the second sub-viewing history data as viewing history data for a target user;

matching video feature distribution data of a periodically played video feature model for at least one channel with video feature distribution data of a periodically watched video feature model, and obtaining recommendation data for a target user based on a matching result;

the method comprises the following steps of obtaining viewing history data or playing history data in a preset statistical period, representing the viewing history data or the playing history data by a periodic feature structure representation method, determining a periodic feature structure based on the dimensional features of the preset statistical period, representing video features by using a theme label based on the periodic feature structure representation method, and the periodic feature structure representation method comprises the following steps:

the method comprises the following steps: constructing a periodic playing video characteristic model aiming at a channel;

step two: constructing a periodic watching video feature model for a target user;

further, each channel in the step one corresponds to one periodically playing video feature model, and each periodically playing video feature model is realized based on the following method:

selecting a television channel, and collecting play history data of a video program played by the television channel in a certain time period; based on the play history data, a periodic play video feature model for the channel is constructed and obtained by using the probability distribution feature of the channel topic label, and the periodic play video feature model of the channel is characterized as follows:

the periodic broadcast video characteristic model of the channel is characterized in that Pr (· c) represents conditional probability, Pr (·) represents probability, c represents channel, l represents video characteristic of video program, p represents video program, u represents user, t represents time represented by the periodic characteristic structure, and the formula (1) is obtained by carrying out Bayesian transformation on Pr (l | c; t); the formula (2) is obtained by representing the channel by using a television program set Prog (c, t) played by the channel c at time t and expanding the channel by using a probability addition formula; the formula (3) is obtained by performing Bayesian transformation on Pr (p | l) in the formula (2); the formula (4) is obtained by reducing the same items in the numerator and denominator in the formula (3) and adding a normalization factor Zc, which is specifically calculated as follows: zc ═ Σ L' ∈ L Σ p ∈ Prog (c, t) Pr (L | p) · Pr (p), where L is the topic label corpus;

further, the second step further comprises the following steps: selecting a target user, and collecting historical data of watching television watched by the target user within a certain time period; based on the watching history data, a periodic watching video feature model is constructed and obtained by using the probability distribution feature of the subject label of the target user, and the periodic watching video feature model is obtained by adopting a derivation mode similar to that of the periodic playing video feature model, and is characterized as follows:

prog (u, t) in formula (5) represents a video program list watched by the target user u within the time t, and Pr (l | p), Pr (p) and Zu adopt the same calculation mode as the feature model of the periodically played video;

carrying out balance processing on the characteristic model of the periodically watched video, wherein the smoothed characteristic model of the periodically watched video is as follows:

prog (u, t ') in equation (6) represents a list of video programs watched by the target user u in a time t' adjacent to the time t of the preset statistical period, and Prog (u) represents all the video programs watched by the target user u; neig (t) is a set of all neighborhood times of time t of a preset statistical period;

the method comprises the following steps of measuring the matching degree between a periodically played video feature model and a periodically watched video feature model by using the relative entropy between the probability distribution of the theme label of the television channel and the probability distribution of the theme label of a target user, wherein the specific calculation mode is as follows:

dist (c, u; t) of all candidate channels c for the target user can be obtained through calculation based on the formula (7), and finally, the first M channels with the maximum Dist (c, u; t) values are selected and recommended to the user.

2. The method of claim 1, wherein obtaining recommendation data for the target user based on the matching result comprises:

obtaining viewing period recommendation data of the target user on a target channel in at least one channel based on the matching result; and/or the presence of a gas in the gas,

and obtaining the channel recommendation data of the target user in the selected watching time period based on the matching result.

3. The method of claim 1, wherein the step of obtaining a periodically playing video feature model for a channel comprises:

based on the play history data aiming at the channel, determining video characteristic distribution data of the video programs played by the channel in the statistical period and probability distribution data of the video programs played by the channel in the statistical period;

and obtaining a periodic playing video feature model aiming at the channel based on the video feature distribution data of the video program played by the channel in the statistical period and the probability distribution data of the video program played by the channel in the statistical period.

4. The method of claim 1, wherein the step of deriving a model of periodically viewed video features for the target user comprises:

based on the watching history data aiming at the target user, determining video characteristic distribution data of the video program watched by the target user at least in the counting period and probability distribution data of the video program watched by the target user at least in the counting period;

and obtaining a periodic watching video feature model aiming at the target user based on the video feature distribution data of the video program watched by the target user at least in the statistical period and the probability distribution data of the video program watched by the target user at least in the statistical period.

5. The method of claim 1, wherein matching video feature distribution data of a periodically playing video feature model for at least one channel with video feature distribution data of a periodically watching video feature model, and obtaining recommendation data for a target user based on the matching result comprises:

determining a characteristic difference value between video characteristic distribution data representing the periodic regularity of a channel and video characteristic distribution data representing the periodic regularity of a target user;

and sequencing the characteristic difference values, and taking data with a sequencing result meeting a preset rule as recommended data.

6. An apparatus for channel-based recommendation, the apparatus comprising:

the model processing unit is used for obtaining a periodic playing video feature model for at least one channel based on the obtained playing history data for the channel in at least one preset statistical period, wherein the periodic playing video feature model can represent video feature distribution data of the periodic regularity of the channel, and is a model corresponding to the channel, and the method for determining the playing history data of the channel comprises the following steps: acquiring playing characteristics of a video program played by a channel in at least one preset statistical period and video characteristics of the video program as playing history data aiming at the channel;

obtaining a periodic watching video feature model for the target user based on the obtained watching history data for the target user in at least one preset statistical period, wherein the periodic watching video feature model can represent video feature distribution data of the periodic regularity of the target user, and the watching history data for the target user comprises: the method comprises the following steps of obtaining watching characteristics and video characteristics of a video program watched by a target user in at least one preset statistical period, and watching characteristics and video characteristics of a video program watched by the target user in neighborhood time, wherein the method further comprises the following steps: acquiring the watching characteristics and video characteristics of a video program watched by a target user in at least one preset statistical period as first sub-watching history data aiming at the target user; acquiring the watching characteristics and the video characteristics of the video program watched by the target user in the determined neighborhood time as second sub-watching history data aiming at the target user; wherein the neighborhood time is a time associated with the preset statistical period; taking the first sub-viewing history data and the second sub-viewing history data as viewing history data for a target user;

viewing history data or playing history data acquired in a preset statistical period are represented by a periodic feature structure representation method, the periodic feature structure is determined based on the dimensional features of the preset statistical period, video features are represented by using a theme label based on the periodic feature structure representation method, and the periodic feature structure representation method comprises the following steps:

the method comprises the following steps: constructing a periodic playing video characteristic model aiming at a channel;

step two: constructing a periodic watching video feature model for a target user;

further, each channel in the step one corresponds to one periodically playing video feature model, and each periodically playing video feature model is realized based on the following method:

selecting a television channel, and collecting play history data of a video program played by the television channel in a certain time period; based on the play history data, a periodic play video feature model for the channel is constructed and obtained by using the probability distribution feature of the channel topic label, and the periodic play video feature model of the channel is characterized as follows:

the periodic broadcast video characteristic model of the channel is characterized in that Pr (· c) represents conditional probability, Pr (·) represents probability, c represents channel, l represents video characteristic of video program, p represents video program, u represents user, t represents time represented by the periodic characteristic structure, and the formula (1) is obtained by carrying out Bayesian transformation on Pr (l | c; t); the formula (2) is obtained by representing the channel by using a television program set Prog (c, t) played by the channel c at time t and expanding the channel by using a probability addition formula; the formula (3) is obtained by performing Bayesian transformation on Pr (p | l) in the formula (2); the formula (4) is obtained by reducing the same items in the numerator and denominator in the formula (3) and adding a normalization factor Zc, which is specifically calculated as follows: zc ═ Σ L' ∈ L Σ p ∈ Prog (c, t) Pr (L | p) · Pr (p), where L is the topic label corpus;

further, the second step further comprises the following steps: selecting a target user, and collecting historical data of watching television watched by the target user within a certain time period; based on the watching history data, a periodic watching video feature model is constructed and obtained by using the probability distribution feature of the subject label of the target user, and the periodic watching video feature model is obtained by adopting a derivation mode similar to that of the periodic playing video feature model, and is characterized as follows:

prog (u, t) in formula (5) represents a video program list watched by the target user u within the time t, and Pr (l | p), Pr (p) and Zu adopt the same calculation mode as the feature model of the periodically played video;

carrying out balance processing on the characteristic model of the periodically watched video, wherein the smoothed characteristic model of the periodically watched video is as follows:

prog (u, t ') in equation (6) represents a list of video programs watched by the target user u in a time t' adjacent to the time t of the preset statistical period, and Prog (u) represents all the video programs watched by the target user u; neig (t) is a set of all neighborhood times of time t of a preset statistical period;

a recommending unit, configured to match video feature distribution data of a periodically playing video feature model for at least one channel with video feature distribution data of the periodically watching video feature model, and obtain recommended data for a target user based on a matching result, where a relative entropy between a topic tag probability distribution of a television channel and a topic tag probability distribution of the target user is used to measure a matching degree between the periodically playing video feature model and the periodically watching video feature model, and a specific calculation manner is as follows:

dist (c, u; t) of all candidate channels c for the target user can be obtained through calculation based on the formula (7), and finally, the first M channels with the maximum Dist (c, u; t) values are selected and recommended to the user.

7. A digital television channel-based recommender, comprising:

one or more processors;

a memory communicatively coupled to the one or more processors;

one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more programs configured to perform the method of any of claims 1-5.

8. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the method according to any one of claims 1 to 5.

Images