AI-driven multi-dimensional short video matrix analysis and management system

Abstract

The present application provides an AI-driven multi-dimensional short video matrix analysis and management system, which relates to the field of information prediction technology. Audience modal information is obtained when a short video is played by performing audience verification on a traffic feature set, and click attribute information of a short video user when clicking on a short video is determined based on the audience modal information; and the audience acceptance of the short video content is determined based on the access and likes data, and the delivery matching coefficient when the short video is delivered is determined based on the audience acceptance and click attribute information; the perceived weirdness of the short video user when watching the short video is determined based on the behavioral feature information; and then the dynamic interest tags of the short video user are trend analyzed based on the delivery matching coefficient and the user behavior sequence after the correction constraint, so as to recommend short videos to the short video user. The present application can improve the user experience and platform activity under the influence of the difficulty in accurately capturing the dynamic interests and complex preferences of users from the recommended content.

Description

Technical Field

[0001] The present application relates to the field of information prediction technology, and more specifically, to an AI-driven multi-dimensional short video matrix analysis and management system. Background Art

[0002] Information prediction refers to the use of algorithms and models to predict future trends, behaviors or events by analyzing historical data, user behavior, environmental changes and other factors. It is based on existing data and existing rules to infer unknown situations or things that may happen in the future. Information prediction is widely used in various fields, such as e-commerce recommendations, stock market forecasts, weather forecasts, traffic flow analysis, etc. It is optimized through technologies such as big data, machine learning, and artificial intelligence to make predictions more accurate and real-time. In short video recommendation systems, information prediction usually refers to predicting what users may be interested in in the future by analyzing user historical viewing data, behavioral characteristics, social interactions and other information.

[0003] In the prior art, AI-driven multi-dimensional short video matrix analysis and management systems usually use artificial intelligence, big data analysis and machine learning algorithms to conduct multi-level analysis of short video content and user behavior, optimize video recommendations and content distribution, and then collect and process user viewing history, interactive behavior (such as likes, comments, sharing, etc.) and social data to build a user interest model and match this information with the content characteristics of short videos. Through deep learning algorithms, the system can identify and predict changes in user interests and dynamically adjust recommended content. However, short video recommendation systems usually rely on traditional user interest models and behavioral data analysis. These models are prone to problems such as reduced accuracy and insufficient personalization when processing massive, multi-dimensional user behavior data, making it difficult for recommended content to accurately capture users' dynamic interests and complex preferences, resulting in the gradual weakening of the effect of recommendation results, and thus failing to effectively improve user experience and platform activity. Therefore, how to achieve the improvement of user experience and platform activity under the influence of the difficulty in accurately capturing users' dynamic interests and complex preferences from recommended content is a problem faced by the industry. Summary of the invention

[0004] This application provides an AI-driven multi-dimensional short video matrix analysis and management system, which can improve user experience and platform activity when it is difficult to accurately capture users' dynamic interests and complex preferences from recommended content.

[0005] This application provides an AI-driven multi-dimensional short video matrix analysis and management system, which includes:

[0006] A model building module, which is used to build a set of traffic features when playing short videos using AI-driven big data structures;

[0007] An attribute determination module is used to perform audience verification on the traffic feature set to obtain audience modality information when the short video is played, and determine click attribute information when the short video user clicks on the short video based on the audience modality information;

[0008] A delivery matching module is used to obtain the access and likes data of the target short video, determine the audience acceptance of the short video content according to the access and likes data, and determine the delivery matching coefficient when delivering the short video according to the audience acceptance and the click attribute information;

[0009] A correction constraint module is used to determine the behavior feature information of the short video user when operating the short video, and determine the perceived weirdness of the short video user when watching the short video according to the behavior feature information; and correct the user behavior sequence of the short video user by using the perceived weirdness;

[0010] The video recommendation module is used to perform trend analysis on the dynamic interest tags of short video users according to the delivery matching coefficient and the user behavior sequence after the correction constraints, so as to recommend short videos to the short video users.

[0011] In this embodiment, the traffic feature set represents a structured data set of various types of interactive data (such as viewing time, click frequency, bounce rate, etc.) generated by users during the short video playback process.

[0012] In this embodiment, the audience verification of the traffic feature set to obtain audience modality information when the short video is played specifically includes:

[0013] Extracting a playback index dimension when the short video is played from the traffic feature set;

[0014] The text modality information and the audio text information when the short video is played are matched according to the playback index dimension to obtain the audience modality information when the short video is played.

[0015] In this embodiment, the audience modality information represents preference data of the main interests and behavioral characteristics of the viewing group.

[0016] In this embodiment, audience acceptance is an indicator for measuring the degree of acceptance and preference of users for a short video.

[0017] In this embodiment, determining the delivery matching coefficient when delivering the short video based on the audience acceptance and the click attribute information specifically includes:

[0018] Determining the preference breadth information of short video users when browsing recommended videos according to the audience acceptance;

[0019] Extracting the preference adoption features when delivering short videos from the click attribute information;

[0020] The delivery matching coefficient when delivering the short video is determined according to the preference breadth information and the preference adoption characteristics.

[0021] In this embodiment, determining the behavior characteristic information of the short video user when operating the short video specifically includes:

[0022] Determine the data sample points when the short video user operates the short video;

[0023] Determine the density of operation behaviors of short video users when operating short videos;

[0024] Behavior feature information of a short video user when operating a short video is determined according to the data sample points and the operation behavior density.

[0025] In this embodiment, determining the perceived weirdness of a short video user when watching a short video according to the behavior characteristic information specifically includes:

[0026] Determine the perception change nodes of short video users when watching short videos;

[0027] Performing multi-dimensional analysis on the behavior feature information through the perception change node to obtain a behavior proximity distance;

[0028] The perceived weirdness of a short video user when watching the short video is determined according to the behavior proximity distance.

[0029] In this embodiment, the perceived weirdness indicates the degree of behavioral consistency and emotional coherence of the user when watching the short video.

[0030] In this embodiment, the user behavior sequence is a record of continuous interactive behaviors performed by the user when watching a short video.

[0031] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects:

[0032] A traffic feature set for short video playback is established by using an AI-driven big data structure; audience verification is performed on the traffic feature set to obtain audience modal information when the short video is played, and click attribute information of the short video user when clicking on the short video is determined based on the audience modal information; by obtaining the access and like data of the target short video, the audience acceptance of the short video content is determined based on the access and like data, and the delivery matching coefficient when the short video is delivered is determined based on the audience acceptance and the click attribute information; by determining the behavioral feature information when the short video user operates the short video, the perceived weirdness of the short video user when watching the short video playback is determined based on the behavioral feature information; the user behavior sequence of the short video user is corrected and constrained by the perceived weirdness; the dynamic interest tags of the short video user are trend analyzed based on the delivery matching coefficient and the user behavior sequence after the correction and constraint, so as to recommend short videos to the short video user.

[0033] It can be seen that in this application, matrix analysis and management of multi-dimensional short videos are achieved under the influence of the difficulty in accurately capturing the dynamic interests and complex preferences of users from the recommended content; among them, by building a comprehensive set of traffic characteristics, the system can accurately capture and analyze key data such as users' viewing behavior and interaction habits, provide data support for subsequent personalized recommendations, and improve the accuracy and timeliness of recommendations; audience verification ensures in-depth analysis of different user groups, and can more accurately identify the preferences and needs of each user, so that content recommendations are closer to users' interests, avoid the push of irrelevant content, and improve user participation and satisfaction; by analyzing user click attributes Through precise analysis, the system can gain an in-depth understanding of users’ specific behavioral preferences, such as video type and duration, so as to more effectively match user needs when making recommendations, optimize click-through rate and viewing time, and improve user experience; by analyzing likes data, the system can understand the audience feedback and popularity of short videos in real time, so as to optimize content distribution strategies, enhance the attractiveness of platform content, and increase user activity and platform stickiness; through trend analysis and continuous updating of dynamic interest tags, the system can make real-time adjustments based on changes in user interests, ensure that recommended content always matches the user’s current interests, improve the personalization and accuracy of recommendations, and enhance the user’s long-term stickiness and platform activity.

[0034] To sum up, the technical solution adopted in this application can improve user experience and platform activity when it is difficult to accurately capture users' dynamic interests and complex preferences from recommended content. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 It is a module structure diagram of the AI-driven multi-dimensional short video matrix analysis and management system provided by this application;

[0037] Figure 2 is a schematic diagram of a process for determining click attribute information provided by this application;

[0038] Figure 3 It is a flowchart of determining audience acceptance according to the application. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0040] The embodiment of the present application provides an AI-driven multi-dimensional short video matrix analysis and management system, the core of which is to use an AI-driven big data structure to establish a flow feature set when a short video is played; to perform audience verification on the flow feature set to obtain audience modal information when the short video is played, and to determine the click attribute information when the short video user clicks on the short video based on the audience modal information; to obtain the access like data of the target short video, to determine the audience acceptance of the short video content based on the access like data, and to determine the delivery matching coefficient when the short video is delivered based on the audience acceptance and the click attribute information; to determine the behavioral feature information when the short video user operates the short video, and to determine the perceived weirdness when the short video user watches the short video according to the behavioral feature information; to correct and constrain the user behavior sequence of the short video user through the perceived weirdness; to perform trend analysis on the dynamic interest tags of the short video user based on the delivery matching coefficient and the corrected user behavior sequence, so as to recommend short videos to the short video user. The above scheme is adopted to improve the user experience and platform activity under the influence of the difficulty in accurately capturing the dynamic interests and complex preferences of users from the recommended content.

[0041] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods. Figure 1 As shown, this figure is a module structure diagram of the AI-driven multi-dimensional short video matrix analysis and management system shown in this embodiment of the present application. The analysis and management system includes: a model building module 100, an attribute determination module 200, a delivery matching module 300, a correction constraint module 400 and a video recommendation module 500, which are respectively described as follows:

[0042] The model building module 100 is used to establish a set of traffic features when playing short videos using an AI-driven big data structure.

[0043] In the specific implementation, the interactive data of users during the short video playback process is first collected through AI algorithms, including video viewing time, click frequency, bounce rate, playback completion rate and other behavioral characteristics; then, these multi-source data are preprocessed (such as removing outliers and normalizing) to ensure the consistency and reliability of the data. Next, a feature extraction algorithm (such as principal component analysis PCA or clustering algorithm) is used to compress the high-dimensional data into a more streamlined flow feature set. In other embodiments, other methods can also be used to determine the flow feature set, which will not be repeated here.

[0044] It should be noted that, in the present application, the traffic feature set represents a structured data set of various types of interactive data (such as viewing time, click frequency, bounce rate, etc.) generated by users during the short video playback process.

[0045] The attribute determination module 200 is used to perform audience verification on the traffic feature set to obtain audience modality information when the short video is played, and determine click attribute information when the short video user clicks on the short video based on the audience modality information.

[0046] In this embodiment, audience verification of the traffic feature set to obtain audience modality information when the short video is played can be implemented by the following steps:

[0047] Extracting a playback index dimension when the short video is played from the traffic feature set;

[0048] The text modality information and the audio text information when the short video is played are matched according to the playback index dimension to obtain the audience modality information when the short video is played.

[0049] In the specific implementation, first, the key data points that can reflect the user's short video viewing behavior are screened out from the traffic feature set, such as viewing time, stay frequency, skip rate, click-through rate, etc. In order to identify different viewing modes, indicators such as user device type and viewing time period can also be added. Next, the data segmentation technology is used to divide the dimensions of these data to generate "play index dimensions", which will serve as the core reference for audience verification; then, the natural language processing (NLP) algorithm is used to process the text information (such as title, description, subtitle) and audio-to-text information (audio content is converted into text through speech recognition technology) of the short video content, and these text information are matched with the user's playback data based on the playback index dimension. For example, when a user has a high frequency of clicking and watching a certain type of video, the system will analyze the text and audio content of this video to determine the user's preferred theme. For example: if the user frequently watches short videos containing the keyword "fitness" and has a high viewing completion rate, the system can identify the "fitness" theme from the text modal information and verify the user's preference for "fitness" content. Finally, the system summarizes the specific points of interest of different audiences when watching short videos as "audience modal information". This information usually includes dimensions such as the interest categories of the viewing group, preferred content features (such as the user's preferred video type and viewing time period).

[0050] It should be noted that in this application, the playback index dimension represents key data indicators closely related to user playback behavior, which are used to distinguish and compare the interaction patterns of different audience groups when watching videos; the audience modal information represents the preference data of the main interests and behavioral characteristics of the viewing group.

[0051] Preferably, in this embodiment, the click attribute information when the short video user clicks on the short video is determined according to the audience modality information, referring to Figure 2 As described above, the figure is a schematic diagram of a process for determining click attribute information when a short video user clicks on a short video in some embodiments of the present application. In this embodiment, the click attribute information can be determined by the following steps:

[0052] In step S21, the click distribution pattern of short video users when clicking on the short video is determined according to the audience modality information;

[0053] In step S22, the interest anchor point of the short video user is determined according to the click distribution rule;

[0054] In step S23, a push decision information sequence for short video push is generated according to the interest anchor point;

[0055] In step S24, click attribute information of the short video user when clicking on the short video is determined according to the push decision information sequence.

[0056] In the specific implementation, first, by analyzing a large number of user click behaviors, the click distribution statistics of various audience groups on different content types are formed, such as peak click time periods, types of content with high click volume, etc. Then, according to the click distribution law, the core content features that users pay attention to in short videos (such as specific topics, theme types, etc.) are identified, and these features are marked as "interest anchors". Interest anchors reflect users' strong preference for certain content and are the key factors in locking user interests in the recommendation process; then, a push decision information sequence is generated to describe the content type preferred by each user and the appropriate push time period and frequency. This sequence is generated by a push strategy model (such as a recommendation algorithm based on user behavior) and records the content type, time period, push frequency, etc. that the user prefers to receive. For example, for users who like "food" short videos, the system can set a push decision information sequence according to their click rules, such as "push a food short video every lunch period"; finally, the system further determines the user's click attribute information based on the push decision information sequence and the user's actual click behavior. The click attribute information describes the user's click pattern on the short video platform, such as click tendency, content interest type, viewing time period, etc. For example, if a user frequently clicks on the pushed "fitness" short videos and watches them in full, the system will use "fitness content, high-frequency clicks, evening time period" as the user's click attribute information.

[0057] It should be noted that, in the present application, the click distribution law describes the distribution characteristics of user click preferences and is used to analyze user interaction patterns; interest anchors represent special attention paid by users to certain subjects or topics and are used to locate user interest centers and provide content preference directions for the recommendation system; push decision information sequences represent decision sequences that describe user preferred content types and push schedules and provide strategy references for the push system; click attribute information represents a set of preference data that describes user click behaviors on the platform, including content type preferences, click frequency, time periods, etc. and is used to guide the content push strategies of the recommendation system.

[0058] The delivery matching module 300 is used to obtain the access and likes data of the target short video, determine the audience acceptance of the short video content based on the access and likes data, and determine the delivery matching coefficient when delivering the short video through the audience acceptance and the click attribute information.

[0059] In specific implementation, obtaining the access and likes data of the target short video can be achieved in the following manner, namely: first, extracting the access and likes data of the target short video from the platform database, including the number of times the video has been viewed, the frequency of user likes, the time distribution of likes, and other information. Through the data interface or query module, the system aggregates and organizes these real-time updated user interaction data to build a complete access and likes data set for analyzing the user's acceptance and interaction preferences for the short video. In addition, in order to improve the timeliness and accuracy of the data, the system will update and clean these data regularly or in real time to remove duplicate and abnormal data to ensure that the data set reflects the actual user interaction situation. In other embodiments, other methods can also be used to determine the access and likes data, which are not limited here.

[0060] It should be noted that in this application, the visit and like data refers to the collection of user visit and like behavior data for a target short video on the short video platform. It contains information such as the number of video views, total number of likes, like frequency, and time distribution of likes, which is used to reflect the user's interest and acceptance of the video.

[0061] Preferably, in this embodiment, the audience acceptance of the short video content is determined according to the access likes data, referring to Figure 3 The figure is a schematic diagram of a process for determining audience acceptance of short video content in some embodiments of the present application. In this embodiment, determining audience acceptance can be achieved by using the following steps:

[0062] In step S31, multi-dimensional feature data corresponding to the real-time behavior of the short video user is obtained;

[0063] In step S32, the multi-dimensional feature data is extracted to obtain a preference fusion amount corresponding to the real-time behavior of the short video user;

[0064] In step S33, the recommendation discrete quantity when recommending the short video is extracted from the access likes data;

[0065] In step S34, the audience acceptance of the short video content is determined according to the preference fusion amount and the recommendation discrete amount.

[0066] In the specific implementation, first, the user's interactive behavior is monitored in real time through the data acquisition module to obtain multi-dimensional feature data, including the user's playback time, viewing times, like frequency, comments, sharing times, etc. These multi-dimensional features reflect the user's real-time interactive preferences and behavioral characteristics on the video; then, the multi-dimensional feature data is extracted using a data processing algorithm (such as weighted average or feature fusion algorithm), and features such as viewing time, like frequency, and number of comments are integrated to form the user's preference fusion amount on the short video. The preference fusion amount can be used to quantify the user's overall interest in the content. For example, if a user shows a high playback time and multiple likes on the video, these features will be given a higher weight in the calculation, resulting in a higher preference fusion amount; then, the user's likes and viewing behaviors in the recommendation process are screened out from the access likes data, and the data distribution is quantified using statistical analysis to form a recommendation discrete amount. The recommended discrete quantity usually reflects the interaction frequency and distribution characteristics of different user groups when accepting recommended videos; finally, the system compares the preference fusion quantity with the recommended discrete quantity, integrates the numerical results of the two through weight calculation or correlation analysis methods (such as regression analysis), and generates an "audience acceptance" indicator to describe the overall acceptance of short videos by users. For example, when a video performs high in both preference fusion quantity and recommended discrete quantity, the system will mark the audience acceptance of the video as "high", reflecting the user's strong interest and interaction tendency in the content.

[0067] It should be noted that in this application, multidimensional feature data refers to a data set that describes the various behavioral characteristics of users on the short video platform, including viewing time, interaction frequency, etc., which is used to analyze the user's acceptance of the content; the preference fusion quantity is a comprehensive value generated based on the user's multidimensional feature data, which is used to measure the user's interest in a certain video and is one of the core indicators of audience acceptance analysis; the recommendation discrete quantity describes the distribution characteristics of users' recommended short video content, reflecting the differences in acceptance of the content by different groups; audience acceptance is an indicator to measure the user's acceptance and preference for a certain short video.

[0068] In this embodiment, the following steps can be used to determine the delivery matching coefficient when delivering the short video based on the audience acceptance and the click attribute information:

[0069] Determining the preference breadth information of short video users when browsing recommended videos according to the audience acceptance;

[0070] Extracting the preference adoption features when delivering short videos from the click attribute information;

[0071] The delivery matching coefficient when delivering the short video is determined according to the preference breadth information and the preference adoption characteristics.

[0072] In the specific implementation, first, based on the audience acceptance, analyze the user's acceptance range of short videos of different types and themes. The user's preference breadth information can be determined by counting the types and frequency of content that the user watches and interacts with; then, based on the interactive features in the click attribute information, such as click frequency, viewing time, content preference type, etc., extract the user's adoption characteristics for specific content, that is, the user's willingness to receive recommended content. This feature reflects the user's acceptance and behavioral preferences when delivering short video content; finally, combined with the preference breadth information (interest range) and the preference adoption characteristics (acceptance tendency), a delivery matching coefficient is generated through an algorithm model (such as similarity calculation or regression analysis). For example, for a user, if the preference breadth information covers the themes of "travel, food", and the preference adoption characteristics show that the user is more likely to receive food recommendations in the morning, the delivery matching coefficient of pushing food-related short videos in the morning will be higher, which will not be repeated here.

[0073] It should be noted that in this application, preference breadth information describes the user's willingness to accept content on different topics, reflects the range and diversity of user interests, and provides guidance for content delivery strategies; preference adoption features represent the user's willingness to accept recommended content in specific situations and times, and are key data for the recommendation system to optimize content delivery; the delivery matching coefficient represents the user's acceptance of matching specific content, and is used to improve the accuracy and interactive effect of content delivery.

[0074] The correction constraint module 400 is used to determine the behavioral characteristic information of the short video user when operating the short video, and determine the perceived weirdness of the short video user when watching the short video according to the behavioral characteristic information; and correct the user behavior sequence of the short video user through the perceived weirdness.

[0075] In this real-time example, determining the behavioral feature information of the short video user when operating the short video can be achieved by using the following steps:

[0076] Determine the data sample points when the short video user operates the short video;

[0077] Determine the density of operation behaviors of short video users when operating short videos;

[0078] Behavior feature information of a short video user when operating a short video is determined according to the data sample points and the operation behavior density.

[0079] In the specific implementation, first, record each operation performed by the user while watching the short video, such as play, pause, like, comment, share, etc. The time point when these operations occur and the associated operation types are recorded as data sample points, which constitute the basic data describing the user's operation; then, based on the data sample points, count the number of operations performed by the user per unit time to form the operation behavior density, reflecting the frequency and intensity of user interaction. A high operation behavior density indicates that the user frequently performs interactive operations in a short period of time, which may indicate a high interest in the video content; finally, combine the data sample points and the operation behavior density to model the behavioral characteristic information of the user during the video playback. The behavioral characteristic information quantifies the user's operation preference for short videos, such as whether there is a habit of high-frequency likes or comments, whether frequent interactions in the early stage of watching, etc. For example, for a user with a high operation behavior density and a high frequency of likes, his behavioral characteristic information will be expressed as "high frequency likes, active interaction", which is helpful for the subsequent push of similar popular content, which will not be repeated here.

[0080] It should be noted that in the present application, data sample points represent specific operations and their time point records that occur during the user's viewing of short videos, which constitute the basic data of user interactive behavior; operation behavior density represents the frequency of user operations per unit time, which is used to measure the intensity of user interaction on short video content; behavioral feature information represents comprehensive information describing user operation preferences based on data sample points and operation behavior density, which is used to identify users' interaction patterns and preferences for content.

[0081] In this embodiment, determining the perceived weirdness of a short video user when watching a short video according to the behavior characteristic information can be achieved by using the following steps:

[0082] Determine the perception change nodes of short video users when watching short videos;

[0083] Performing multi-dimensional analysis on the behavior feature information through the perception change node to obtain a behavior proximity distance;

[0084] The perceived weirdness of a short video user when watching the short video is determined according to the behavior proximity distance.

[0085] In the specific implementation, the user's perception change nodes are first identified by detecting the key interactive changes (such as likes, pauses, jumps, comments, etc.) shown by the user during the process of watching short videos. These nodes reflect the different reaction points of users to the video content, usually concentrated in the places where the video emotions suddenly change or the plot changes; then, the user's behavioral feature information is multi-dimensionally analyzed through the perception change nodes, and the behavioral similarities or differences between each node are calculated to form the behavioral proximity distance, which can be calculated by algorithms such as Euclidean distance or cosine similarity; finally, the behavioral proximity distance between multiple perception change nodes is calculated. If the proximity distance is large, it indicates that the user has shown more abnormal behavioral changes during the viewing process, which is the perceived weirdness. The perceived weirdness quantifies the abnormality of the user's reaction to the video content. The higher the value, the more incoherent or abnormal the user's behavior is. For example, if the user shows jump playback behavior when watching a funny video, or suddenly reduces interaction in the tense part of the plot, the system will determine that the user has a high perceived weirdness based on the larger behavioral proximity distance.

[0086] It should be noted that in the present application, the perception change node refers to the time node when the user shows significant interactive behavior due to content changes during the short video viewing process, which is usually the key point of user perception change or emotional fluctuation; the behavioral proximity distance describes the similarity of the user's behavior at different nodes, which is used to identify the consistency of the user's behavior during the video viewing process; the perceived weirdness indicates the degree of behavioral consistency and emotional coherence of the user when watching short videos, which is used to identify the user's abnormal interaction pattern.

[0087] In specific implementation, the correction constraint of the user behavior sequence of short video users by the perceived weirdness can be implemented in the following manner, namely: first, the system marks the time intervals with high perceived weirdness according to the user's behavior sequence when watching short videos and the calculated perceived weirdness. These intervals represent the time periods when users show abnormal behavior in video playback (for example, frequent jumps or repeated operations of liking and pausing, etc.); then, in the high perceived weirdness interval, the system determines the reason for the user's reaction by analyzing the motivation of the user's behavior (such as specific content is not suitable or the plot is boring). For example, if the user jumps frequently, it may be because of lack of interest in the content or misunderstanding of the video plot. At this time, the system will automatically reduce the recommendation priority of similar content for the user. When recording and learning user behavior, the system constrains abnormal behavior and filters out behaviors that may mislead recommendations to avoid excessive influence on subsequent recommendations. For example, in future recommendations, reduce the push frequency of similar content, or adjust the push timing; finally, after executing the correction constraint, the system will update the user's behavior sequence data model to remove or reduce the impact of high perceived weirdness behavior so that it does not interfere with normal interest tag extraction. The model will gradually learn and adapt to the user's true preferences without being disturbed by individual abnormal behaviors. For example, if a user performs a large number of abnormal operations (such as frequent jumps) when watching a short video, the system will downplay the impact of this part of the behavior in future recommendation models to ensure that the recommendation results will not be completely changed due to a single abnormal behavior. I will not go into details here.

[0088] It should be noted that in this application, the user behavior sequence is a record of the continuous interactive behavior of the user when watching a short video, arranged in chronological order, reflecting the user's changes in interest, focus and interaction pattern in the video content.

[0089] The video recommendation module 500 is used to perform trend analysis on the dynamic interest tags of short video users according to the delivery matching coefficient and the user behavior sequence after the correction constraints, so as to recommend short videos to the short video users.

[0090] In the specific implementation, the user's content acceptance and matching are first determined. The delivery matching coefficient is an indicator to measure the degree of fit between content and user interests. The higher the value, the higher the fit between content and user interests. The system combines this coefficient with the user's corrected and constrained behavior sequence to screen out content that has a high degree of match with the user as the basis for interest tag parsing; then extract the user's core interactive behavior from the corrected and constrained user behavior sequence, and identify the content tags corresponding to the behavior, such as "funny", "popular science", "music", etc., as the user's interest tags; then the system tracks the changing trend of the user's interest tags, and monitors the dynamic changes of user preferences through time series analysis or sliding window methods. For example, when the system observes that the weight of the "music" tag gradually increases and the weight of the "popular science" tag decreases, it determines that the user's interest may be shifting towards music content; finally, based on the latest dynamic interest tag trends, the short video content that highly matches the current interest tag is included in the recommendation pool, and is sorted and prioritized according to the delivery matching coefficient to ensure that the recommended content is in line with the user's current interests. For example, for users whose current interest tag is "funny" and the trend is stable, the system gives priority to pushing funny videos, and further refines the matching in content format and theme, which will not be elaborated here.

[0091] It can be seen that in this application, matrix analysis and management of multi-dimensional short videos are achieved under the influence of the difficulty in accurately capturing the dynamic interests and complex preferences of users from the recommended content; among them, by building a comprehensive set of traffic characteristics, the system can accurately capture and analyze key data such as users' viewing behavior and interaction habits, provide data support for subsequent personalized recommendations, and improve the accuracy and timeliness of recommendations; audience verification ensures in-depth analysis of different user groups, and can more accurately identify the preferences and needs of each user, so that content recommendations are closer to users' interests, avoid the push of irrelevant content, and improve user participation and satisfaction; by analyzing user click attributes Through precise analysis, the system can gain an in-depth understanding of users’ specific behavioral preferences, such as video type and duration, so as to more effectively match user needs when making recommendations, optimize click-through rate and viewing time, and improve user experience; by analyzing likes data, the system can understand the audience feedback and popularity of short videos in real time, so as to optimize content distribution strategies, enhance the attractiveness of platform content, and increase user activity and platform stickiness; through trend analysis and continuous updating of dynamic interest tags, the system can make real-time adjustments based on changes in user interests, ensure that recommended content always matches the user’s current interests, improve the personalization and accuracy of recommendations, and enhance the user’s long-term stickiness and platform activity.

[0092] To sum up, the technical solution adopted in this application can improve user experience and platform activity when it is difficult to accurately capture users' dynamic interests and complex preferences from recommended content.

[0093] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0094] A person skilled in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, the storage medium including a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically-erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0095] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

Claims

1. AI-driven multi-dimensional short video matrix analysis and management system, characterized by: The analysis and management system includes: A model building module, used to establish a traffic feature set when playing short videos using an AI-driven big data structure, where the traffic feature set represents a structured data set of various types of interactive data generated by users during the short video playback process; An attribute determination module is used to perform audience verification on the traffic feature set to obtain audience modality information when the short video is played, and determine click attribute information when the short video user clicks on the short video based on the audience modality information, wherein the audience modality information represents preference data of interests and behavioral characteristics of the viewing group; A delivery matching module is used to obtain the access and likes data of the target short video, determine the audience acceptance of the short video content according to the access and likes data, and determine the delivery matching coefficient when delivering the short video according to the audience acceptance and the click attribute information; The correction constraint module is used to determine the behavior characteristic information of the short video user when operating the short video, and determine the perceived weirdness of the short video user when watching the short video according to the behavior characteristic information; correct the user behavior sequence of the short video user by using the perceived weirdness, wherein determining the perceived weirdness of the short video user when watching the short video according to the behavior characteristic information specifically includes: Determine the perception change nodes of short video users when watching short videos; Performing multi-dimensional analysis on the behavior feature information through the perception change node to obtain a behavior proximity distance; Determining the perceived weirdness of the short video user when watching the short video according to the behavior proximity distance; The video recommendation module is used to perform trend analysis on the dynamic interest tags of short video users according to the delivery matching coefficient and the user behavior sequence after the correction constraints, so as to recommend short videos to the short video users.

2. The AI-driven multi-dimensional short video matrix analysis and management system according to claim 1, characterized in that: Audience verification of the traffic feature set to obtain audience modality information when the short video is played specifically includes: Extracting a playback index dimension when the short video is played from the traffic feature set; The text modality information and the audio text information when the short video is played are matched according to the playback index dimension to obtain the audience modality information when the short video is played.

3. The AI-driven multi-dimensional short video matrix analysis and management system according to claim 1, characterized in that: Audience acceptance is an indicator that measures the degree of users' acceptance and preference for a short video.

4. The AI-driven multi-dimensional short video matrix analysis and management system according to claim 1, characterized in that: Determining the delivery matching coefficient when delivering a short video by using the audience acceptance and the click attribute information specifically includes: Determining the preference breadth information of short video users when browsing recommended videos according to the audience acceptance; Extracting the preference adoption features when delivering short videos from the click attribute information; The delivery matching coefficient when delivering the short video is determined according to the preference breadth information and the preference adoption characteristics.

5. The AI-driven multi-dimensional short video matrix analysis and management system according to claim 1, characterized in that: The behavioral characteristic information of the short video user when operating the short video specifically includes: Determine the data sample points when the short video user operates the short video; Determine the density of operation behaviors of short video users when operating short videos; Behavior feature information of a short video user when operating a short video is determined according to the data sample points and the operation behavior density.

6. The AI-driven multi-dimensional short video matrix analysis and management system according to claim 1, characterized in that: Perceived weirdness indicates the degree of behavioral consistency and emotional coherence of users when watching short videos.

7. The AI-driven multi-dimensional short video matrix analysis and management system according to claim 1, characterized in that: User behavior sequence is a record of the continuous interactive behaviors of users when watching short videos.

Images