An automatic traffic diversion system for short video live broadcast based on real-time content recommendation algorithm

Abstract

The present invention discloses a short video live broadcast automatic diversion system based on a real-time content recommendation algorithm, including a user data collection module, a real-time content analysis module, a content matching module, a personalized recommendation module and a diversion strategy module. The user data collection module collects user data, the real-time content analysis module performs real-time analysis on user data, extracts user interests and keywords, and the content matching module selects video live broadcast content that meets user interests from the content library of the short video live broadcast platform according to the user interests and keywords; the personalized recommendation module recommends the most suitable short video live broadcast content to the user, and the diversion strategy module is used to formulate a diversion strategy. Beneficial effects of the present invention: Through the system provided by the present invention, users can find content of interest more conveniently, thereby improving user satisfaction and the platform's traffic conversion effect.

Description

Technical Field

[0001] The present invention relates to the fields of e-commerce platforms and machine learning, and in particular to an automatic traffic diversion system for short video live broadcasts based on a real-time content recommendation algorithm. Background Art

[0002] With the popularization of Internet technology and mobile devices, short video live streaming platforms have rapidly emerged and become an important way for people to obtain entertainment, information and interaction. Short video live streaming platforms have attracted a large number of users by providing rich and diverse content and real-time interactive functions. However, due to the huge and diverse content on short video live streaming platforms, users face information overload and difficulty in content selection when browsing and watching. In order to improve user experience and participation, short video live streaming platforms need to provide users with personalized and accurate content recommendations and diversion services.

[0003] At present, the content recommendation and traffic diversion of short video live streaming platforms mainly rely on traditional recommendation algorithms, such as content-based recommendation and collaborative filtering recommendation. Although these algorithms can provide a certain degree of recommendation effect, they still have certain limitations for real-time and personalization needs. Traditional recommendation algorithms are often based on static user portraits and offline analysis data, which cannot accurately reflect users' real-time interests and preferences. In addition, traditional algorithms do not handle the data characteristics and interaction patterns unique to short video live streaming platforms well, resulting in a mismatch between the recommended content and users' actual needs and interests, reducing user participation and satisfaction.

[0004] In order to solve these problems, the present invention proposes a short video live broadcast automatic diversion system based on a real-time content recommendation algorithm. The system utilizes real-time collected user data and an efficient algorithm module to provide users with personalized and accurate content recommendation and diversion services. By real-time analysis of users' interests, keywords, emotional states and preferences, the system can accurately grasp users' real-time needs and recommend the most suitable short video live broadcast content for them. The system also provides content related to users' interests through content matching and personalized recommendation modules, and effectively guides users to watch and participate through a diversion strategy module.

[0005] The research background of the present invention is based on an in-depth analysis and research on the development status and user needs of short video live streaming platforms. Through the understanding of the shortcomings of existing recommendation algorithms and user needs, it is realized that real-time and personalization are the key elements to improve the user experience of short video live streaming platforms. Therefore, the present invention aims to utilize real-time content recommendation algorithms, combined with user data collection, real-time content analysis, content matching, personalized recommendation and diversion strategy modules, to build an innovative short video live streaming automatic diversion system to improve users' click-through rate, viewing time and interactive behavior, and enhance the user stickiness and competitiveness of short video live streaming platforms. Summary of the invention

[0006] In view of the above problems, the purpose of the present invention is to provide a short video live broadcast automatic diversion system based on a real-time content recommendation algorithm to solve the problems raised in the above background technology.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] The invention provides a short video live broadcast automatic diversion system based on real-time content recommendation algorithm, including a user data collection module, a real-time content analysis module, a content matching module, a personalized recommendation module and a diversion strategy module, wherein the user data collection module is responsible for collecting the user's viewing history, interests and hobbies, and behavior pattern data, and updating them in real time. By collecting and analyzing user data, the system understands the user's interests and preferences. The real-time content analysis module analyzes the user data in real time, extracts the user's interests and keywords, and identifies the user's emotional state and preferences, so as to quickly and accurately understand the user's needs. The content matching module selects video live broadcast content that meets the user's interests from the content library of the short video live broadcast platform according to the user's interests and keywords, and the system finds the short video live broadcast content related to the user's interests. The personalized recommendation module recommends the most suitable short video live broadcast content to the user based on the user's interests and preferences. According to the user's historical viewing records and behavior patterns, the system provides personalized recommendation results. The diversion strategy module formulates a diversion strategy according to the user's behavior pattern and the overall strategy of the platform, and diverts the recommended short video live broadcast content to the user. The system flexibly adjusts the diversion strategy according to the user's preferences and the platform's goals to improve user participation and platform traffic conversion.

[0009] Furthermore, the implementation methods of the user data collection module include data acquisition, data storage, data update, data cleaning and processing. Relevant user data are collected through the user's operations and behaviors on the short video live broadcast platform. These data include the user's viewing history, likes and comment information, sharing behavior, followed anchors or topics that can be obtained by this system. Data is acquired through user login accounts, browsing records, and operation logs. The collected user data needs to be stored for subsequent real-time content analysis and personalized recommendations. Database technology is used to store user data in a structured manner, and ensure data security and privacy protection. By monitoring the user's operations and behaviors in this system, user data is updated in a timely manner to reflect the user's latest interests and preferences. Timed updates and real-time updates are used to update and synchronize data according to user activities. If the collected user data contains noise and redundant information, it needs to be cleaned and processed. Data preprocessing technology, including removing duplicate data and processing outliers, is used to ensure the accuracy and availability of user data.

[0010] Furthermore, the real-time content analysis module extracts users' interests and keywords and identifies users' emotional states and preferences by performing real-time analysis on user data. By analyzing users' viewing history, likes and comments, the real-time content analysis module identifies users' interests and extracts keywords from users' viewing history and interactive behaviors. These keywords reflect users' focus and areas of interest, which are helpful for subsequent content matching and personalized recommendations. In order to better understand users' emotional states and preferences, the real-time content analysis module performs sentiment analysis to analyze the emotions expressed by users in comments and likes, such as likes, dislikes, and excitement, in order to better understand users' preferences and feedback. At the same time, the real-time content analysis module needs to update user data in real time. As users' behaviors and preferences change, the module should be able to update the analysis results in a timely manner to provide accurate real-time recommendations and traffic diversion services.

[0011] Furthermore, the content matching module uses the real-time content analysis module to systematically understand the user's interests, keywords, emotional state and preferences, providing a basis for subsequent content matching and personalized recommendations. By analyzing user data in real time, it provides users with short video live content related to their interests, thereby improving user experience and platform traffic conversion effects. The content matching module first needs to build a content library containing various types of short video live content. The content library includes short video live broadcasts of different themes and types, including entertainment, sports, food, and music. The construction of the content library is carried out through manual screening, automatic crawling, and content creator uploading. The content matching module analyzes the user's keywords and interests, and performs keyword matching with the video live content in the content library. By matching user interests and keywords, the module filters out video live content related to user interests.

[0012] Furthermore, in order to more accurately match the interests of users, the content matching module uses the labels and classification information of the live video content to achieve more accurate matching by labeling and classifying the live video content, ensuring that the content recommended to users is consistent with their interests, and conducting quality assessments on the screened live video content. The assessment indicators include the number of views, the number of likes, and the number of comments. By evaluating the quality of the content, the module prioritizes recommending high-quality live video content to users, thereby improving the user experience and the attractiveness of the content.

[0013] Furthermore, in the content matching module, the user preference matrix is ​​defined as Among them, S 11 is the recommendation value of the first user for the first short video real-time content, S 12 is the recommendation value of the first user for the real-time content of the second short video, S 1N is the recommendation value of the first user for the real-time content of the Nth short video, S 21is the recommendation value of the second user for the real-time content of the first short video, S 22 is the recommendation value of the second user for the real-time content of the second short video, S 2N is the recommendation value of the second user for the real-time content of the Nth short video, S M1 is the recommendation value of the Mth user for the first short video real-time content, S M2 is the recommendation value of the Mth user for the real-time content of the second short video, S MN is the recommendation value of the Mth user for the Nth short video real-time content, and S ij ∈S, where S ij is the recommendation value of the i-th user for the j-th short video real-time content, and i∈M, j∈N, for S ij , define the user vector as U i , the short video real-time content quantization value vector is V j ,satisfy: in Indicates the Mth i Group user vector, Indicates the Nth j A group of short video real-time content quantization value vectors.

[0014] Furthermore, for Construct the overall characteristic index w1 ij , w2 ij ,w3 ij , define the first comprehensive characteristic index w1 ij It is expressed as follows:

[0015]

[0016] Among them, ||U i || means U i The scale of the vector, ||V j || represents vector V j The scale of θ represents the overall threshold. The first overall feature index tests the difference between the user vector and the short video real-time content quantization value vector. If w1 ij The higher the value, the higher the short video live broadcast automatic diversion system based on the real-time content recommendation algorithm will give priority to the short video real-time content quantization value vector V j Recommended to U i , the second overall characteristic index w2 ij It is expressed as follows:

[0017]

[0018] Among them, S i，kis the recommendation value of the i-th user for the k-th short video real-time content, k∈{1, max(M, N)} is a constant. The logarithmic function enables the sum value to converge quickly, speeds up the response speed of the system, and quickly improves the difference between the short video real-time content recommended by the system to the user's preference and the short video real-time content recommended by the system to the user's non-preference. The third comprehensive feature index w3 ij It is expressed as follows:

[0019]

[0020] in, represents the substitute domain of the user vector, A substitute domain representing the quantitative value of the real-time content of short videos, Representing the splicing operation of the substitute domain, the present invention defines the constraint functions of three comprehensive feature indicators as follows:

[0021] aw1 ij +bw2 ij +cw3 ij =1

[0022] Among them, a, b, and c are the debugging factors of the constraint function. In order to recommend the most suitable short video real-time content to users, the difference with the true value Y ij The present invention proposes a deviation function L ij , defined as follows:

[0023]

[0024] According to the meaning of the three comprehensive characteristic indicators, the above formula is updated to a constrained deviation function:

[0025]

[0026] Among them, ||Y ij || indicates Y ij The scale of , λ represents the regularization factor, and the definition It is a subjective coordination item, that is, a direct relationship is established between the user preference matrix and the true value. As the first objective coordination item, For the second objective coordination item, that is, the relationship between the user vector, the quantized value of the real-time content of the short video, and the true value, in order to prevent the coordination overflow phenomenon of the deviation function, the regularization factor λ is added to weaken the coordination overflow effect brought by the subjective coordination item.

[0027] Furthermore, for the first objective coordination item, the present invention further optimizes it, and sets g ij is the first objective coordination item, that is Its gradient vector is:

[0028]

[0029] Since the user vector and the short video real-time content quantization value proposed in the present invention are interrelated, the present invention improves the ant system algorithm so that the deviation function explores the path in two directions: the user vector and the short video real-time content quantization value. Assume that at time t, ant k explores user U i and short video real-time content quantification value V j Probability for:

[0030]

[0031] The above formula satisfies the following constraints:

[0032]

[0033] Among them, τ ij (t) represents the number of times ant k searches for user U i and short video real-time content quantification value V j The pheromone, η ij (t) represents the number of times ant k searches for user U i and short video real-time content quantification value V j The heuristic factor, τ i,q (t) represents the number of times ant k searches for user U i and short video real-time content quantification value V q The pheromone, η i,q (t) represents the number of times ant k searches for user U i and short video real-time content quantification value V q , α is the pheromone hyperparameter, β is the heuristic factor hyperparameter, α1 and α2 are the sub-hyperparameters of the pheromone of α, β1 and β2 are the sub-hyperparameters of the heuristic factor of β, q is a constant, when ant k completes a traversal, the pheromone is updated to: τ ij (t+Δt)=(1-ρ)τ ij (t)+Δτ ij , where Δt is the intermittent time, ρ (0<ρ<1) represents the pheromone evaporation coefficient on the path, 1-ρ represents the pheromone persistence coefficient, and the Δτ ij represents the pheromone increment of this iteration. For the improved ant system algorithm provided by the present invention, the best recommendation time found by the crawler history is T best , then the improved pheromone is: Where ε is the adjustment T best Parameters affecting weights, improved optimal pheromone It is expressed as follows:

[0034]

[0035] Among them, L k i ,V j > is the length of the optimal path of the ant k exploration deviation function in both directions of the user vector and the quantized value of the short video real-time content.

[0036] Furthermore, the personalized recommendation module recommends the most suitable short video live content to the user based on the user's interests and preferences. The personalized recommendation module establishes the user's interest model by analyzing the user's historical behavior and preferences. The interest model includes the user's interest tags and interest weight information, which is used to evaluate the user's preference for different types of short video live content. The recommendation priority is determined by calculating the similarity between the user's interest model and the video live content. The similarity calculation is based on multiple factors such as the similarity of the content and the similarity of the user's behavior. Finally, the recommendation result is generated based on the user's interest model and the results of the similarity calculation. The recommendation result is a group of top-ranked short video live content, sorted according to the user's personalized preferences. The recommendation result includes the live content that the user is interested in and related popular live content.

[0037] Furthermore, the diversion strategy module formulates a suitable diversion strategy, that is, to effectively guide the recommended short video live broadcast content to users, so as to increase users' click-through rate and viewing time. The diversion strategy module first sets the goal of diversion, which is to increase users' click-through rate, increase users' viewing time, and enhance users' interactive behavior. According to the needs and goals of the system, the diversion strategy module formulates corresponding strategies to optimize the diversion position of the recommended content to improve the diversion effect. The diversion position is displayed on the user's homepage, related video page, and recommendation list. The module determines the best diversion position based on the user's browsing behavior and page layout factors. At the same time, the diversion strategy module also needs to be optimized in real time. According to user feedback and behavior data, the diversion strategy is adjusted. By real-time monitoring of users' click behavior, viewing time, and interactive behavior indicators, the module adjusts the diversion strategy in time to provide recommendation results that are more in line with user interests and needs.

[0038] ​Beneficial effects of the present invention: the system of the present invention can accurately understand the user's interests, keywords and preferences by collecting and analyzing user data in real time, thereby realizing personalized recommendation. Through personalized recommendation, the system can recommend to users the short video live broadcast content that best suits their interests and preferences, thereby improving the user's click-through rate and viewing time. Through personalized recommendation and diversion strategies, the system of the present invention can guide users to watch and participate in more short video live broadcast content, and users can more easily find content of interest, thereby improving their participation and interactive behavior, including liking, commenting, and sharing, thereby enhancing the interaction and stickiness between users and the platform. Through the real-time content analysis module, the short video live broadcast content can be quickly and accurately analyzed and evaluated, and the system can screen out high-quality content, avoid recommending low-quality or repetitive content, and improve the content quality of the platform and user satisfaction. At the same time, the system can also recommend diversified content to meet the different interests and needs of users. The innovation of the present invention lies in that three comprehensive feature indicators are provided. The subjective coordination item, the first objective coordination item and the second objective coordination item are mapped respectively. The subjective coordination item guides the system to find the user's preference. The logarithmic function is used to make the sum value converge quickly, which speeds up the response speed of the system and can quickly improve the difference between the short video real-time content recommended to the user's preference and the short video real-time content recommended to the user's non-preference. The three coordinated feature indicators are used to establish the functional relationship between the user vector and the short video real-time content quantization value and the relationship between the real value. In order to prevent the coordinated overflow of the deviation function, the regularization factor λ is added to weaken the coordinated overflow effect brought by the subjective coordination item. Furthermore, the problem that the gradient descent strategy under the traditional one-dimensional condition can only find the optimal gradient in one direction is solved. The improved ant system algorithm proposed in the present invention enables the deviation function to explore the path in two directions of the user vector and the short video real-time content quantization value, ensuring that the fastest and optimal gradient descent strategy is obtained without affecting the correlation between the user vector and the short video real-time content quantization value. The system of the present invention can effectively guide users to watch and participate in short video live broadcast content through the diversion strategy module, thereby improving users' click-through rate and viewing time. The system can accurately divert the content most suitable for users to users through weight calculation and diversion position optimization, thereby improving users' click and viewing conversion rates for recommended content, thereby improving the platform's traffic conversion effect. By adopting real-time data collection and analysis, it can accurately grasp users' real-time needs and interests, and timely adjust recommendation strategies and diversion positions. The system can provide real-time and personalized recommendation results based on users' real-time behavior and feedback, thereby enhancing user experience and participation. In addition, by providing personalized and accurate content recommendation and diversion services, the system of the present invention can increase user stickiness and loyalty, improve the platform's user retention rate, and users can obtain better experience and satisfaction on the platform, and are more likely to become loyal users, thereby increasing the platform's user base and influence, and improving the platform's competitiveness in the market. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The invention is further illustrated by the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0040] Figure 1 It is a structural schematic diagram of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] See also Figure 1 The present invention will be further described in conjunction with the following examples.

[0043] See also Figure 1 The present invention aims to provide a short video live broadcast automatic diversion system based on a real-time content recommendation algorithm, including a user data collection module, a real-time content analysis module, a content matching module, a personalized recommendation module and a diversion strategy module, wherein the user data collection module is responsible for collecting the user's viewing history, interests and hobbies, and behavior pattern data, and updating it in real time. By collecting and analyzing user data, the system understands the user's interests and preferences. The real-time content analysis module analyzes the user data in real time, extracts the user's interests and keywords, and identifies the user's emotional state and preferences, and quickly and accurately understands the user's needs. The content matching module selects video live broadcast content that meets the user's interests from the content library of the short video live broadcast platform according to the user's interests and keywords. The system finds short video live broadcast content related to the user's interests. The personalized recommendation module recommends the most suitable short video live broadcast content to the user based on the user's interests and preferences. According to the user's historical viewing records and behavior patterns, the system provides personalized recommendation results. The diversion strategy module formulates a diversion strategy based on the user's behavior pattern and the overall strategy of the platform, and diverts the recommended short video live broadcast content to the user. The system flexibly adjusts the diversion strategy according to the user's preferences and the platform's goals to improve user participation and platform traffic conversion.

[0044] Specifically, it is characterized in that the implementation method of the user data collection module includes data acquisition, data storage, data update, data cleaning and processing. Relevant user data is collected through the user's operations and behaviors on the short video live broadcast platform. These data include the user's viewing history, likes and comment information, sharing behavior, followed anchors or topics that can be obtained by this system. Data acquisition is carried out through user login account, browsing history, and operation log. The collected user data needs to be stored for subsequent real-time content analysis and personalized recommendation. Database technology is used to store user data in a structured manner, and ensure data security and privacy protection. By monitoring the user's operations and behaviors in this system, user data is updated in time to reflect the user's latest interests and preferences. Timed updates and real-time updates are used to update and synchronize data according to the user's activities. If the collected user data contains noise and redundant information, it needs to be cleaned and processed. Data preprocessing technology, including removing duplicate data and processing outliers, is used to ensure the accuracy and availability of user data.

[0045] Specifically, the real-time content analysis module extracts users' interests and keywords and identifies users' emotional states and preferences by performing real-time analysis on user data. By analyzing users' viewing history, likes and comments, the real-time content analysis module identifies users' interests and extracts keywords from their viewing history and interactive behaviors. These keywords reflect users' focus and areas of interest, which are helpful for subsequent content matching and personalized recommendations. In order to better understand users' emotional states and preferences, the real-time content analysis module performs sentiment analysis to analyze the emotions expressed by users in comments and likes, such as likes, dislikes, and excitement, in order to better understand users' preferences and feedback. At the same time, the real-time content analysis module needs to update user data in real time. As users' behaviors and preferences change, the module should be able to update the analysis results in a timely manner to provide accurate real-time recommendations and traffic diversion services.

[0046] Specifically, the content matching module uses the real-time content analysis module to systematically understand the user's interests, keywords, emotional state and preferences, providing a basis for subsequent content matching and personalized recommendations. By analyzing user data in real time, it provides users with short video live content related to their interests, thereby improving user experience and platform traffic conversion effects. The content matching module first needs to build a content library containing various types of short video live content. The content library includes short video live broadcasts of different themes and types, including entertainment, sports, food, and music. The construction of the content library is carried out through manual screening, automatic crawling, and content creator uploading. The content matching module analyzes the user's keywords and interests, and performs keyword matching with the video live content in the content library. By matching user interests and keywords, the module filters out video live content related to user interests.

[0047] Specifically, in order to more accurately match the interests of users, the content matching module uses the labels and classification information of the live video content to achieve more accurate matching by labeling and classifying the live video content, ensuring that the content recommended to users is consistent with their interests, and conducting quality assessments on the screened live video content. The assessment indicators include the number of views, likes, and comments. By evaluating the quality of the content, the module prioritizes recommending high-quality live video content to users, thereby improving user experience and the attractiveness of the content.

[0048] Specifically, in the content matching module, the user preference matrix is ​​defined as Among them, S 11 is the recommendation value of the first user for the first short video real-time content, S 12 is the recommendation value of the first user for the real-time content of the second short video, S 1N is the recommendation value of the first user for the real-time content of the Nth short video, S 21 is the recommendation value of the second user for the real-time content of the first short video, S 22 is the recommendation value of the second user for the real-time content of the second short video, S 2N is the recommendation value of the second user for the real-time content of the Nth short video, S M1 is the recommendation value of the Mth user for the first short video real-time content, S M2 is the recommendation value of the Mth user for the real-time content of the second short video, S MN is the recommendation value of the Mth user for the Nth short video real-time content, and S ij ∈S, where S ij is the recommendation value of the i-th user for the j-th short video real-time content, and i∈M, j∈N, for S ij , define the user vector as U i , the short video real-time content quantization value vector is V j ,satisfy: in Indicates the Mth i Group user vector, Indicates the Nth j A group of short video real-time content quantization value vectors.

[0049] Preferably, for Construct the overall characteristic index w1 ij , w2 ij ,w3 ij , define the first comprehensive characteristic index w1 ij It is expressed as follows:

[0050]

[0051] Among them, ||U i || means U i The scale of the vector, ||V j || represents vector V j The scale of θ represents the overall threshold. The first overall feature index tests the difference between the user vector and the short video real-time content quantization value vector. If w1 ij The higher the value, the higher the short video live broadcast automatic diversion system based on the real-time content recommendation algorithm will give priority to the short video real-time content quantization value vector V j Recommended to U i , the second overall characteristic index w2 ij It is expressed as follows:

[0052]

[0053] Among them, S i,k is the recommendation value of the i-th user for the k-th short video real-time content, k∈{1, max(M,N)} is a constant. The logarithmic function enables the sum value to converge quickly, speeds up the response speed of the system, and quickly improves the difference between the short video real-time content recommended by the system to the user's preference and the short video real-time content recommended by the system to the user's non-preference. The third comprehensive feature index w3 ij It is expressed as follows:

[0054]

[0055] in, represents the substitute domain of the user vector, A substitute domain representing the quantitative value of the real-time content of short videos, Representing the splicing operation of the substitute domain, the present invention defines the constraint functions of three comprehensive feature indicators as follows:

[0056] aw1 ij +bw2 ij +cw3 ij =1

[0057] Among them, a, b, and c are the debugging factors of the constraint function. In order to recommend the most suitable short video real-time content to users, the difference with the true value Y ij The present invention proposes a deviation function L ij , defined as follows:

[0058]

[0059] According to the meaning of the three comprehensive characteristic indicators, the above formula is updated to a constrained deviation function:

[0060]

[0061] Among them, ||Y ij || indicates Y ij The scale of , λ represents the regularization factor, and the definition It is a subjective coordination item, that is, a direct relationship is established between the user preference matrix and the true value. As the first objective coordination item, For the second objective coordination item, that is, the relationship between the user vector, the quantized value of the real-time content of the short video, and the true value, in order to prevent the coordination overflow phenomenon of the deviation function, the regularization factor λ is added to weaken the coordination overflow effect brought by the subjective coordination item.

[0062] Preferably, for the first objective coordination item, the present invention further optimizes it, and sets g ij is the first objective coordination item, that is Its gradient vector is:

[0063]

[0064] Since the user vector and the short video real-time content quantization value proposed in the present invention are interrelated, the present invention improves the ant system algorithm so that the deviation function explores the path in two directions: the user vector and the short video real-time content quantization value. Assume that at time t, ant k explores user U i and short video real-time content quantification value V j Probability for:

[0065]

[0066] The above formula satisfies the following constraints:

[0067]

[0068] Among them, τ ij (t) represents the number of times ant k searches for user U i and short video real-time content quantification value V j The pheromone, η ij (t) represents the number of times ant k searches for user U i and short video real-time content quantification value V j The heuristic factor, τ i,q (t) represents the number of times ant k searches for user U i and short video real-time content quantification value V q The pheromone, η i,q (t) represents the number of times ant k searches for user U i and short video real-time content quantification value V q, α is the pheromone hyperparameter, β is the heuristic factor hyperparameter, α1 and α2 are the sub-hyperparameters of the pheromone of α, β1 and β2 are the sub-hyperparameters of the heuristic factor of β, q is a constant, when ant k completes a traversal, the pheromone is updated to: τ ij (t+Δt)=(1-ρ)τ ij (t)+Δτ ij , where Δt is the intermittent time, ρ (0<ρ<1) represents the pheromone evaporation coefficient on the path, 1-ρ represents the pheromone persistence coefficient, and the Δτ ij represents the pheromone increment of this iteration. For the improved ant system algorithm provided by the present invention, the best recommendation time found by the crawler history is T best , then the improved pheromone is: Where ε is the adjustment T best Parameters affecting weights, improved optimal pheromone It is expressed as follows:

[0069]

[0070] Among them, L k i ,V j > is the length of the optimal path of the ant k exploration deviation function in both directions of the user vector and the quantized value of the short video real-time content.

[0071] Preferably, the personalized recommendation module recommends the most suitable short video live content to the user based on the user's interests and preferences. The personalized recommendation module establishes the user's interest model by analyzing the user's historical behavior and preferences. The interest model includes the user's interest tags and interest weight information, which is used to evaluate the user's preference for different types of short video live content. The recommendation priority is determined by calculating the similarity between the user's interest model and the video live content. The similarity calculation is based on multiple factors such as the similarity of the content and the similarity of the user's behavior. Finally, the recommendation result is generated based on the user's interest model and the results of the similarity calculation. The recommendation result is a group of top-ranked short video live content, which is sorted according to the user's personalized preferences. The recommendation result includes the live content that the user is interested in and related popular live content.

[0072] ​Preferably, the diversion strategy module formulates a suitable diversion strategy, that is, to effectively guide the recommended short video live broadcast content to users, so as to increase users' click-through rate and viewing time. The diversion strategy module first sets the goal of diversion, which is to increase users' click-through rate, increase users' viewing time, and enhance users' interactive behavior. According to the needs and goals of the system, the diversion strategy module formulates corresponding strategies to optimize the diversion position of the recommended content to improve the diversion effect. The diversion position is displayed on the user's homepage, related video page, and recommendation list. The module determines the best diversion position based on the user's browsing behavior and page layout factors. At the same time, the diversion strategy module also needs to be optimized in real time. According to user feedback and behavior data, the diversion strategy is adjusted. By real-time monitoring of users' click behavior, viewing time, and interactive behavior indicators, the module adjusts the diversion strategy in time to provide recommendation results that are more in line with user interests and needs.

[0073] Beneficial effects of the present invention: the system of the present invention can accurately understand the user's interests, keywords and preferences by collecting and analyzing user data in real time, thereby realizing personalized recommendation. Through personalized recommendation, the system can recommend to users the short video live broadcast content that best suits their interests and preferences, thereby improving the user's click-through rate and viewing time. Through personalized recommendation and diversion strategies, the system of the present invention can guide users to watch and participate in more short video live broadcast content, and users can more easily find content of interest, thereby improving their participation and interactive behavior, including liking, commenting, and sharing, thereby enhancing the interaction and stickiness between users and the platform. Through the real-time content analysis module, the short video live broadcast content can be quickly and accurately analyzed and evaluated, and the system can screen out high-quality content, avoid recommending low-quality or repetitive content, and improve the content quality of the platform and user satisfaction. At the same time, the system can also recommend diversified content to meet the different interests and needs of users. The innovation of the present invention lies in that three comprehensive feature indicators are provided. The subjective coordination item, the first objective coordination item and the second objective coordination item are mapped respectively. The subjective coordination item guides the system to find the user's preference. The logarithmic function is used to make the sum value converge quickly, which speeds up the response speed of the system and can quickly improve the difference between the short video real-time content recommended to the user's preference and the short video real-time content recommended to the user's non-preference. The three coordinated feature indicators are used to establish the functional relationship between the user vector and the short video real-time content quantization value and the relationship between the real value. In order to prevent the coordinated overflow of the deviation function, the regularization factor λ is added to weaken the coordinated overflow effect brought by the subjective coordination item. Furthermore, the problem that the gradient descent strategy under the traditional one-dimensional condition can only find the optimal gradient in one direction is solved. The improved ant system algorithm proposed in the present invention enables the deviation function to explore the path in two directions of the user vector and the short video real-time content quantization value, ensuring that the fastest and optimal gradient descent strategy is obtained without affecting the correlation between the user vector and the short video real-time content quantization value. :The system of the present invention can effectively guide users to watch and participate in short video live broadcast content through the diversion strategy module, thereby improving users' click-through rate and viewing time. The system can accurately divert the content that is most suitable for users to users through weight calculation and diversion position optimization, thereby improving users' click and viewing conversion rates for recommended content, thereby improving the platform's traffic conversion effect. By using real-time data collection and analysis, it can accurately grasp users' real-time needs and interests, and timely adjust recommendation strategies and diversion positions. The system can provide real-time and personalized recommendation results based on users' real-time behavior and feedback, thereby enhancing user experience and participation. In addition, by providing personalized and accurate content recommendation and diversion services, the system of the present invention can increase user stickiness and loyalty, improve the platform's user retention rate, and users can get better experience and satisfaction on the platform, and are more likely to become loyal users, thereby increasing the platform's user base and influence, and improving the platform's competitiveness in the market.

[0074] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A short video live broadcast automatic diversion system based on real-time content recommendation algorithm, characterized in that: It includes user data collection module, real-time content analysis module, content matching module, personalized recommendation module and diversion strategy module; among them, the user data collection module is responsible for collecting the user's viewing history, interests and hobbies, and behavior pattern data, and updating it in real time. By collecting and analyzing user data, the system understands the user's interests and preferences; the real-time content analysis module analyzes user data in real time, extracts the user's interests and keywords, and identifies the user's emotional state and preferences, and quickly and accurately understands the user's needs; the content matching module selects video live broadcast content that meets the user's interests from the content library of the short video live broadcast platform based on the user's interests and keywords, and the system finds short video live broadcast content related to the user's interests; the personalized recommendation module recommends the most suitable short video live broadcast content to the user based on the user's interests and preferences, and the system provides personalized recommendation results based on the user's historical viewing records and behavior patterns; the diversion strategy module formulates a diversion strategy based on the user's behavior pattern and the overall strategy of the platform, and diverts the recommended short video live broadcast content to the user. The system adjusts the diversion strategy according to the user's preferences and the platform's goals to improve user participation and platform traffic conversion; The content matching module uses the real-time content analysis module to help the system understand the user's interests, keywords, emotional state, and preferences, providing a basis for subsequent content matching and personalized recommendations; The content matching module uses the labels and classification information of the live video content to label and classify the live video content, and conducts quality assessment on the selected live video content. The assessment indicators include the number of views, the number of likes, and the number of comments. The content matching module gives priority to recommending high-quality live video content to users. In the content matching module, the user preference matrix is ​​defined as Among them, S 11 is the recommendation value of the proposed system for the first user for the first short video real-time content, S 12 is the recommendation value of the proposed system for the first user for the second short video real-time content, S1N is the recommendation value of the proposed system for the first user for the Nth short video real-time content, S 21 is the recommendation value of the proposed system for the second user for the real-time content of the first short video, S 22 is the recommendation value of the proposed system for the second user for the second short video real-time content, M2N is the recommendation value of the proposed system for the second user for the Nth short video real-time content, S M1 is the recommendation value of the proposed system for the real-time content of the first short video for the Mth user, S M2 is the recommendation value of the proposed system for the real-time content of the second short video for the Mth user, S MN is the recommendation value of the proposed system for the real-time content of the Nth short video for the Mth user, and there is S ij ∈S, where S ij is the recommendation value of the proposed system for the jth short video real-time content by the i-th user, and i∈M, j∈N, for S ij , define the user vector as U i , the short video real-time content quantization value vector is V j ,satisfy: in Indicates the Mth i Group user vector, Indicates the Nth j Group short video real-time content quantization value vector; for Construct the overall characteristic index w1 ij ,w2 ij ,w3 ij , define the first comprehensive characteristic index w1 ij It is expressed as follows: Among them, ||U i || means U i The scale of the vector, ||V j || represents vector V j The scale of θ represents the overall threshold. The first overall characteristic index tests the difference between the user vector and the short video real-time content quantization value vector. If w1 ij The higher the value is, the more the short video live broadcast automatic diversion system based on the real-time content recommendation algorithm will give priority to the short video real-time content quantization value vector V j Recommended to U i , the second overall characteristic index w2 ij It is expressed as follows: Among them, S i,k is the recommendation value of the proposed system for the kth short video real-time content of the ith user, k∈{1, max(M,N)} is a constant. The logarithmic function is used to make the sum value converge quickly, speed up the response speed of the system, and quickly improve the difference between the short video real-time content recommended by the system to the user's preference and the short video real-time content recommended by the system to the user's non-preference. The third comprehensive feature index w3 ij It is expressed as follows: in, represents the substitute domain of the user vector, A substitute domain representing the quantitative value of the real-time content of short videos, Represents the concatenation operation of the substitute domain, and defines the constraint functions of the three overall characteristic indicators as: <h2 style=";text-align:left;direction:ltr">aw1<h2 style=";text-align:left;direction:ltr"> ij <h2 style=";text-align:left;direction:ltr"> +bw2<h2 style=";text-align:left;direction:ltr"> ij <h2 style=";text-align:left;direction:ltr"> +cw3<h2 style=";text-align:left;direction:ltr"> ij <h2 style=";text-align:left;direction:ltr"> =1 Among them, a, b, c are the debugging factors of the constraint function, which reduce the error with the true value Y ij The difference between ij , defined as follows: According to the meaning of the three comprehensive characteristic indicators, the above formula is updated to a constrained deviation function: Among them, ||Y ij || represents Y ij The scale of , λ represents the regularization factor, and the definition It is a subjective coordination item, that is, a direct relationship is established between the user preference matrix and the true value. As the first objective coordination item, To establish a relationship between the second objective coordination item, that is, the interaction relationship between the user vector and the quantized value of the real-time content of the short video, and the true value, a regularization factor λ is added to weaken the coordination spillover effect brought by the subjective coordination item.

2. According to claim 1, a short video live broadcast automatic diversion system based on real-time content recommendation algorithm is characterized in that: The implementation methods of the user data collection module include data acquisition, data storage, data update, data cleaning and processing, and collect relevant user data through the user's operations and behaviors on the short video live broadcast platform.

3. According to claim 1, a short video live broadcast automatic diversion system based on real-time content recommendation algorithm is characterized in that: The real-time content analysis module extracts users' interests and keywords and identifies users' emotional states and preferences by performing real-time analysis on user data. By analyzing users' viewing history, likes and comments, the real-time content analysis module identifies users' interests and extracts keywords from their viewing history and interactive behaviors. Keywords reflect users' focus and areas of interest, which are helpful for subsequent content matching and personalized recommendations. The real-time content analysis module performs sentiment analysis and analyzes the emotions expressed by users in comments and likes.

4. According to claim 1, a short video live broadcast automatic diversion system based on real-time content recommendation algorithm is characterized in that: The personalized recommendation module recommends the most suitable short video live content to users based on their interests and preferences. The personalized recommendation module establishes a user interest model by analyzing the user's historical behavior and preferences. The interest model includes the user's interest tags and interest weight information, which is used to evaluate the user's preference for different types of short video live content. The priority of the recommendation is determined by calculating the similarity between the user's interest model and the live video content. The similarity calculation is based on multiple factors such as the similarity of the content and the similarity of the user's behavior. Finally, the recommendation result is generated based on the user's interest model and the results of the similarity calculation.

5. According to claim 1, a short video live broadcast automatic diversion system based on real-time content recommendation algorithm is characterized in that: The traffic diversion strategy module formulates appropriate traffic diversion strategies, that is, effectively directing the recommended short video live broadcast content to users to increase users' click-through rate and viewing time. The traffic diversion strategy module first sets the goal of traffic diversion, which is to increase users' click-through rate, increase users' viewing time, and enhance users' interactive behavior. According to the needs and goals of the system, the traffic diversion strategy module formulates corresponding strategies to optimize the diversion position of recommended content to improve the diversion effect.

Images