Account recommendation system based on user behaviors

Abstract

The invention relates to the technical field of account recommendation, in particular to an account recommendation system based on user behaviors, which comprises an account content screening module, a video hotspot analysis module, a user behavior analysis module, a behavior portrait module and a customization recommendation module. According to the method, accurate recognition and trend prediction of user behavior modes are realized by deeply analyzing fine-grained records of user behavior data and timestamps, interest points of the users can be quickly recognized, and activity and preference of the users are distinguished by utilizing behavior tags through contrastive analysis of group behavior data and construction of detailed user behavior portraits. The recommendation system can more accurately match user requirements, thereby optimizing network experience, improving viscidity of the user to the platform, enhancing relevance and individuation level of account recommendation, improving user satisfaction, and remarkably promoting the degree that the user deeply participates in platform content and interaction.

Description

Technical Field

[0001] The present invention relates to the technical field of account recommendation, and in particular to an account recommendation system based on user behavior. Background Art

[0002] The field of account recommendation technology involves using machine learning, data mining and artificial intelligence algorithms to analyze user data and provide personalized user experience based on this. The core goal is to recommend the most suitable account or content to users by understanding their historical behaviors, preferences and interactions. It usually involves collecting user behavior data, such as browsing history, purchase records and search queries, and then applying algorithm models to predict new accounts or services that users may be interested in. It is widely used in social media, e-commerce, online advertising and content streaming platforms to enhance user engagement and improve the personalization level of services.

[0003] Among them, the account recommendation system based on user behavior refers to a system that uses the user's online behavior data to recommend accounts. The system automatically identifies and recommends accounts that the user may be interested in or related to by analyzing the user's interactions, interests and network behavior patterns. Its main purpose is to improve user satisfaction and user stickiness of the platform, optimize the user's network experience by recommending highly relevant accounts, and promote users to participate more deeply in the platform's content and interaction. It is particularly common in social networking services, news aggregation and online entertainment, and helps users discover new accounts or content that matches their interests, thereby improving the overall user experience.

[0004] Existing technologies usually rely on shallow user behavior analysis, such as browsing history and search queries, which limits the in-depth understanding of users' true preferences and results in recommendations that do not match users' actual needs. In addition, the algorithm models in existing methods often fail to fully utilize time series data to predict user behavior, and are unable to cope with dynamically changing user behavior patterns. For example, unrecognized changes in behavior patterns may cause users to receive outdated or irrelevant recommendations, affecting user experience and engagement. This is particularly evident on rapidly changing platforms such as social media and online entertainment, which can easily lead to user churn and reduce the attractiveness of the platform. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an account recommendation system based on user behavior.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical solution: An account recommendation system based on user behavior comprises:

[0007] The account content screening module matches and screens the content tags, industry classifications, video release times and account attributes of the account data according to the query parameters entered by the user, verifies the data matching degree item by item, accumulates the matching results, and outputs the account screening results;

[0008] The video hotspot analysis module receives the account screening results, collects corresponding video data, extracts video titles and cooperative categories, identifies keywords and calculates keyword frequencies through statistical analysis, determines hot topics, and generates video hotspot analysis data;

[0009] The user behavior analysis module collects user interaction data based on the video hotspot analysis data, records user activities using timestamps, filters the data, identifies the interaction time period and content type, partitions the data according to user groups based on the identification results, performs statistical analysis on the user behavior data in each partition, identifies key behavior patterns, and generates user behavior analysis results;

[0010] The behavior profiling module compares the behavior patterns of different user groups based on the user behavior analysis results, uses behavior tags to distinguish differences in user activity and preferences, and compares the behavior data of each user with the group average data based on the distinction results, thereby revealing the behavior characteristics of each user group and constructing a group user behavior profile;

[0011] The customized recommendation module performs correlation evaluation based on the user behavior profile of the group, analyzes user preferences and behavior cycles, and adjusts the recommended content in real time to generate a dynamic recommendation list.

[0012] As a further solution of the present invention, the steps of obtaining the account screening result are:

[0013] According to the query parameters input by the user, the content tags, industry classifications, video release time and account attributes of the account data are standardized, a standardized matrix of various parameters is established, each parameter is normalized, and the weight vector of each parameter is calculated based on the query parameters to generate a standardized parameter matrix and a weight vector set;

[0014] Based on the standardized parameter matrix and weight vector set, the formula is adopted:

[0015]

[0016] Calculate the comprehensive matching degree M of the i-th account in terms of content tags, industry classification, video release time and account attributes i , and construct the account matching matrix, where w j represents the weight of the jth parameter, P ij represents the standardized value of the i-th account on the j-th parameter, Q ijrepresents the normalized value of the query parameter at the jth parameter, and n represents the number of parameters;

[0017] Based on the account matching matrix, the matching results are cumulatively calculated, the matching degrees of all accounts are sorted, a matching threshold is set to filter out a set of accounts that meet the requirements, and a comprehensive analysis is performed on the account's content tags, industry classifications, video release times, and account attributes to output the account screening results.

[0018] As a further solution of the present invention, the step of obtaining the video hotspot analysis data is:

[0019] According to the account screening results, the corresponding video data is collected, the video titles and cooperative category contents are extracted, special characters, punctuation marks and stop words are removed, the text content is standardized, the text units are split according to word boundaries, and the number of occurrences of all words is counted, and the video text data set is established in combination with the cooperative category information;

[0020] Based on the video text data set, correlation analysis is performed in combination with the cooperative categories, screening conditions are set and keywords are determined, using the formula:

[0021]

[0022] Calculate the relative frequency K of the keywords and get the keyword frequency distribution result, where f i represents the number of occurrences of the ith keyword, Z represents the total number of occurrences of all keywords in the video title, and N represents the total number of keywords;

[0023] Based on the keyword frequency distribution results, a screening threshold is set to determine high-frequency keywords, the hot topic coverage is calculated according to the category to which the keyword belongs, the proportion of hot topics is calculated based on the number of videos, and video hot topic analysis data is generated.

[0024] As a further solution of the present invention, the step of identifying the interactive time period and content type is:

[0025] Based on the video hotspot analysis data, user interaction data is collected, user behaviors during video playback are recorded, timestamps of user interactions are extracted, user interaction records are associated with video IDs, and all user interaction behaviors are arranged in chronological order to generate a user interaction time series;

[0026] Based on the user interaction time series, repeated interaction records are removed, a time window is set for data smoothing, and the user interactions are classified according to the time period, using the formula:

[0027]

[0028] Calculate the interaction concentration C within the time period and obtain the distribution record of the interaction time period, where A i represents the frequency of the i-th user interaction behavior, U represents the number of users who interacted in this time period, and M represents the total number of user interaction behaviors;

[0029] Based on the interaction time period distribution records, cluster analysis is performed according to the time periods when the interaction behaviors occur, and high interaction time intervals are extracted. The user interaction behaviors are associated with the video content types, and the interaction proportions of different content types in each time period are statistically analyzed. The interaction concentration of the target content type is analyzed to obtain the interaction time period and content type identification results.

[0030] As a further solution of the present invention, the steps of obtaining the user behavior analysis results are:

[0031] Based on the interaction time period and content type identification results, users are classified according to their interaction behavior characteristics, and the interaction frequency, interaction method and content preference of users in different time periods are counted, and a partitioning standard is simultaneously established to generate a user group partitioning result;

[0032] Based on the user group partition results, analyze the interactive behavior patterns of users in each partition, count the frequency of each interactive behavior in the partition, calculate the proportion of each behavior category in different time periods, compare the behavioral feature differences between different partitions, identify the key interactive trends and behavioral changes of each partition, and obtain the user behavior trend analysis results in the partition;

[0033] Based on the user behavior trend analysis results within the partition, key behavior patterns are identified, key interaction types of users in the target time period are extracted, the stability of user behavior and changes in interaction methods over time are analyzed, the behavior concentration of users in different partitions is compared, and combined with the overall changes in user group behavior patterns, user behavior analysis results are generated.

[0034] As a further solution of the present invention, the steps of distinguishing the user activity and preference differences are:

[0035] According to the user behavior analysis results, statistically analyze the interaction frequency, behavior stability, and content preference data of different user groups, compare the key behavior characteristics between groups, calculate the distribution differences of different user groups in each behavior category, and obtain the behavior pattern comparison results of different user groups;

[0036] Based on the comparison results of the behavioral patterns of the different user groups, the core behavioral characteristics of each group are extracted, and the activity level is set in combination with the interaction frequency and time characteristics, using the formula:

[0037]

[0038] Calculate the user's activity S and obtain the classification results of user activity and preference differences, where I i represents the number of interactions of the user in the i-th time period, U represents the total number of users, and m represents the total number of time periods;

[0039] Based on the classification results of the user activity and preference differences, use behavioral labels to mark the user's interactive activity and content tendencies, set label classification standards, classify users according to behavioral characteristics, reflect the key behavioral characteristics of users under different categories, and generate user behavior labels.

[0040] As a further solution of the present invention, the steps of obtaining the group user behavior portrait are:

[0041] According to the user behavior tags, the behavior average of each group is calculated, the common behavior characteristics of users within the group are determined, and the group average behavior data is established;

[0042] Based on the group average behavior data, the individual behavior data of each user is compared with the average data of the group to which they belong, the deviation of the individual behavior data relative to the group mean is calculated, the consistency and difference between the individual behavior and the group behavior are analyzed, and the user group behavior characteristics are output;

[0043] Based on the behavioral characteristics of the user groups, the key behavioral trends of different groups are summarized, and combined with the users' interactive activity, time distribution and content preferences, the core interaction patterns and characteristics of the group users are reflected to construct a behavioral portrait of the group users.

[0044] As a further solution of the present invention, the step of obtaining the dynamic recommendation list is:

[0045] Based on the user behavior portrait of the group, collect the user interaction frequency, content browsing time, and the number of times the target behavior occurs, calculate the behavior distribution within the different user groups, and normalize the characteristic values ​​of each group to obtain user behavior distribution data;

[0046] According to the user behavior distribution data, the behavior correlation between user groups is analyzed, matching is performed based on the similarity of feature values, and the degree of difference in behavior patterns is calculated, a user group behavior pattern comparison matrix is ​​established, groups are classified based on the degree of matching, and user group behavior pattern classification results are generated;

[0047] Based on the classification results of the user group behavior patterns, the periodic changes of the behavior patterns within each group are analyzed, the distribution of key behavior patterns in different time periods is extracted, the changes in user demand are predicted according to the behavior change trends, the recommended content is adjusted, and a dynamic recommended content list is established.

[0048] Compared with the prior art, the advantages and positive effects of the present invention are:

[0049] In the present invention, by deeply analyzing the fine-grained records of user behavior data and timestamps, accurate identification and trend prediction of user behavior patterns are achieved, and user interests can be quickly identified. Through comparative analysis of group behavior data, detailed user behavior portraits are constructed, and behavior tags are used to distinguish user activity and preferences. The recommendation system can more accurately match user needs, thereby optimizing the network experience and improving user stickiness to the platform, enhancing the relevance and personalization level of account recommendations, improving user satisfaction, and significantly promoting the degree of user in-depth participation in platform content and interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a system flow chart of the present invention;

[0051] Figure 2 A flowchart for obtaining account screening results of the present invention;

[0052] Figure 3 It is a flow chart for obtaining video hotspot analysis data of the present invention;

[0053] Figure 4 A flow chart for identifying interactive time periods and content types of the present invention;

[0054] Figure 5 A flowchart for obtaining the user behavior analysis results of the present invention;

[0055] Figure 6 A flow chart for distinguishing user activity and preference differences of the present invention;

[0056] Figure 7 A flowchart for obtaining a group user behavior portrait of the present invention;

[0057] Figure 8 This is a flow chart of obtaining a dynamic recommendation list according to the present invention. DETAILED DESCRIPTION

[0058] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0059] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0060] See also Figure 1 , an account recommendation system based on user behavior includes:

[0061] The account content screening module matches and screens the content tags, industry classifications, video release times and account attributes of the account data according to the query parameters entered by the user, verifies the data matching degree item by item, accumulates the matching results, and outputs the account screening results;

[0062] The video hotspot analysis module receives the account screening results, collects the corresponding video data, extracts the video title and cooperative category, identifies keywords through statistical analysis and calculates the keyword frequency, determines the hot topics, and generates video hotspot analysis data;

[0063] The user behavior analysis module collects user interaction data based on video hotspot analysis data, uses timestamps to record user activities, filters the data, identifies the interaction time period and content type, partitions the data according to user groups based on the identification results, performs statistical analysis on the user behavior data in each partition, identifies key behavior patterns, and generates user behavior analysis results;

[0064] The behavior portrait module compares the behavior patterns of different user groups based on the results of user behavior analysis, uses behavior tags to distinguish differences in user activity and preferences, and compares each user's behavior data with the group average data based on the distinction results, revealing the behavior characteristics of each user group and building a group user behavior portrait;

[0065] The customized recommendation module conducts correlation assessment based on group user behavior portraits, analyzes user preferences and behavior cycles, and adjusts recommended content in real time to generate a dynamic recommendation list.

[0066] The account screening results include account attribute matching results, content tag matching results, industry classification matching results and video time matching results. The video hot spot analysis data includes keyword frequency, hot topic identification records, video title data and cooperation category data. The user behavior analysis results include user interaction data statistics, activity time records, interactive content analysis records and user group partitioning results. The group user behavior portrait includes behavior tag classification, activity analysis records, preference difference comparison records and behavior feature disclosure results. The dynamic recommendation list includes recommended content adjustment results, user preference analysis results and behavior cycle evaluation records.

[0067] See also Figure 2 , the steps to obtain the account screening results are:

[0068] According to the query parameters input by the user, the content tags, industry classifications, video release time and account attributes of the account data are standardized, a standardized matrix of various parameters is established, each parameter is normalized, and the weight vector of each parameter is calculated based on the query parameters to generate a standardized parameter matrix and a weight vector set;

[0069] Call the query parameters entered by the user, standardize the content tags, industry classifications, video release time and account attributes of the account data, extract the original values ​​of all account data, and calculate the maximum and minimum values ​​of each parameter respectively. Use the minimum-maximum normalization method to perform standardization conversion to ensure that all parameters are mapped to the same numerical range. During the standardization calculation, read the parameter values ​​of each account in turn, calculate the minimum and maximum values ​​of the current parameters, and follow the formula A normalization transformation is performed to ensure that the data will not affect subsequent calculations due to differences in magnitude. After the calculation is completed, the standardized parameter values ​​are stored in the parameter matrix. At the same time, the query parameters entered by the user are also standardized to ensure that the query parameters and account data are in the same numerical range. Then, the mapping values ​​of the query parameters in the standardized matrix are calculated. Based on the keywords, options, and filter conditions entered by the user, the core elements of user concern are analyzed, the priority of the query parameters is determined, and the weight vector is calculated according to the distribution of the query parameters. When calculating the weight, a parameter importance comparison matrix is ​​first constructed, and the importance of each parameter is compared pairwise. The eigenvalues ​​of the parameter comparison matrix are calculated, and the weight vector is obtained after normalization to ensure that the parameter weights can objectively reflect the focus of user needs. Finally, a standardized parameter matrix and weight vector set are generated.

[0070] Based on the standardized parameter matrix and weight vector set, the formula is adopted:

[0071]

[0072] Calculate the comprehensive matching degree M of the i-th account in terms of content tags, industry classification, video release time and account attributes i , and construct the account matching matrix, where w j represents the weight of the jth parameter, P ij represents the standardized value of the i-th account on the j-th parameter, Q ij represents the normalized value of the query parameter at the jth parameter, and n represents the number of parameters;

[0073] w j : The weight of the jth parameter is determined by the analytic hierarchy process (AHP).

[0074] P ij : The standardized value of the i-th account at the j-th parameter is calculated as follows:

[0075]

[0076] Among them, X ij is the original value of the i-th account at the j-th parameter, X min,j and X max,j are the minimum and maximum values ​​of the j-th parameter, respectively.

[0077] Q ij : The normalized value of the query parameter at the jth parameter, calculated in the same way as above.

[0078] Weight (calculated based on AHP):

[0079] w 1 =0.4, w 2 =0.3, w 3 =0.2, w 4 =0.1;

[0080] Standardized account data (a certain account):

[0081] P i1 =0.7, P i2 =0.5, P i3 =0.6, P i4 =0.8;

[0082] Normalized query parameter data:

[0083] Q i1 =0.6, Q i2 =0.5, Q i3 =0.7, Q i4 =0.9;

[0084] Compute the weighted sum of absolute differences:

[0085]

[0086] Calculate the denominator, which is the sum of the weights:

[0087]

[0088] Calculate the final matching degree:

[0089]

[0090] The results show that the match between the first account and the user's query parameters is 0.93, indicating that the account is highly matched with the query requirements and is close to a perfect match (1 indicates a perfect match, and 0 indicates a complete mismatch). If the match is lower than the set threshold (such as 0.8), the data needs to be re-screened.

[0091] Based on the account matching matrix, the matching results are accumulated and calculated, the matching degrees of all accounts are sorted, and the matching threshold is set to filter the account set that meets the requirements. The account content tags, industry classifications, video release time and account attributes are comprehensively analyzed to output the account screening results.

[0092] Based on the account matching matrix, the matching values ​​of each account are extracted and sorted. When sorting the matching values, they are arranged in order from high to low, and the matching values ​​corresponding to each account are stored. A matching screening threshold is set, and accounts that meet the requirements are screened according to the threshold. Accounts with matching values ​​higher than the set threshold are extracted to form a set of accounts to be screened. On this basis, content analysis is performed on the screened account set. When analyzing, the content label categories of the screened accounts are first counted, the main content labels of each account are extracted, and the frequency of occurrence of each label is calculated. Based on the frequency distribution of the labels, the content concentration is calculated, and the aggregation of the screened accounts in specific content categories is analyzed. Then, the industry classification is analyzed. , extract industry category information, calculate the account proportion of each industry category, determine the mainstream distribution trend of the industry to which the filtered accounts belong, further analyze the distribution of video release time, extract the release time data of each account, calculate the release time frequency in different time periods, draw the time distribution curve of the release time, analyze the active time period of the filtered accounts, then statistically organize the account attribute data, extract the main characteristic values ​​of the account, such as the number of fans, interaction rate, playback volume and other data, classify and calculate each data, calculate statistical indicators such as mean, median, standard deviation, and finally form the overall distribution characteristics of the filtered accounts, combine all analysis results, and output the final account filtering results.

[0093] See also Figure 3 ,The steps to obtain video hotspot analysis data are:

[0094] According to the account screening results, collect the corresponding video data, extract the video titles and the content of the cooperation categories, remove special characters, punctuation marks and stop words, perform standardized conversion on the text content, split the text units according to word boundaries, and count the occurrence times of all words. Combine the cooperation category information to establish a video text data set;

[0095] Call the account screening results, collect the corresponding video data, extract the video titles and the content of the cooperation categories, read the text information of all videos, convert all characters into a unified format to ensure the consistency of text processing. First, remove all punctuation marks, special characters, HTML tags and other non-text elements, and convert all letters to lowercase to eliminate the impact of case on keyword statistics. Then perform stop word filtering, removing high-frequency but semantically meaningless words from the text data, such as common auxiliary words and conjunctions like "de", "shi", "he", etc., to avoid meaningless words interfering with keyword extraction. Next, split the text according to word boundaries in natural language. During the word segmentation process, it is necessary to ensure that each word can form a single word independently and avoid incorrect splitting or merging of words due to hyphens or abbreviations. After completing the basic text processing, construct a vocabulary mapping table to number the words in all video titles, enabling subsequent data processing to perform fast matching and statistics based on indexes. Then, perform data screening on the content of the cooperation categories, extract the category information that has a direct relevance to the video titles, and剔除 descriptive or less relevant category labels to avoid irrelevant categories interfering with data analysis. At the same time, analyze the matching degree between the categories and the video content. If the occurrence times of a certain category label are much lower than the number of title keywords, it may indicate that the category information is incomplete and the data collection scope needs to be further adjusted to ensure the consistency of the data dimensions of the video titles and the cooperation categories. Finally, establish a video text data set and save the structured data containing video titles, keywords, and category labels.

[0096] Based on the video text data set, conduct a relevance analysis in combination with the cooperation categories, set screening conditions and determine keywords, using the formula:

[0097]

[0098] Calculate the relative occurrence frequency K of the keywords to obtain the keyword frequency distribution result. Among them, f i represents the occurrence times of the i-th keyword, Z represents the total occurrence times of all keywords in the video titles, and N represents the total number of keywords;

[0099] The occurrence times of the keyword f i : Traverse all video titles, count the occurrence times of each keyword, and ensure that the cumulative calculation of the same words is not missed;

[0100] Total occurrences Z: Calculate the total frequency of all keywords to measure the overall frequency distribution and ensure that the calculation results of different keywords are comparable. Keyword statistics results:

[0101] Short videos = 25, product promotion = 30, promotion = 20, live streaming = 40;

[0102] Total keyword occurrences:

[0103] Z = 25 + 30 + 20 + 40 = 115;

[0104] Calculate the relative frequency of "short videos":

[0105]

[0106] Calculate the relative frequency of "carrying goods":

[0107]

[0108] Calculate the relative frequency of "planting grass":

[0109]

[0110] Calculate the relative frequency of "live":

[0111]

[0112] The results show that the keyword "live broadcast" has the highest relative frequency, accounting for 34.8%, followed by "bringing goods", while "planting grass" has the lowest relative frequency, only 17.4%, indicating that the keyword is used less frequently in video titles. The calculation results can be used to screen high-frequency keywords for subsequent hot topic analysis.

[0113] Based on the results of keyword frequency distribution, set the screening threshold to determine the high-frequency keywords, calculate the coverage of hot topics according to the categories to which the keywords belong, calculate the proportion of hot topics based on the number of videos, and generate video hotspot analysis data;

[0114] Based on the results of keyword frequency distribution, set the screening threshold to screen out keywords with higher frequency. First, sort the keywords according to their relative frequency, determine the keywords that appear most frequently in all video titles, and select the top several high-frequency words as analysis targets. In order to ensure that the screened keywords can truly reflect the hot content of the video, it is necessary to further calculate the distribution of keywords. First, classify and count the keywords according to their category affiliation, extract the keywords with the highest frequency in each cooperative category, and calculate the keyword share of each category to measure the industry distribution of hot keywords. On this basis, analyze the distribution of hot topics in different cooperative categories, group high-frequency keywords by category, and calculate the concentration of keywords in each category. If the keywords in a certain category are relatively concentrated, it means that the product is relatively concentrated. The hot spots of a category are relatively fixed, otherwise it means that the hot spots of this category are relatively scattered. In order to ensure that the calculation of hot topics is based on the actual appearance ratio of keywords, it is necessary to cross-analyze the matching relationship between cooperative categories and keywords, screen out keywords with too low a proportion or too small a coverage, and ensure that the coverage of hot topics is wide enough. Finally, based on the statistical weight of keywords and the number of videos, the influence range of hot topics is calculated, and the distribution of hot topics is sorted out. When calculating the influence range of hot topics, the extended influence index of keywords is calculated according to the distribution of keywords in the video to measure whether the keyword can cover enough video content and ensure that the hot topics are not limited to individual videos, but have greater dissemination. Finally, the industry distribution of keywords, the proportion of hot topics and the influence range are sorted out, and the video hot spot analysis data is output.

[0115] See also Figure 4 ,The steps for identifying the interaction time period and content type are:

[0116] Based on the video hotspot analysis data, collect user interaction data, record user behavior during video playback, extract the timestamp of user interaction, associate user interaction records according to video ID, arrange all user interaction behaviors in chronological order, and generate a user interaction time series;

[0117] Based on the video hotspot analysis data, user interaction data is collected, user activities are recorded using timestamps, and the data is screened to identify the interaction time periods and content types. First, the video hotspot analysis data is obtained to extract the user's interactive behavior data during video playback, including likes, comments, shares and other operations. An accurate timestamp is added to each interactive behavior to record the specific activity time of the user during video playback. Next, the collected user interaction data is screened to remove abnormal or invalid data points to ensure the accuracy and reliability of the data. Then, the time distribution of user interaction behavior is analyzed to identify the time periods when users are active and determine the peak periods of concentrated user interaction. At the same time, the content types of user interactions are classified and sorted, and different interaction methods are distinguished, such as comment content, like objects, sharing channels, etc. Through the above steps, the user's interactive time periods and content types in video hotspots can be fully grasped to provide a basis for subsequent analysis.

[0118] Based on the user interaction time series, duplicate interaction records are removed, a time window is set for data smoothing, and the data is classified according to the time period in which the user interaction occurs, using the formula:

[0119]

[0120] Calculate the interaction concentration C within the time period and obtain the distribution record of the interaction time period, where A i represents the frequency of the i-th user interaction behavior, U represents the number of users who interacted in this time period, and M represents the total number of user interaction behaviors;

[0121] User interaction behavior frequency A i :Extract the number of user interactions in a certain time period from the user interaction time series, and calculate the interaction behavior frequency of all users in this time period;

[0122] Number of users U: Count the number of users who interacted during the time period, excluding users who did not interact.

[0123] Set the user interaction frequency within a certain time period:

[0124] A={5, 8, 12, 4, 6, 10, 15, 7, 9, 11}

[0125] Compute the sum of squares:

[0126]

[0127] Calculate the interaction concentration:

[0128]

[0129] The results show that the interaction concentration during this time period is 2.94, which means that the users' interaction behaviors during this time period are relatively concentrated. A higher concentration indicates that the interaction behaviors of some users during this time period are much higher than those of other users, while a lower concentration indicates that the interaction behaviors during this time period are more evenly distributed. The calculation results can be used to identify the active time periods of users and provide data support for further analysis of interaction behavior trends.

[0130] Based on the interactive time period distribution records, cluster analysis is performed according to the time period when the interactive behavior occurs, and the high interactive time interval is extracted. The user interactive behavior is associated with the video content type, and the interactive proportion of different content types in each time period is counted. The interactive concentration of the target content type is analyzed to obtain the interactive time period and content type identification results;

[0131] Based on the distribution records of interaction time periods, a clustering algorithm is used to classify the interaction behaviors in different time periods. K-means or DBSCAN algorithms are used to analyze the concentration trend of interaction behaviors. First, the interaction concentration data of each time period is normalized to ensure that the data of different time windows are comparable. The elbow method is used to determine the optimal number of clusters, and the K-means algorithm is used for clustering. The time periods with high interaction behavior and low interaction behavior are divided. The high interaction behavior time intervals are further analyzed, and the intervals are associated with the video content type. The video content labels are obtained by querying the user interaction logs, and the interaction proportions of different content types in each time period are calculated. Finally, the interaction distribution of the target content type in different time periods is obtained, and the interaction time period and content type recognition results are output.

[0132] See also Figure 5 , the steps to obtain the user behavior analysis results are:

[0133] Based on the interaction time period and content type identification results, users are classified according to their interaction behavior characteristics, and the interaction frequency, interaction method and content preference of users in different time periods are counted. Partitioning standards are established simultaneously to generate user group partitioning results;

[0134] Call the identified interaction time period and content type data, and classify them according to the user's interaction behavior characteristics. First, extract all user interaction records from the collected user interaction data, and classify and store them according to time period and content type, ensuring that each interaction record contains information such as user ID, interaction type, interaction time and content category. Next, calculate the interaction frequency of each user in different time periods, identify high-active users and low-active user groups, and count the total number of user interactions in different time periods to measure the overall activity. Subsequently, analyze the interaction preferences of each user in different content categories based on the user's interaction content, calculate the interaction ratio of each user in different categories of content, and classify user interest tags according to the ratio to ensure that each user's interest characteristics are clearly visible. Finally, establish user classification standards based on the above data, and partition users according to multiple dimensions such as interaction activity, interaction type preference and content interest points to ensure that the division of user groups is consistent and logical, and finally generate user group partition results.

[0135] Based on the user group partition results, analyze the interactive behavior patterns of users in each partition, count the frequency of each interactive behavior in the partition, calculate the proportion of each behavior category in different time periods, compare the behavioral characteristics between different partitions, identify the key interactive trends and behavioral changes in each partition, and obtain the user behavior trend analysis results in the partition;

[0136] Based on the user group partitioning results, the interactive behavior patterns of users in each partition are analyzed. First, the frequency of occurrence of each interactive behavior in the partition is counted, and the distribution of users' interactive methods in different time periods is obtained to ensure that all interactive data are compared along the time dimension. Next, the interactive methods, active time and content preference characteristics of users in each partition are extracted, the proportion of different interactive behaviors is calculated, and key interactive trends are extracted according to the distribution of time periods. Subsequently, the behavioral characteristics between different partitions are compared to analyze whether there are similarities in the interactive methods of different user groups, and the behavioral stability of each partition is evaluated. The range of changes in user behavior in each time period is calculated, the time periods with large changes in interaction are identified, and stable user behavior patterns are determined. Finally, the behavioral changes of user groups in each time period are compared to ensure the integrity and interpretability of the data and obtain the user behavior trend analysis results within the partition.

[0137] Based on the analysis results of user behavior trends within the partition, identify key behavior patterns, extract key interaction types of users in the target time period, analyze the stability of user behavior and changes in interaction methods over time, compare the behavior concentration of users in different partitions, and generate user behavior analysis results based on the overall changes in user group behavior patterns;

[0138] Based on the user behavior trends within the partitions, key behavior patterns are identified. First, the interaction data of each user group partition is obtained, and the main interaction types of each group in different time periods are analyzed, including interaction methods such as likes, comments, and sharing, and the proportion of each interaction method is calculated to extract the most representative user behavior characteristics. Next, the behavior concentration of different user groups in each interaction time period is analyzed, and the relative proportion of each interaction category is calculated to ensure that the significance of certain behaviors in certain time periods is quantified. Subsequently, the behavior concentration between different user groups is compared, and the key behavior characteristics of each group are summarized. Combined with the overall changes in the user group behavior patterns, it is analyzed whether the behavior change trend has periodicity or a specific time pattern, and finally the user behavior analysis results are generated.

[0139] See also Figure 6 , the steps to distinguish user activity and preference differences are:

[0140] According to the results of user behavior analysis, statistics are collected on the interaction frequency, behavior stability, and content preference data of different user groups, key behavior characteristics between groups are compared, distribution differences of different user groups in each behavior category are calculated, and the behavior pattern comparison results of different user groups are obtained;

[0141] Based on the results of user behavior analysis, the behavior patterns of different user groups are extracted. First, the interaction data of users in multiple time periods are collected, including operations such as likes, comments, shares, and collections, and the interaction behaviors of each user are arranged in chronological order to ensure the time correlation of the data. Next, the behavioral characteristics of different user groups are counted, and the interaction frequency of each group is calculated, that is, the total number of interactions within a specific time, and the distribution of interactive behaviors in different time periods is analyzed to identify the active time periods of users. Subsequently, the content preference characteristics of users are extracted, and the proportion of users' interactions with different content categories, such as video content categories, theme types, etc., are calculated, and the similarities of users within the group in these characteristics are compared to ensure that the behavior patterns within each user group have a certain stability. Next, the behavior patterns between different user groups are compared, and the differences between different groups in terms of interaction frequency, content preference, time distribution, etc. are calculated, and the standard deviation within the user group is used to measure the consistency within the group. Finally, the behavioral characteristics and difference indicators of each group are combined to obtain the comparison results of the behavior patterns of different user groups, and the key behavioral characteristics of each group are output for further classification and analysis.

[0142] Based on the comparison results of the behavioral patterns of different user groups, the core behavioral characteristics of each group are extracted, and the activity level is set in combination with the interaction frequency and time characteristics. The formula is:

[0143]

[0144] Calculate the user's activity S and obtain the classification results of user activity and preference differences, where I i represents the number of interactions of the user in the i-th time period, U represents the total number of users, and m represents the total number of time periods;

[0145] Interactions I i : Extract user interaction records in each time period from user behavior analysis data and calculate the total amount of interaction;

[0146] Total number of users U: Count the total number of users in a specific behavior category, excluding users who have not had effective interactions to ensure the accuracy of the calculation.

[0147] Set the interaction frequency for 5 user groups:

[0148] I={16, 25, 36, 49, 64};

[0149] Total number of users:

[0150] U=5;

[0151] Compute the sum of square roots:

[0152]

[0153] Calculate activity:

[0154]

[0155] The results show that the calculated activity value is 6, indicating that the average interactive activity of this user group is high. This value can be used to distinguish the activity levels of different user groups. A higher activity value indicates that the user interaction frequency of this group is high and the users remain active in multiple time periods, while a lower activity value indicates that the user interaction of this group is relatively scattered, and there may be situations where the users are active in some time periods and inactive in other time periods.

[0156] Based on the classification results of user activity and preference differences, use behavioral tags to mark users' interactive activity and content tendencies, set tag classification standards, classify users according to their behavioral characteristics, reflect the key behavioral characteristics of users under different categories, and generate user behavior tags;

[0157] Based on the classification results of user activity and preference differences, behavioral labels are used to annotate users' interactive activity and content tendencies. First, the classification criteria for user behavior labels are set, and different user groups are graded according to their interactive activity ranges. For example, users are divided into three categories: high activity, medium activity, and low activity, and the active threshold corresponding to each category is defined. Next, the main interaction methods of different user groups are analyzed, and the proportion of users in different interaction types such as likes, comments, shares, and favorites is calculated. Users are classified into different behavioral label categories based on their interactive habits. For example, some users are more inclined to like and less likely to comment, so their behavior labels can be marked as "prefer to like users". Subsequently, the distribution of users under different categories is calculated to ensure the rationality of the classification, and the consistency of user behavior within different categories is evaluated to select the label system with the best classification effect. Finally, the calculated behavior label is attached to each user record, and the corresponding relationship between user behavior labels is stored for subsequent personalized recommendations or user behavior analysis. Finally, user behavior labels are generated, and the main behavioral characteristics of users under different categories are output.

[0158] See also Figure 7 , the steps to obtain the group user behavior portrait are:

[0159] According to user behavior tags, the average behavior of each group is calculated to determine the common behavior characteristics of users within the group and establish the group average behavior data;

[0160] According to user behavior labels, user behavior data is extracted, including interaction frequency, content preference, time distribution and other characteristics. First, according to the division results of user groups, all user behavior data in each group are aggregated and counted, and the total number of interactions, active time distribution and content category selection ratio of users in the group are calculated. Then, the behavior data of each group is normalized to ensure that the behavior characteristics between different groups can be compared horizontally. Subsequently, the average values ​​of users within the group in terms of interaction methods, active time periods, content interests, etc. are calculated, and the sliding window method is used to smooth out outliers to ensure that the calculated group mean can accurately reflect the overall behavior pattern of the group. Finally, the calculated group average behavior data is stored to provide a benchmark for subsequent comparative analysis of user individual data, and finally the group average behavior data is established.

[0161] Based on the group average behavior data, compare each user's individual behavior data with the average data of the group to which they belong, calculate the deviation of the individual behavior data relative to the group mean, analyze the consistency and difference between individual behavior and group behavior, and output the user group behavior characteristics;

[0162] Based on the average behavior data of the group, the individual behavior data of each user is compared with the average data of the group to which he belongs. First, the user data in multiple dimensions such as interaction frequency, content preference, and active time distribution are extracted, and the deviation value of each dimension is calculated. Then, the degree of deviation of the individual behavior data relative to the group mean is calculated, and the different features are normalized using a standardized method to ensure the comparability between the feature dimensions. Subsequently, the consistency and difference between individual behavior and group behavior are analyzed, the behavioral feature deviation value of each user is calculated, and the user's uniqueness in the group is determined based on the deviation value. Finally, based on the calculation results, the unique characteristics of the user behavior pattern are extracted, the preference degree and interaction method of different users are classified, and the behavioral characteristics of the user group are obtained.

[0163] Based on the behavioral characteristics of user groups, the key behavioral trends of different groups are summarized. Combined with the user's interactive activity, time distribution and content preference, the core interactive mode and characteristics of group users are reflected to build a behavioral portrait of group users.

[0164] Based on the behavioral characteristics of user groups, the core interaction patterns of each user are integrated. First, the interaction data of users in different time periods are collected, and the proportion of users in behaviors such as likes, comments, shares, and browsing is calculated. Then, the behavioral patterns of different groups are compared, and the interaction characteristics of different users are classified. Users are classified according to their interaction tendencies, such as high-activity groups, medium-activity groups, and low-activity groups. Subsequently, the behavioral trends of different user groups are analyzed, and the differences in users in terms of content preference, time activity, interaction methods, and other dimensions are calculated, and the main behavioral patterns of each group are summarized. Finally, based on the calculated characteristic data of each group, a group user behavior portrait is constructed, and the final group behavior pattern is generated for subsequent user personalized analysis.

[0165] See also Figure 8 , the steps to obtain the dynamic recommendation list are:

[0166] Based on the group user behavior portrait, collect user interaction frequency, content browsing time, and the number of target behavior occurrences, calculate the behavior distribution within different user groups, and normalize the characteristic values ​​of each group to obtain user behavior distribution data;

[0167] Based on group user behavior portraits, we first need to extract the core behavior characteristics of different user groups, call user behavior data, and filter according to multiple dimensions such as time dimension, interaction frequency, content preference, historical click records, etc., and extract users' access behavior, stay duration, interaction frequency and preferred content type in different time periods. Then, we aggregate the data to form a preliminary set of behavior characteristics. Then, for the behavior characteristic set, we calculate the proportion of each type of behavior data in the overall group user portrait, determine the main behavior trends of different groups, further subdivide the behavior habits of different user groups, classify the behavior habits according to feature labels, and calculate the behavior weights corresponding to each label, and finally obtain user behavior distribution data.

[0168] According to the user behavior distribution data, analyze the behavioral correlation between user groups, match them according to the similarity of feature values, calculate the degree of difference in behavior patterns, establish a user group behavior pattern comparison matrix, classify the groups according to the matching degree, and generate the user group behavior pattern classification results;

[0169] After obtaining the core behavioral feature data of the user group, the data is called to analyze the behavioral correlation between different groups, and the behavioral intersections between different user groups are extracted, including the similarity of content interaction, overlap of visit time, common trends in click hotspots, and other aspects. The correlation indicators are quantified, and the common characteristics between similar groups are calculated through the behavioral feature matching degree. The common characteristics are then classified, and different user preference levels are divided according to standards such as behavior frequency, content attention, and interaction methods. The proportion of users at each level is also counted to form a classification result of user group behavior patterns.

[0170] Based on the classification results of user group behavior patterns, analyze the periodic changes of behavior patterns within each group, extract the distribution of key behavior patterns in different time periods, predict changes in user needs based on behavior change trends, adjust recommended content, and establish a dynamic recommendation content list;

[0171] Based on the classification results of user group behavior patterns and combined with user behavior cycle data, the behavior change trends of users in different time periods are statistically analyzed, short-term behavior patterns, long-term behavior patterns and periodic behavior characteristics are extracted, data backtracking is performed on short-term behavior patterns, and the content interaction trends of users within a certain time window are analyzed. Trend fitting is performed on long-term behavior patterns, and the stability and fluctuation of user behavior are observed. Finally, peak statistics are performed on periodic behavior characteristics, high-frequency interaction time points are calculated, and the laws of behavior cycle changes are summarized. Based on the analysis results, user behavior cycle data is established and input into the dynamic recommendation adjustment module to realize dynamic updating of user personalized recommendation content.

[0172] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. An account recommendation system based on user behavior, characterized in that: The system comprises: The account content screening module matches and screens the content tags, industry classifications, video release times and account attributes of the account data according to the query parameters entered by the user, verifies the data matching degree item by item, accumulates the matching results, and outputs the account screening results; The video hotspot analysis module receives the account screening results, collects corresponding video data, extracts video titles and cooperative categories, identifies keywords and calculates keyword frequencies through statistical analysis, determines hot topics, and generates video hotspot analysis data; The user behavior analysis module collects user interaction data based on the video hotspot analysis data, records user activities using timestamps, filters the data, identifies the interaction time period and content type, partitions the data according to user groups based on the identification results, performs statistical analysis on the user behavior data in each partition, identifies key behavior patterns, and generates user behavior analysis results; The behavior profiling module compares the behavior patterns of different user groups based on the user behavior analysis results, uses behavior tags to distinguish differences in user activity and preferences, and compares the behavior data of each user with the group average data based on the distinction results, thereby revealing the behavior characteristics of each user group and constructing a group user behavior profile; The customized recommendation module performs correlation evaluation based on the user behavior profile of the group, analyzes user preferences and behavior cycles, and adjusts the recommended content in real time to generate a dynamic recommendation list.

2. The account recommendation system based on user behavior according to claim 1, characterized in that: The steps for obtaining the account screening result are as follows: According to the query parameters input by the user, the content tags, industry classifications, video release time and account attributes of the account data are standardized, a standardized matrix of various parameters is established, each parameter is normalized, and the weight vector of each parameter is calculated based on the query parameters to generate a standardized parameter matrix and a weight vector set; Based on the standardized parameter matrix and weight vector set, the formula is adopted: Calculate the comprehensive matching degree M of the i-th account in terms of content tags, industry classification, video release time and account attributes i , and construct the account matching matrix, where w j represents the weight of the jth parameter, P ij represents the standardized value of the i-th account on the j-th parameter, Q ij represents the normalized value of the query parameter at the jth parameter, and n represents the number of parameters; Based on the account matching matrix, the matching results are cumulatively calculated, the matching degrees of all accounts are sorted, a matching threshold is set to filter out a set of accounts that meet the requirements, and a comprehensive analysis is performed on the account's content tags, industry classifications, video release times, and account attributes to output the account screening results.

3. The account recommendation system based on user behavior according to claim 2, characterized in that: The steps for obtaining the video hotspot analysis data are as follows: According to the account screening results, the corresponding video data is collected, the video titles and cooperative category contents are extracted, special characters, punctuation marks and stop words are removed, the text content is standardized, the text units are split according to word boundaries, and the number of occurrences of all words is counted, and the video text data set is established in combination with the cooperative category information; Based on the video text data set, correlation analysis is performed in combination with the cooperative categories, screening conditions are set and keywords are determined, using the formula: Calculate the relative frequency K of the keywords and get the keyword frequency distribution result, where f i represents the number of occurrences of the ith keyword, Z represents the total number of occurrences of all keywords in the video title, and N represents the total number of keywords; Based on the keyword frequency distribution results, a screening threshold is set to determine high-frequency keywords, the hot topic coverage is calculated according to the category to which the keyword belongs, the proportion of hot topics is calculated based on the number of videos, and video hot topic analysis data is generated.

4. The account recommendation system based on user behavior according to claim 3, characterized in that: The steps for identifying the interaction time period and content type are: Based on the video hotspot analysis data, user interaction data is collected, user behaviors during video playback are recorded, timestamps of user interactions are extracted, user interaction records are associated with video IDs, and all user interaction behaviors are arranged in chronological order to generate a user interaction time series; Based on the user interaction time series, repeated interaction records are removed, a time window is set for data smoothing, and the data is classified according to the time period in which the user interaction occurs, using the formula: Calculate the interaction concentration C within the time period and obtain the distribution record of the interaction time period, where A i represents the frequency of the i-th user interaction behavior, U represents the number of users who interacted in this time period, and M represents the total number of user interaction behaviors; Based on the interaction time period distribution records, cluster analysis is performed according to the time periods when the interaction behaviors occur, and high interaction time intervals are extracted. The user interaction behaviors are associated with the video content types, and the interaction proportions of different content types in each time period are statistically analyzed. The interaction concentration of the target content type is analyzed to obtain the interaction time period and content type identification results.

5. The account recommendation system based on user behavior according to claim 4, characterized in that: The steps for obtaining the user behavior analysis results are: Based on the interaction time period and content type identification results, users are classified according to their interaction behavior characteristics, and the interaction frequency, interaction method and content preference of users in different time periods are counted, and a partitioning standard is simultaneously established to generate a user group partitioning result; Based on the user group partition results, analyze the interactive behavior patterns of users in each partition, count the frequency of each interactive behavior in the partition, calculate the proportion of each behavior category in different time periods, compare the behavioral feature differences between different partitions, identify the key interactive trends and behavioral changes of each partition, and obtain the user behavior trend analysis results in the partition; Based on the user behavior trend analysis results within the partition, key behavior patterns are identified, key interaction types of users in the target time period are extracted, the stability of user behavior and changes in interaction methods over time are analyzed, the behavior concentration of users in different partitions is compared, and combined with the overall changes in user group behavior patterns, user behavior analysis results are generated.

6. The account recommendation system based on user behavior according to claim 5, characterized in that: The steps for distinguishing the user activity and preference differences are as follows: According to the user behavior analysis results, statistically analyze the interaction frequency, behavior stability, and content preference data of different user groups, compare the key behavior characteristics between groups, calculate the distribution differences of different user groups in each behavior category, and obtain the behavior pattern comparison results of different user groups; Based on the comparison results of the behavioral patterns of the different user groups, the core behavioral characteristics of each group are extracted, and the activity level is set in combination with the interaction frequency and time characteristics, using the formula: Calculate the user's activity S and obtain the classification results of user activity and preference differences, where I i represents the number of interactions of the user in the i-th time period, U represents the total number of users, and m represents the total number of time periods; Based on the classification results of the user activity and preference differences, use behavioral labels to mark the user's interactive activity and content tendencies, set label classification standards, classify users according to behavioral characteristics, reflect the key behavioral characteristics of users under different categories, and generate user behavior labels.

7. The account recommendation system based on user behavior according to claim 6, characterized in that: The steps for obtaining the group user behavior portrait are as follows: According to the user behavior tags, the behavior average of each group is calculated, the common behavior characteristics of users within the group are determined, and the group average behavior data is established; Based on the group average behavior data, the individual behavior data of each user is compared with the average data of the group to which they belong, the deviation of the individual behavior data relative to the group mean is calculated, the consistency and difference between the individual behavior and the group behavior are analyzed, and the user group behavior characteristics are output; Based on the behavioral characteristics of the user groups, the key behavioral trends of different groups are summarized, and combined with the users' interactive activity, time distribution and content preferences, the core interaction patterns and characteristics of the group users are reflected to construct a behavioral portrait of the group users.

8. The account recommendation system based on user behavior according to claim 7, characterized in that: The steps for obtaining the dynamic recommendation list are: Based on the user behavior portrait of the group, collect the user interaction frequency, content browsing time, and the number of times the target behavior occurs, calculate the behavior distribution within the different user groups, and normalize the characteristic values ​​of each group to obtain user behavior distribution data; According to the user behavior distribution data, the behavior correlation between user groups is analyzed, matching is performed based on the similarity of feature values, and the degree of difference in behavior patterns is calculated, a user group behavior pattern comparison matrix is ​​established, groups are classified based on the degree of matching, and user group behavior pattern classification results are generated; Based on the classification results of the user group behavior patterns, the periodic changes of the behavior patterns within each group are analyzed, the distribution of key behavior patterns in different time periods is extracted, the changes in user demand are predicted according to the behavior change trends, the recommended content is adjusted, and a dynamic recommended content list is established.

Images