CN117176982A

CN117176982A - Video recommendation method, device, terminal equipment and storage medium

Info

Publication number: CN117176982A
Application number: CN202310943876.9A
Authority: CN
Inventors: 智丹丹; 李展鹏
Original assignee: China Mobile Communications Group Co Ltd; China Mobile Financial Technology Co Ltd
Current assignee: China Mobile Communications Group Co Ltd; China Mobile Financial Technology Co Ltd
Priority date: 2023-07-28
Filing date: 2023-07-28
Publication date: 2023-12-05

Abstract

The invention discloses a video recommendation method, a video recommendation device, terminal equipment and a storage medium, wherein the method comprises the following steps: classifying videos to be recommended to obtain a video set to be recommended; updating the user conversion rate of the video set to be recommended through a multi-arm slot machine MAB algorithm according to the service real-time flow data; and selecting the video set to be recommended with the highest user conversion rate to conduct video recommendation. The method solves the problem that the recommendation type cannot be adjusted in real time according to the user interest and the recommendation effect is not ideal due to lack of user feedback, and improves the accuracy of video recommendation.

Description

Video recommendation method, device, terminal equipment and storage medium

Technical Field

The present invention relates to the field of video content recommendation technologies, and in particular, to a video recommendation method, a video recommendation device, a terminal device, and a storage medium.

Background

Along with the explosive development of the internet and communication technology, the video service also has explosive growth, how to improve the accuracy and richness of video recommendation, and promote the use experience of users to become the core problem of concern of video recommendation, the video recommendation algorithm of the current mainstream can be divided into a recall layer, a coarse arrangement layer and a fine arrangement layer from functions, the video quantity and the video association degree of the recall layer determine the effect of the recommendation algorithm, but the recall result can only reflect the interests of users to a certain extent, real-time exploration cannot be performed on the interests of users, the lack of richness of the recommended content can lead the users to be in interest bottlenecks, interest fatigue is generated, and the interest change of the users cannot be effectively captured, so that optimization of the recommendation algorithm and improvement of the follow-up recommendation conversion rate are not facilitated.

Currently, the main stream ways of improving the video recommendation richness are divided into two main types: 1. rule-based diversity policy: the method mainly uses means such as a duplication removal strategy, a frequency control strategy, a scattering strategy, a sub-bucket scattering and the like, and performs video category control when the recall video after fine discharge is manually controlled and displayed to a user again in a manual rule mode so as to achieve the effect of improving the video richness; 2. based on an algorithm diversity strategy: the method comprises the steps of long tail weighting, a maximum boundary algorithm, a graph relation algorithm, a multi-arm slot machine algorithm and the like, and the types of videos in a list to be recommended are controlled through similarity, entropy and weight relation of the videos in the list to be recommended.

However, the method has the problems that the recommendation type cannot be adjusted in real time according to the user interest, and the recommendation effect is not ideal due to lack of user feedback.

The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art.

Disclosure of Invention

The invention mainly aims to provide a video recommendation method, a video recommendation device, a video recommendation terminal and a video recommendation storage medium, and aims to solve the technical problem that recommendation effects are not ideal due to the fact that recommendation types cannot be adjusted in real time according to user interests and lack of user feedback.

In order to achieve the above object, the present invention provides a video recommendation method, including:

classifying videos to be recommended to obtain a video set to be recommended;

updating the user conversion rate of the video set to be recommended through a multi-arm slot machine MAB algorithm according to the service real-time flow data;

and selecting the video set to be recommended with the highest user conversion rate to conduct video recommendation.

Optionally, the step of classifying the video to be recommended and acquiring the video set to be recommended further includes:

acquiring the heat information of the video to be recommended;

and storing the video set to be recommended according to the heat information to obtain a video heat library.

Optionally, before the step of updating the user conversion rate of the video set to be recommended by using the multi-arm slot machine MAB algorithm according to the service real-time traffic data, the method further includes:

calculating the user conversion rate of the video set to be recommended through preset initial parameters, and obtaining the initial user conversion rate;

and sending the initial user conversion rate to a user interest library for storage.

Optionally, the step of updating the user conversion rate of the video set to be recommended by using a multi-arm slot machine MAB algorithm according to the service real-time traffic data includes:

Acquiring real-time feedback data of a user through the service real-time flow data;

according to the real-time feedback data, carrying out parameter updating on the initial parameters to obtain final parameters;

and updating the initial user conversion rate through a multi-arm slot machine MAB algorithm according to the final parameters to obtain the final user conversion rate.

Optionally, the step of selecting the video set to be recommended with the highest user conversion rate for video recommendation includes:

sequencing the video sets to be recommended according to the conversion rate of the end user, and obtaining the video set to be recommended with the highest conversion rate of the user;

and recommending the video by using the video set to be recommended with the highest user conversion rate.

Optionally, the step of updating the initial parameter according to the real-time feedback data to obtain a final parameter includes:

inputting the real-time feedback data into the video hotness library for updating, and obtaining an updated video hotness library;

and carrying out parameter updating on the initial parameters according to the updated video heat library to obtain final parameters.

Optionally, the step of updating the initial user conversion rate by using a multi-arm slot machine MAB algorithm according to the final parameter and obtaining an end user conversion rate further includes:

Acquiring algorithm parameters in an initial multi-arm slot machine MAB algorithm;

weight distribution is carried out on preset user operation, and a distribution result is obtained;

according to the distribution result, improving the algorithm parameters to obtain final algorithm parameters;

and acquiring a final multi-arm slot machine MAB algorithm according to the final algorithm parameters.

The embodiment of the invention also provides a video recommending device, which comprises:

the classification module is used for classifying the video to be recommended and acquiring a video set to be recommended;

the updating module is used for updating the user conversion rate of the video set to be recommended through a multi-arm slot machine MAB algorithm according to the service real-time flow data;

and the recommending module is used for selecting the video set to be recommended with the highest user conversion rate to conduct video recommendation.

The embodiment of the invention also provides a terminal device which comprises a memory, a processor and a video recommendation program stored in the memory and capable of running on the processor, wherein the video recommendation program realizes the steps of the video recommendation method when being executed by the processor.

The embodiment of the invention also provides a computer readable storage medium, wherein the computer readable storage medium stores a video recommendation program, and the video recommendation program realizes the steps of the video recommendation method when being executed by a processor.

The embodiment of the invention provides a video recommendation method, a video recommendation device, terminal equipment and a storage medium, wherein videos to be recommended are classified, and a video set to be recommended is obtained; updating the user conversion rate of the video set to be recommended through a multi-arm slot machine MAB algorithm according to the service real-time flow data; and selecting the video set to be recommended with the highest user conversion rate to conduct video recommendation. The improved multi-arm slot machine MAB algorithm is used for calculating the real-time traffic data of the service, so that video recommendation is realized, the problem that recommendation effects are not ideal due to the fact that recommendation types cannot be adjusted in real time according to user interests and user feedback is lacked is solved, and the video recommendation accuracy is improved.

Drawings

FIG. 1 is a schematic diagram of functional blocks of a terminal device to which a video recommendation apparatus of the present invention belongs;

FIG. 2 is a flowchart illustrating an exemplary embodiment of a video recommendation method according to the present invention;

FIG. 3 is a flowchart illustrating a video recommendation method according to another exemplary embodiment of the present invention;

FIG. 4 is a flowchart illustrating a video recommendation method according to another exemplary embodiment of the present invention;

FIG. 5 is a flow chart of the video recommendation method of the present invention involving updating user conversion rate;

FIG. 6 is a schematic diagram of a video recommendation method according to the present invention involving obtaining a video set with a highest user conversion rate;

FIG. 7 is a schematic flow chart of a video recommendation method according to the present invention;

FIG. 8 is a flow chart of the video recommendation method according to the present invention, which involves obtaining final parameters;

FIG. 9 is a flow chart of the video recommendation method of the present invention relating to improving the MAB algorithm of the multi-arm slot machine.

The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

The main solutions of the embodiments of the present invention are: acquiring the heat information of the video to be recommended; and storing the video set to be recommended according to the heat information to obtain a video heat library. Calculating the user conversion rate of the video set to be recommended through preset initial parameters, and obtaining the initial user conversion rate; and sending the initial user conversion rate to a user interest library for storage. Acquiring real-time feedback data of a user through the service real-time flow data; according to the real-time feedback data, carrying out parameter updating on the initial parameters to obtain final parameters; and updating the initial user conversion rate through a multi-arm slot machine MAB algorithm according to the final parameters to obtain the final user conversion rate. Sequencing the video sets to be recommended according to the conversion rate of the end user, and obtaining the video set to be recommended with the highest conversion rate of the user; and recommending the video by using the video set to be recommended with the highest user conversion rate. Inputting the real-time feedback data into the video hotness library for updating, and obtaining an updated video hotness library; and carrying out parameter updating on the initial parameters according to the updated video heat library to obtain final parameters. Acquiring algorithm parameters in an initial multi-arm slot machine MAB algorithm; weight distribution is carried out on preset user operation, and a distribution result is obtained; according to the distribution result, improving the algorithm parameters to obtain final algorithm parameters; and acquiring a final multi-arm slot machine MAB algorithm according to the final algorithm parameters. Therefore, the problem that the recommendation type cannot be adjusted in real time according to the user interest and the recommendation effect is not ideal due to lack of user feedback is solved, video recommendation is realized, and the video recommendation accuracy is improved. Based on the scheme of the invention, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interest and the lack of user feedback exists in the video recommendation in reality is solved, the effectiveness of the video recommendation method is verified when the video is recommended, and finally the accuracy of video recommendation is obviously improved through the method.

Technical terms related to the embodiment of the invention:

MAB: is a short for Multi-arm gambling machine (Multi-arm bandwidth) which is a common algorithmic framework in reinforcement learning and optimization problems in which there are multiple tie rods (arms), each with a different potential prize distribution, with the goal of finding the best tie rod within a limited number of attempts to maximize the cumulative prize, and the MAB algorithm aims at a strategy of balancing the Exploration (expression) against the Exploitation (expression) to balance the Exploitation of known good tie rods with the Exploration of unknown tie rods, some common MAB algorithms: epsilon-greedy algorithm: the ratio of exploration and utilization is controlled according to a parameter epsilon. Selecting the pull rod with the probability of 1-epsilon and the best estimation currently (utilization), and randomly selecting other pull rods with the probability of epsilon (exploration); UCB (Upper Confidence Bound) algorithm: the selection is made by using the upper bound of confidence intervals for each tie rod. This upper bound takes into account the rewards that have been earned and the number of times the tie rod is explored in order to trade off exploration and utilization; thompson sampling algorithm: the potential rewards for each tie are modeled as a probability distribution, and the ties are selected by sampling and updating. After each selection of a pull rod, sampling a value from the respective probability distribution, wherein the probability of the pull rod with higher rewards being selected is higher; gradient gambling algorithm: a probability distribution is established based on the average prize for each tie rod and the probability distribution is adjusted based on the prizes. The tie rods with higher average rewards will be selected with higher probability while taking into account the factors of exploration and utilization, which will balance between exploration and utilization to solve the MAB problem, which algorithm is specifically selected depending on the specific application context, goals and constraints, and other MAB algorithms and improved methods, which may require different algorithms to achieve optimal performance.

Cold Start (Cold Start): refers to problems encountered during initial stages of the system or in the face of new users, products, content, etc., due to the lack of sufficient a priori information. In the case of cold starts, conventional data driven algorithms may not work effectively because they require learning and recommendation based on historical data, and cold start problems may have different manifestations and solutions in different application areas, as follows are some common cold start problems and corresponding solutions: user cold start: when the system is faced with new users, there is a lack of historical data for their personalized preferences. The solution includes recommendation based on user attributes (such as age, gender, geographic location, etc.), utilization of collaborative filtering algorithm and mixed recommendation algorithm, etc.; cold starting of the product: when a new product or commodity is introduced into the system, related user feedback and evaluation information are lacked, and the solution method comprises recommending (such as category, label, keyword and the like) based on the product attribute, utilizing a content filtering algorithm, a collaborative filtering algorithm, a mixed recommending algorithm and the like; content cold start: when the system is faced with new content (e.g., news, articles, videos, etc.), it lacks its associated historical interaction data. The solution includes content-based recommendations (such as text features, keywords, topics, etc.), utilization of collaborative filtering algorithms, hybrid recommendation algorithms, tag-based recommendations, etc.; context cold start: when the system needs to provide personalized recommendations based on the user's current context (e.g., time, geographic location, device, etc.), but lacks historical data for context information. The method for solving the cold start problem generally needs to combine domain knowledge, user feedback and some heuristic strategies to acquire preliminary data and signals, gradually accumulate and optimize a recommendation model, and can also evaluate and improve the effect of the cold start strategy by adopting methods such as experimental design, AB test and the like.

EE: exploration and utilization (Exploration and Exploitation) refers to strategies that balance exploration and utilization in the decision process, which concept is often applied in the field of machine learning, optimization and decision science, and in EE exploration refers to exploration of unknown fields and possibilities by trying new choices and strategies. This helps to gain new knowledge, discover new solutions and opportunities, but may require some risk or cost, while leveraging is based on known information and experience to maximize current benefits. By exploiting existing knowledge and experience, decisions and goals can be made more efficiently, but some unknown opportunities may be missed, in many cases exploration and exploitation are competing requirements, excessive exploration may result in failure to fully exploit current resources and knowledge, and excessive exploitation may result in sinking into locally optimal solutions and missing better solutions, so in practice, a balance point needs to be found between exploration and exploitation for best results, which can be achieved by employing different strategies and algorithms, such as epsilon-greedy algorithms, multi-arm gambling algorithms, etc. These approaches have been explored to some extent to find new possibilities, while also taking advantage of existing knowledge to maximize benefit.

Fine discharge layer: in an information retrieval system, a stage or module for fine ordering and ranking of search results is usually based on basic retrieval technologies such as inverted index, and the like, search results are further ordered and screened through a series of algorithms and models to provide search results with more relevant and high quality for users, and a fine ranking layer aims to rank the most relevant results in front according to query intention and preference of users, and also considers a plurality of factors such as quality, credibility, timeliness of content and the like, and in order to achieve the aim, various machine learning, natural language processing and information retrieval technologies are generally used for the fine ranking layer.

Rearrangement layer: is a stage or module in an information retrieval system for re-ordering existing search results, and its main purpose is to make the search results more fit to the needs and intentions of the user by taking more factors into account and adopting a more complex ordering algorithm, and the rearrangement layer is usually performed after the fine-ranking layer, and possibly combined with the fine-ranking layer into a single module, which can adjust the order of the search results according to the feedback of the user, the history data and other higher-level signals, so as to provide better search experience.

According to the embodiment of the invention, when video recommendation is carried out, the related technology cannot realize real-time adjustment of recommendation types according to the user interests, and the video recommendation richness is improved through similarity, entropy and weight relation of videos in a user recommendation list, so that a larger weight library is always required to be maintained or real-time computing resources of real-time recommendation are increased, the computing cost is increased, and meanwhile, the recommendation effect is always not ideal due to lack of feedback of users.

Therefore, based on the scheme of the invention, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interest and the lack of user feedback exists in the actual video recommendation process, the video recommendation method is designed, the effectiveness of the video recommendation method is verified when the video is recommended, and finally the accuracy of video recommendation through the method is obviously improved.

Specifically, referring to fig. 1, fig. 1 is a schematic diagram of functional blocks of a terminal device to which the video recommendation apparatus of the present invention belongs. The video recommendation device may be independent of the device of the terminal device capable of video recommendation, and may be carried on the terminal device in the form of hardware or software. The terminal equipment can be intelligent mobile equipment with a data processing function such as a mobile phone and a tablet personal computer, and can also be fixed terminal equipment or a server with a data processing function.

In this embodiment, the terminal device to which the video recommendation apparatus belongs at least includes an output module 110, a processor 120, a memory 130, and a communication module 140.

The memory 130 stores an operating system and a video recommendation program, and the video recommendation device can classify videos to be recommended to obtain a video set to be recommended; updating the user conversion rate of the video set to be recommended through a multi-arm slot machine MAB algorithm according to the service real-time flow data; and selecting the video set to be recommended with the highest user conversion rate to conduct video recommendation. Video recommendation is performed by the video recommendation program, and information such as results of recommendation, etc. is obtained and stored in the memory 130; the output module 110 may be a display screen or the like. The communication module 140 may include a WIFI module, a mobile communication module, a bluetooth module, and the like, and communicates with an external device or a server through the communication module 140.

Wherein the video recommendation program in the memory 130 when executed by the processor performs the steps of:

classifying videos to be recommended to obtain a video set to be recommended;

Further, the video recommendation program in the memory 130, when executed by the processor, further performs the steps of:

acquiring the heat information of the video to be recommended;

According to the scheme, the video to be recommended is classified, and the video set to be recommended is obtained; updating the user conversion rate of the video set to be recommended through a multi-arm slot machine MAB algorithm according to the service real-time flow data; and selecting the video set to be recommended with the highest user conversion rate to conduct video recommendation. The method comprises the steps of carrying out improvement based on a preset initial multi-arm slot machine MAB algorithm, obtaining an improved multi-arm slot machine MAB algorithm, and carrying out video recommendation by using the improved multi-arm slot machine MAB algorithm, so that the problem that recommendation effects are not ideal due to the fact that recommendation types cannot be adjusted in real time according to user interests and lack of user feedback can be solved. Based on the scheme of the invention, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interest and the lack of user feedback exists in the video recommendation in reality is solved, the effectiveness of the video recommendation method is verified when the video is recommended, and finally the accuracy of video recommendation is obviously improved through the method.

The method embodiments of the present invention are presented based on the above-described terminal device architecture but not limited to the above-described framework.

Referring to fig. 2, fig. 2 is a flowchart illustrating an exemplary embodiment of a video recommendation method according to the present invention. The video recommendation method comprises the following steps:

step S01: classifying videos to be recommended to obtain a video set to be recommended;

the main implementation body of the method of the embodiment may be a video recommendation device, or may be a video recommendation terminal device or a server, and the embodiment uses the video recommendation device as an example, where the video recommendation device may be integrated on a terminal device with a data processing function.

In order to acquire a video set to be recommended, the following steps are adopted:

first, all videos in a database are acquired, wherein all videos refer to videos previously stored in the database, including but not limited to contribution videos, official videos and the like;

then, classifying according to all the acquired videos, wherein the classified types include, but are not limited to, sports, entertainment, games, civilian, and the like;

finally, generating a video set according to the classified videos;

further, in this embodiment, a recommendation arm is defined for each type of video, where a gambling Machine (MAB) algorithm is used, and each type of video serves as a horn.

Step S06, updating the user conversion rate of the video set to be recommended through a multi-arm slot machine MAB algorithm according to the service real-time flow data;

after the video set is acquired, in order to update the user conversion rate of the video set to be recommended, the following steps are adopted to realize:

firstly, acquiring business real-time flow data, wherein the business real-time flow data is formed for each user, and comprises a series of operations of praying, forwarding, collecting and the like on the current video, and data of watching time, watching times and the like of the current video;

and finally, updating the user conversion rate of each video set through a multi-arm slot machine MAB algorithm according to the service real-time flow data, wherein the updating process adopts the steps of converting the service real-time flow data to form feedback data, updating parameters of an algorithm by using the feedback data, carrying out algorithm operation on the parameters after updating the parameters, and then obtaining the user conversion rate of each video set.

And S07, selecting a video set to be recommended with the highest user conversion rate to conduct video recommendation.

After the user conversion rate of the video set is obtained, video recommendation is completed through the following steps:

Firstly, sequencing the user conversion rate of each video set obtained previously;

finally, the video set with the highest user conversion rate is selected as the recommended video, wherein in this embodiment, a single user is used for illustration, so it can be understood that the user conversion rate of each type of video is updated according to the operation of each user and the acquisition of the real-time traffic data, and then the video category with the highest behavior frequency of frequent viewing, sharing, praying and the like of the current user can be known, so that the video is recommended.

Further, in the embodiment, for example, for a single user, in actual application, the interest of each user can be known and recommended by calculating according to the requirements of actual services, for example, for different users, and the user who first uses the video player recommends the video player according to the user conversion rate of all video sets and the user facing of the current video player, so that the video recommendation effect is excellent when the video player is started in a cold state.

According to the scheme, the video to be recommended is classified, and the video set to be recommended is obtained; updating the user conversion rate of the video set to be recommended through the service real-time flow data; and selecting the video set to be recommended with the highest user conversion rate to conduct video recommendation. The video recommendation method and device can be used for recommending videos, the problem that recommendation types cannot be adjusted in real time according to user interests and recommendation effects are not ideal due to lack of user feedback is solved, and accuracy of video recommendation is improved.

Referring to fig. 3, fig. 3 is a flowchart illustrating another exemplary embodiment of a video recommendation method according to the present invention.

Based on the embodiment shown in fig. 2, the step S01 of classifying the video to be recommended, and the step of obtaining the video set to be recommended further includes:

step S02, acquiring the heat information of the video to be recommended;

and step S03, storing the video set to be recommended according to the heat information, and obtaining a video heat library.

Specifically, in order to obtain the video hotness library, the method is realized by the following steps:

firstly, a video popularity database is a visual representation for representing video popularity in a current database, wherein the establishment of the popularity database is to record according to original video popularity information to obtain the popularity information of the current video, the popularity information of the video is an index for measuring popularity and popularity of the video, and the following are some common video popularity information: the play amount refers to the number of times the video is watched, and a higher play amount generally indicates that the video is widely concerned and browsed; the praise number refers to the number of likes or approvals received by the video, and the praise number indicates that the video is approved and loved by audiences; the comment number refers to the number of comments received by the video, and a higher comment number may represent that the video causes discussion and interaction of audience; the sharing number refers to the number of times that the video is shared to other platforms or social media, and the sharing number is more that the audience considers the video to be worth sharing with others; the barrage interaction refers to real-time comments sent by audiences in the video playing process, the barrage interaction refers to interaction in the form of barrage sent by the audiences when the videos are played, and more barrage interactions can reflect the attractive force and the interactivity of the videos; subscription number, which refers to the number of subscribers to a channel or an uploader, an increase in subscription number represents a continued interest of the viewer in the content of the channel or uploader.

And finally, storing the heat information of each video in a video set mode to obtain a video heat library, wherein the video heat library is updated according to the real-time service flow.

According to the embodiment, the method comprises the steps of obtaining the heat information of the video to be recommended; and storing the video set to be recommended according to the heat information to obtain a video heat library. Therefore, the construction of the video hotness library is realized, the problem of unsatisfactory recommendation effect caused by lack of user feedback is solved, and the accuracy of video recommendation is improved.

Referring to fig. 4, fig. 4 is a flowchart illustrating another exemplary embodiment of a video recommendation method according to the present invention.

Based on the embodiment shown in fig. 2, the step S06, according to the service real-time traffic data, further includes, before the step of updating the user conversion rate of the video set to be recommended by using the multi-arm slot machine MAB algorithm:

step S04, calculating the user conversion rate of the video set to be recommended through preset initial parameters, and obtaining the initial user conversion rate;

and step S05, the initial user conversion rate is sent to a user interest library for storage.

Specifically, in order to obtain the user interest library, the following steps are adopted to realize:

Firstly, calculating the user conversion rate of a video set to be recommended by using initial parameters to obtain the initial user conversion rate, wherein the initial parameters refer to Beta distribution in a MAB algorithm, which is often used for Thompson Sampling strategies, the Beta distribution is a probability distribution and is usually used for modeling the condition that the value range of a random variable is between [0,1], the Beta distribution has two parameters, which are usually expressed as Alpha and Beta, wherein Alpha controls the shape parameters of a distribution shape, beta controls the scale parameters of the distribution shape, in the context of the MAB algorithm, the Thompson Sampling strategy uses the Beta distribution to model and estimate the rewarding probability of each gambling machine (arm), and Alpha and Beta can be initially set to a smaller positive number, and then Alpha and Beta parameters are updated according to obtained rewarding information with each attempt/exploration and utilization update;

finally, a user interest library is established based on the initial user conversions obtained, wherein the user interest library refers to a data structure or database storing interest information of users, which is used to record the degree or probability of each user's interest in a different gambling machine (or arm) in order to make a more intelligent decision when selecting an arm.

Further, the user interest library is established and updated along with the real-time flow input of the service, which means that after each time of operation, the user conversion rate of the current video set is obtained, and when the user is cold started, video recommendation is performed according to the user conversion rates of all users, so that good recommendation effect is achieved during cold start.

According to the scheme, specifically, the user conversion rate of the video set to be recommended is calculated through preset initial parameters, and the initial user conversion rate is obtained; and sending the initial user conversion rate to a user interest library for storage. Therefore, the construction of the user interest library is completed, the problem that no user feedback data is updated when video recommendation is performed is solved, the conversion of the user real-time feedback data is realized, and the accuracy of video recommendation is improved.

Referring to fig. 5, fig. 5 is a schematic flow chart of a video recommendation method according to the present invention, which relates to updating user conversion rate.

Based on the embodiment shown in fig. 2, the step S06 of updating the user conversion rate of the video set to be recommended by using the multi-arm slot machine MAB algorithm according to the service real-time traffic data includes:

Step S061, acquiring real-time feedback data of a user through the service real-time flow data;

step S062, carrying out parameter updating on the initial parameters according to the real-time feedback data to obtain final parameters;

and step S067, updating the initial user conversion rate through a multi-arm slot machine MAB algorithm according to the final parameters to obtain the final user conversion rate.

Specifically, to obtain end-user conversion, this is achieved by:

firstly, processing and analyzing the acquired business real-time flow data, wherein the business real-time flow data comprises but is not limited to data such as access quantity, clicking times and sharing quantity of users, the processing and analyzing operations comprise but are not limited to operations such as data cleaning, abnormal value removal and index calculation, and some key indexes or trends such as change of access quantity of users, hot pages and the like can be obtained by analyzing the data;

then, according to the service requirement, determining a required feedback mechanism and a target, for example, setting thresholds of some key indexes, and triggering corresponding feedback measures when the thresholds are reached or exceeded;

then, generating corresponding real-time feedback data according to the real-time flow data and a set feedback mechanism;

Then, updating initial parameters through real-time feedback data to obtain final parameters, wherein the initial parameters are preset values, and reflect the user conversion rate of the current video set;

and finally, calculating final parameters by using a multi-arm slot machine MAB algorithm to obtain the user conversion rate of the current video set.

More specifically, as shown in fig. 6, fig. 6 is a schematic diagram of a video recommendation method according to the present invention, which involves acquiring a video set with the highest user conversion rate.

Firstly, a recall layer selects a candidate set of potentially interested items through quick screening so as to reduce the calculated amount and improve the recommendation efficiency, and the recall layer is used for quickly selecting the items possibly meeting the user interests from a large-scale item library, so that a plurality of types of video sets (namely video types) are obtained;

then, category labels are carried out on the video set through the coarse arrangement layer;

then, through the category of the video and the respective heat information, the most relevant and high-quality articles are ranked in front through a fine ranking layer according to the interests and the preferences of the user, and better recommendation experience is provided;

then, updating through a multi-arm slot machine MAB algorithm according to a user interest library (namely the specific behavior parameters of the current user) to obtain a final recommended video set, namely K-type videos in the embodiment;

And finally, according to feedback data of the user on the K-type video, updating parameters.

According to the scheme, the real-time feedback data of the user is obtained specifically through the service real-time flow data; according to the real-time feedback data, carrying out parameter updating on the initial parameters to obtain final parameters; and updating the initial user conversion rate through a multi-arm slot machine MAB algorithm according to the final parameters to obtain the final user conversion rate. Therefore, the conversion rate of the end user is obtained, the video recommendation is realized, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interest and the lack of user feedback is solved, and the video recommendation accuracy is improved.

Referring to fig. 7, fig. 7 is a schematic flow chart of a video recommendation method according to the present invention.

Based on the embodiment shown in fig. 2, the step S07 of selecting the video set to be recommended with the highest user conversion rate for video recommendation includes:

step 071, sorting the video sets to be recommended according to the end user conversion rate, and obtaining the video set to be recommended with the highest user conversion rate;

And step 072, performing video recommendation by using the video set to be recommended with the highest user conversion rate.

Specifically, in order to select the video set with the highest user conversion rate for recommendation, the following steps are adopted:

firstly, after the user conversion rate of each video set is obtained, sorting the video sets, wherein the user conversion rate refers to the proportion of the user from accessing, browsing, clicking and the like to actually completing the behavior or the target under a certain specific behavior or target, and the user conversion rate is generally used for measuring the effect and the benefit in the fields of marketing, advertising, recommendation systems and the like;

then, according to the sequencing result, obtaining the video set with the highest user conversion rate in all the current video sets;

and finally, recommending the user by using the video set with the highest user conversion rate.

According to the scheme, the video set to be recommended with the highest user conversion rate is obtained by sorting the updated user conversion rates; and recommending the video by using the video set to be recommended with the highest user conversion rate. The video recommendation of the user is completed, the video recommendation is realized, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interest and the lack of user feedback is solved, and the accuracy of the video recommendation is improved.

Referring to fig. 8, fig. 8 is a flowchart illustrating a video recommendation method according to the present invention, which involves obtaining final parameters.

Based on the embodiment shown in fig. 5, in step S062, the step of updating the initial parameters according to the real-time feedback data to obtain final parameters includes:

step S0621, inputting the real-time feedback data into the video hotness library for updating, and obtaining an updated video hotness library;

and step S0622, carrying out parameter updating on the initial parameters according to the updated video heat database to obtain final parameters.

Specifically, to obtain the final parameters, the following steps are taken:

firstly, inputting the acquired real-time feedback data into a video hotness library for updating, wherein the user represented by the video hotness library performs various specific behaviors of the current video, such as praise, forwarding, appreciation and the like, and the actions are added according to different weights, namely the specific hotness data of the current video;

and finally, carrying out parameter updating on the initial parameters according to the obtained specific heat data to obtain final parameters, wherein the final parameters are combined with specific operations of the user, the final parameters are stored and recorded in a user interest library, and the final calculated user conversion rate is also stored in the user interest library.

According to the scheme, the real-time feedback data are input into the video hotness library for updating, and the updated video hotness library is obtained; and carrying out parameter updating on the initial parameters according to the updated video heat library to obtain final parameters. Therefore, the acquisition of final parameters is completed, the update of the user conversion rate according to the real-time feedback data is realized, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interest and the lack of user feedback is solved, and the accuracy of video recommendation is improved.

Referring to fig. 9, fig. 9 is a flowchart illustrating a video recommendation method according to the present invention, which involves improving the MAB algorithm.

Based on the embodiment shown in fig. 5, step S067, according to the final parameter, updates the initial user conversion rate by using the multi-arm slot machine MAB algorithm, and before the step of obtaining the final user conversion rate, further includes:

step S063, obtaining algorithm parameters in an initial multi-arm slot machine MAB algorithm;

step S064, carrying out weight distribution on preset user operation to obtain a distribution result;

step S065, improving the algorithm parameters according to the distribution result to obtain final algorithm parameters;

And step S066, obtaining a final multi-arm slot machine MAB algorithm according to the final algorithm parameters.

Specifically, in order to obtain the final multi-arm slot machine MAB algorithm, the method is realized by the following steps:

firstly, algorithm parameters in an initial MAB algorithm are obtained, wherein the traditional MAB algorithm assumes that the click probability of a user on each type of video obeys a Beta distribution (alpha, beta), and the probability density function is as follows:

alpha and Beta respectively represent two parameters of the algorithm, alpha represents the times of converting the video of the category, beta represents the times of not converting the video of the category, beta distribution parameters of the item are updated according to whether the item is converted by a user, specifically, alpha is increased by 1 if the item is converted, and otherwise Beta is increased by 1.

Then, because the initial multi-arm slot machine MAB algorithm has a certain inadequacy, the multi-arm slot machine MAB algorithm is improved, mainly the calculated parameters are improved, the initial multi-arm slot machine MAB algorithm does not consider the information of the video itself and the hidden information behind different behaviors of the user, the improved multi-arm slot machine MAB algorithm mainly improves the updating mechanism of alpha and beta parameters, the actual behaviors (clicking, praying, collecting, sharing, stepping on the points and the like) of the user and the heat information of the video itself are further captured by the user, and the weight occupied by the clicking, praying, collecting and sharing behaviors of the user is set as omega ₁ 、ω ₂ 、ω ₃ 、ω ₄ The hotness information of the video is p, and the action weight of the user for clicking and stepping on the video is omega ₅ 、ω ₆ The threshold value of the heat video is set as W, the value can be adjusted automatically according to the heat information of the platform video, and in the embodiment, the threshold value is set as 75 quantiles of all the heat values of the video, and the threshold value can be adjusted dynamically according to the heat condition of the video. The updating mechanism of the alpha and beta parameters of the multi-arm slot machine is as follows:

according to the scheme, the method comprises the steps of obtaining algorithm parameters in an initial multi-arm slot machine MAB algorithm; weight distribution is carried out on preset user operation, and a distribution result is obtained; according to the distribution result, improving the algorithm parameters to obtain final algorithm parameters; and acquiring a final multi-arm slot machine MAB algorithm according to the final algorithm parameters. Therefore, the improvement of the MAB algorithm is completed, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interest and the lack of user feedback is solved, and the accuracy of video recommendation is improved.

In addition, an embodiment of the present invention further provides a video recommendation device, where the video recommendation device includes:

In addition, the embodiment of the invention also provides a terminal device, which comprises a memory, a processor and a video recommendation program stored in the memory and capable of running on the processor, wherein the video recommendation program realizes the steps of the video recommendation method when being executed by the processor.

Because the video recommendation program is executed by the processor and adopts all the technical schemes of all the embodiments, the video recommendation program at least has all the beneficial effects brought by all the technical schemes of all the embodiments and is not described in detail herein.

In addition, the embodiment of the invention also provides a computer readable storage medium, wherein the computer readable storage medium stores a video recommendation program, and the video recommendation program realizes the steps of the video recommendation method when being executed by a processor.

Compared with the prior art, the video recommendation method, the video recommendation device, the terminal equipment and the storage medium provided by the embodiment of the invention are used for classifying the video to be recommended to obtain a video set to be recommended; updating the user conversion rate of the video set to be recommended through a multi-arm slot machine MAB algorithm according to the service real-time flow data; and selecting the video set to be recommended with the highest user conversion rate to conduct video recommendation. Therefore, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interests and the lack of user feedback is solved, accurate video recommendation is achieved for the user, and accuracy of video recommendation is improved. Based on the scheme of the invention, the problem that the recommendation effect is not ideal due to the fact that the recommendation type cannot be adjusted in real time according to the user interest and the lack of user feedback exists in the video recommendation in reality is solved, the effectiveness of the video recommendation method is verified when the video is recommended, and finally the accuracy of video recommendation is obviously improved through the method.

Compared with the prior art, the embodiment of the invention has the following advantages:

1. the multi-arm slot machine algorithm of the reinforcement learning idea is introduced, so that the advantages of video recommendation diversity and video recommendation accuracy are effectively combined, and particularly, the multi-arm slot machine algorithm has a good recommendation effect in a cold start scene;

2. The method and the device have the advantages that the algorithm parameters are updated by introducing the video heat information and the specific behaviors of the user, the problem that the multi-arm slot machine algorithm is insufficient in capturing the user interests is solved greatly, the exploration efficiency of the EE problem of video recommendation is improved, the user behaviors and the information of the video are fully utilized on the basis of the original algorithm, the excellent effect can be achieved in the aspect of tracking the user interests, and the user experience in a cold start scene is improved greatly.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.

The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as above, comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, a controlled terminal, or a network device, etc.) to perform the method of each embodiment of the present invention.

The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.

Claims

1. A video recommendation method, characterized in that the video recommendation method comprises the steps of:

Classifying videos to be recommended to obtain a video set to be recommended;

2. The video recommendation method according to claim 1, wherein the step of classifying the video to be recommended and acquiring the video set to be recommended further comprises:

acquiring the heat information of the video to be recommended;

3. The video recommendation method according to claim 2, wherein the step of updating the user conversion rate of the video set to be recommended by the multi-arm slot machine MAB algorithm according to the traffic real-time traffic data further comprises:

4. The video recommendation method according to claim 3, wherein the step of updating the user conversion rate of the video set to be recommended by a multi-arm slot machine MAB algorithm according to the traffic real-time traffic data comprises:

5. The method for recommending video according to claim 4, wherein the step of selecting the video set to be recommended with the highest user conversion rate for video recommendation comprises:

6. The video recommendation method according to claim 4, wherein the step of performing parameter update on the initial parameters according to the real-time feedback data to obtain final parameters comprises:

7. The video recommendation method according to claim 4, wherein the step of updating the initial user conversion rate by a multi-arm slot machine MAB algorithm according to the final parameter, and obtaining an end user conversion rate further comprises, before:

8. A video recommendation device, characterized in that the video recommendation device comprises:

9. A terminal device comprising a memory, a processor and a video recommendation program stored on the memory and executable on the processor, which video recommendation program, when executed by the processor, implements the steps of the video recommendation method according to any of claims 1-7.

10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon a video recommendation program, which when executed by a processor, implements the steps of the video recommendation method according to any of claims 1-7.