User management system based on AI big data

Abstract

The invention belongs to the technical field of artificial intelligence and big data processing, and particularly discloses a user management system based on AI big data, and the system mainly comprises a data collection module which is in butt joint with a third-party service and social media platform through an API interface, and collects user behavior data; the feature extraction module processes basic information and behavior data of the user; the user behavior tag generation module generates behavior tags in combination with user features and interests; the user interest analysis module adopts an interest score TF-IDF calculation method to quantify user interests; the pushing grade module divides pushing grades according to the scores of the user behavior labels; the personalized recommendation module monitors the user participation degree and adjusts a video pushing strategy; the dynamic adjustment module adjusts the participation degree score in real time through an exponential decay function, and optimizes the video distribution and pushing sequence; according to the method, accurate personalized recommendation is realized by deeply mining the user data, and the user satisfaction and the interaction efficiency of the platform are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence and big data processing, and specifically relates to an AI big data user management system. Background Art

[0002] With the popularization of the Internet and mobile devices, the amount of data generated by users has exploded. For different user behaviors, user engagement cannot be significantly improved. By introducing an exponential decay function to dynamically adjust user interest and engagement scores, considering that user interest in content will decay over time and lack of dynamic adjustment technology, it can ensure that the push strategy is consistent with the actual changes in user interests, thereby continuously optimizing the push order and frequency of content. The system can manage and adjust the push strategy more intelligently, so that the user experience can be continuously improved and optimized. Summary of the invention

[0003] The implementation method in this article can not only improve user experience and engagement, but also effectively optimize content push and management strategies, thereby achieving more intelligent and personalized user management and services with the support of big data and AI technologies.

[0004] The technical solution adopted by the present invention to solve its technical problem is: the user management system based on AI big data described in the present invention includes a data acquisition module, a feature extraction module, a user behavior label generation module, a user interest analysis module, a push level module, a personalized recommendation module and a dynamic adjustment module;

[0005] A data collection module, which connects to a third-party service or social media platform through an API to collect user behavior data;

[0006] A feature extraction module, wherein the feature extraction module obtains first feature data and second feature data of the user and performs encoding processing on these data;

[0007] A user behavior tag generation module, wherein the user behavior tag generation module calculates a user behavior tag score by defining the user behavior tag and combining gender, age and interest weights;

[0008] User interest analysis module, which treats user behavior data as a document collection, calculates the user interest score TF-IDF, compares it with the interest score threshold, and labels the user accordingly;

[0009] A push level module, which divides push levels into level one, level two, and level three according to user behavior tag scores;

[0010] A personalized recommendation module, which calculates the engagement score by monitoring the user's engagement with the video, and adjusts the video push strategy based on the engagement score to give priority to pushing videos with high engagement scores;

[0011] A dynamic adjustment module, wherein the dynamic adjustment module dynamically adjusts the engagement score through a decay rate β, and adjusts the video distribution and push sequence according to the adjusted engagement score.

[0012] Preferably, the gender and age of the user are obtained and recorded as the first characteristic data;

[0013] Obtain the user's browsing history, search keywords, and favorites in the application and record them as the second feature data;

[0014] Encode the first characteristic data, convert gender into binary code, male is recorded as 1, female is recorded as 2, under 20 years old is recorded as 3, 21 to 40 years old is recorded as 4, 41 to 60 years old is recorded as 5, and over 61 years old is recorded as 6;

[0015] For the second feature data, the user's text data set is regarded as a document set, where each record is regarded as a document, and the number of documents in which each keyword t appears in the user document d is counted.

[0016] Preferably, the first characteristic data and the second characteristic data of the user are integrated and set as

[0017] User behavior labels are generated, and a user behavior label mapping table is generated based on each feature data.

[0018] Preferably, by formula Calculate word frequency ;

[0019] in, represents the number of times word t appears in document d, Represents the total number of occurrences of all words in document d; TF reflects the frequency of a word in a document;

[0020] By formula Calculate the inverse document frequency IDF;

[0021] Among them, N represents the total number of documents, dft represents the number of documents containing word t, and IDF represents the rarity of the word;

[0022] The user interest score TF-IDF is calculated by the formula TF-IDF=TF*IDF;

[0023] Compare the interest score TF-IDF with the interest score threshold to get the tags that the user is interested in.

[0024] Preferably, the gender weight is assigned as Wgender, the age weight is assigned as Wage, and the interest weight is assigned as Winterest, and Wgender+Wage+Winterest=1 is satisfied;

[0025] The formula S = (Wgender × GenderCode) + (Wage × AgeCode) + (Winterest × TF−IDFm×[TF−IDFm>Tlabel]) is used to calculate the user behavior label score S;

[0026] Compare the user behavior label score S with the preset user behavior label score threshold and add a corresponding label;

[0027] Where [TF−IDFm>Tlabel] is a Boolean expression, TF−IDFm represents the interest score of the mth user, m=1, 2, ..., n; it is 1 if the TF-IDF value is greater than the threshold, otherwise it is 0.

[0028] Preferably, the m users collected are given corresponding label scores and marked with corresponding

[0029] The labels are divided into level one, level two, and level three according to the sports label score Sa, e-sports label score Sb, and food label score Sc from large to small.

[0030] Preferably, taking short video push as an example, it is assumed that the m users collected all meet the requirements of three tags;

[0031] Where m≥1;

[0032] Real-time monitoring of users' viewing of short videos during the period;

[0033] Get the completion rate, number of likes, number of shares, number of comments, viewing time, and number of views of the pushed video;

[0034] The weight of the completion rate is assigned as W, the weight of the number of likes is assigned as Lk, the weight of the number of shares is assigned as Sh, and the weight of the number of comments is assigned as Cm, and W+Lk+Sh+Cm=1;

[0035] Calculating the Level 1 Engagement Score , Secondary Engagement Score and level 3 engagement scores By comparing the values ​​of , the tags with the highest engagement scores are pushed first.

[0036] Preferably, an exponential decay function is introduced to adjust the engagement score to dynamically adjust the push of future videos;

[0037] Assuming that the user's interest decays exponentially over time, the decay rate is controlled by β, and the formula δn= Calculate the attenuation coefficient δn of the nth video;

[0038] According to the decay coefficient, the engagement score is adjusted by the formula

[0039] Calculate the adjusted engagement score ;

[0040] Where i represents the push level, i=1, 2 and 3; Yi is the number of videos in the i-th level, Ti[n], Lki[n], Shi[n] and Cmi[n] are the viewing time, number of likes, number of shares and number of comments of the n-th video respectively.

[0041] Preferably, the decay rate β is the speed at which the user's interest decays over time, and the acquisition process is:

[0042] Through the equation PIi(t) = PIi(0)* To describe the time decay rate β changes with time;

[0043] PIi(t) is the observed engagement score at time t;

[0044] PIi(0) is the initial participation score value, i.e., the value at t=0;

[0045] Collect a series of observations (ti,PIi(t)), where ti is a time point and PIi(t) is the corresponding observation;

[0046] Get the total duration of a series of observations, calculate the ratio of the total duration to the number of time nodes i, and get the average duration ty;

[0047] Get the sum of a series of observations and calculate the ratio with the number of time nodes i to get the average observation value PIiy at the time point;

[0048] By formula The decay rate β is calculated;

[0049] Among them, i=1, 2,...n; PIi is the participation score of the corresponding time node.

[0050] Preferably, based on the calculated engagement score ,Adjust push strategy;

[0051] Assume that it is expected that N videos will be provided;

[0052] N videos are allocated according to the proportion of the engagement scores, along with the adjusted first-level engagement score PIS1, the adjusted second-level engagement score PIS2, and the adjusted third-level engagement score PIS3, and are pushed and sorted in descending order according to the engagement scores.

[0053] The beneficial effects of the present invention are as follows:

[0054] (1) By analyzing user behavior data and interest tags, the system can accurately understand each user's interests and preferences. This personalized recommendation can significantly improve user engagement, for example, by pushing content that is more in line with user interests to increase video viewing completion rate, likes, shares, and comments. This not only increases user activity on the platform, but also improves content dissemination and social interaction.

[0055] (2) An exponential decay function is introduced to dynamically adjust user interest and engagement scores, taking into account that user interest in content will decay over time. This dynamic adjustment ensures that the push strategy is consistent with the actual changes in user interests, thereby continuously optimizing the order and frequency of content push. In this way, the system can manage and adjust the push strategy more intelligently, so that the user experience can be continuously improved and optimized.

[0056] (3) The implementation method in this paper can not only improve user experience and engagement, but also effectively optimize content push and management strategies, thereby achieving more intelligent and personalized user management and services with the support of big data and AI technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The present invention will be further described below in conjunction with the accompanying drawings.

[0058] Figure 1 This is a schematic diagram of the structure of an AI big data user management system of the present invention;

[0059] Figure 2 This is a schematic diagram of the function of an AI big data user management system module of the present invention;

[0060] Figure 3 This is a schematic diagram of a characteristic data processing structure based on an AI big data user management system of the present invention; DETAILED DESCRIPTION

[0061] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the present invention is further explained below in conjunction with specific implementation methods.

[0062] Example 1

[0063] See also Figure 1-2As shown, an AI big data user management system according to an embodiment of the present invention includes a data collection module, a feature extraction module, a user behavior tag generation module, a user interest analysis module, a push level module, a personalized recommendation module and a dynamic adjustment module;

[0064] A data collection module, which connects to a third-party service or social media platform through an API to collect user behavior data;

[0065] A feature extraction module, wherein the feature extraction module obtains the user's gender and age as first feature data, and the browsing history, search keywords, and favorites as second feature data, and encodes and processes these data;

[0066] A user behavior tag generation module, wherein the user behavior tag generation module defines user behavior tags, including sports tags, e-sports tags, and food tags; and calculates user behavior tag scores by combining gender, age, and interest weights;

[0067] A user interest analysis module, which calculates the user interest score TF-IDF by treating the user behavior data as a document set. For each user behavior tag, a TF-IDF threshold is set. Only when the TF-IDF value of a certain keyword of the user exceeds the threshold, the user is considered to be interested in the tag;

[0068] A push level module, which divides push levels into level one, level two, and level three according to user behavior tag scores;

[0069] A personalized recommendation module, which calculates the engagement score by monitoring the user's engagement with the video, including the completion rate, number of likes, number of shares, and number of comments, and adjusts the video push strategy based on the engagement score to give priority to pushing content with high engagement;

[0070] A dynamic adjustment module, wherein the dynamic adjustment module dynamically adjusts the engagement score by a decay rate β by considering the decay of user interest over time, and adjusts the video distribution and push order according to the adjusted engagement score.

[0071] Example 2

[0072] See also Figure 3 As shown, data collection SDK is integrated into mobile applications to track user behavior within the application, including page browsing, clicks, dwell time, and purchase behavior, and connect with third-party services or social media platforms through API interfaces to collect users' public information and behavior data;

[0073] Obtain the user's gender and age, and record them as the first feature data;

[0074] Obtain the user's browsing history, search keywords, and favorites in the application and record them as the second feature data;

[0075] The first characteristic data is used to record the user's gender and age, and the second characteristic data is used to identify the user's interests and hobbies;

[0076] Specifically: Encode the first feature data, convert gender into binary code, record male as 1, female as 2, record under 20 as 3, 21 to 40 as 4, 41 to 60 as 5, and over 61 as 6;

[0077] For the second feature data, the user's text data set is regarded as a document set, where each record (such as a search record, a browsing behavior) is regarded as a document, and the number of documents in which each keyword t appears in the user document d is counted.

[0078] Example 3

[0079] Integrate the first characteristic data and the second characteristic data of the user, set them as user behavior tags, and generate a user behavior tag mapping table based on each characteristic data;

[0080] The user behavior tag mapping table includes: sports tag a, e-sports tag b, and food tag c;

[0081] Specifically: collect m user behaviors; extract the first feature value and the second feature value of each user respectively

[0082] Eigenvalue;

[0083] By formula Calculate word frequency ;

[0084] in, represents the number of times word t appears in document d, Represents the total number of occurrences of all words in document d; TF reflects the frequency of a word in a document. The higher the frequency, the greater the TF value.

[0085] By formula Calculate the inverse document frequency IDF;

[0086] Where N represents the total number of documents, dft represents the number of documents containing word t, and IDF reflects the rarity of a word in the entire document collection. The rarer the word, the larger the IDF value.

[0087] The user interest score TF-IDF is calculated by the formula TF-IDF=TF*IDF;

[0088] It should be noted that for one of the keywords, the TF-IDF value reflects the importance of this word to the user's interest.

[0089] For each user behavior tag, a TF-IDF threshold is set. Only when the TF-IDF value of a user's keyword exceeds the threshold, the user is considered to be interested in the tag.

[0090] Assign gender weight to Wgender, age weight to Wage, and interest weight to Winterest, and satisfy Wgender + Wage + Winterest = 1;

[0091] The formula S = (Wgender × GenderCode) + (Wage × AgeCode) + (Winterest × TF−IDFm×[TF−IDFm>Tlabel]) is used to calculate the user behavior label score S;

[0092] The user behavior label score S is compared with the preset user behavior label score threshold. If the user behavior label score is greater than or equal to the preset user behavior label score threshold, the corresponding user is labeled with the corresponding behavior label; otherwise, the labeling requirement is not met.

[0093] Where [TF−IDFm>Tlabel] is a Boolean expression, TF−IDFm represents the interest score of the mth user, m=1, 2, ..., n; it is 1 if the TF-IDF value is greater than the threshold, otherwise it is 0, which means that only those keywords exceeding the threshold will be included in the calculation of the interest weight.

[0094] For example: Assume that the gender weight Wgender=0.2; the age weight Wage=0.3; the interest weight Winterest=0.5;

[0095] User A's information: Gender code: 1 (male); Age code: 4 (21 to 40 years old);

[0096] Interest score TF-IDF values ​​under the sports tag: 0.6, 0.75, 0.9 (assuming the first two are below the threshold and the last one is above the threshold)

[0097] User A’s motion tag score = (0.2*1) + (0.3*4) + (0.5*0.9*1) = 1.85;

[0098] Assuming that the score of the motion tag of user A is greater than the score threshold of the motion tag, user A will be marked with the corresponding motion tag.

[0099] Example 4

[0100] Obtain the corresponding tag scores for the m collected users respectively, and add corresponding tags, and divide the push levels according to the size of the sports tag score Sa, e-sports tag score Sb, and food tag score Sc;

[0101] Among them, according to the behavior tag scores from large to small, they are divided into level one, level two, and level three;

[0102] Taking short video push as an example, assume that all m users collected meet the requirements of three tags;

[0103] Where m≥1;

[0104] Real-time monitoring of users watching short videos during the period:

[0105] Get the completion rate Wa, number of likes LKa, number of shares Sha, number of comments Cma, browsing time Ta, and number of views Ya of the first-level pushed video;

[0106] Get the completion rate Wb, number of likes LKb, number of shares Shb, number of comments Cmb, browsing time Tb, and number of views Yb of the secondary pushed video;

[0107] Get the completion rate Wc, number of likes LKc, number of shares Shc, number of comments Cmc, browsing time Tc, and number of views Yc of the third-level pushed videos;

[0108] Considering that different participation methods have different impacts on the video, we can set weights for each participation method.

[0109] The weight of the completion rate is assigned as W, the weight of the number of likes is assigned as Lk, the weight of the number of shares is assigned as Sh, and the weight of the number of comments is assigned as Cm, and W+Lk+Sh+Cm=1;

[0110] For the video push of the first-level tag, the formula

[0111] Calculate the first-level engagement score ;

[0112] For video push of secondary tags, the formula

[0113] Calculate the secondary engagement score ;

[0114] For the video push of the third-level tag, the formula

[0115] Calculate the three-level engagement score ;

[0116] Analyze and compare first-level engagement scores , Secondary Engagement Score and level 3 engagement scores Change the push strategy to prioritize tags with the highest engagement scores.

[0117] Example 5

[0118] Since users’ interest in videos decreases with time and the number of videos, we can introduce an exponential decay function to adjust the engagement score to dynamically adjust the push of future videos;

[0119] Assuming that the user's interest decays exponentially over time, the decay rate is controlled by β, and the formula δn= Calculate the attenuation coefficient δn of the nth video;

[0120] Among them, the decay rate β is the speed at which user interest decays over time, and the acquisition process is:

[0121] Through the equation PIi(t) = PIi(0)* To describe the time decay rate β changes with time;

[0122] PIi(t) is the observed engagement score at time t;

[0123] PIi(0) is the initial participation score value, i.e., the value at t=0;

[0124] Collect a series of observations (ti,PIi(t)), where ti is a time point and PIi(t) is the corresponding observation;

[0125] Get the total duration of a series of observations, calculate the ratio of the total duration to the number of time nodes i, and get the average duration ty;

[0126] Get the sum of a series of observations and calculate the ratio with the number of time nodes i to get the average observation value PIiy at the time point;

[0127] By formula The decay rate β is calculated;

[0128] Among them, i=1, 2,...n; PIi is the participation score of the corresponding time node.

[0129] The engagement score is adjusted based on the decay factor:

[0130] Calculate the adjusted engagement score ;

[0131] Where i represents the push level, i=1, 2 and 3; Yi is the number of videos in the i-th level, Ti[n], Lki[n], Shi[n] and Cmi[n] are the viewing time, number of likes, number of shares and number of comments of the n-th video respectively.

[0132] Based on the calculated engagement score ,Adjust push strategy;

[0133] Assume that it is expected that N videos will be provided;

[0134] Get the adjusted first-level participation score PIS1;

[0135] Get the adjusted secondary participation score PIS2;

[0136] Get the adjusted third-level participation score PIS3;

[0137] N videos are allocated according to the proportion of the engagement scores, along with the adjusted first-level engagement score PIS1, the adjusted second-level engagement score PIS2, and the adjusted third-level engagement score PIS3, and are pushed and sorted in descending order according to the engagement scores.

[0138] It should be noted that: According to the calculated participation score To dynamically adjust the push strategy for future videos, increase the push frequency of tagged content with high engagement, and optimize video sorting to make it easier for users to access content that they may be interested in.

[0139] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A user management system based on AI big data, characterized by: include; A data collection module, which connects to a third-party service or social media platform through an API to collect user behavior data; A feature extraction module, wherein the feature extraction module obtains the user's first feature data and the second feature data by processing the user's behavior data, and performs encoding processing on the data; A user behavior tag generation module, wherein the user behavior tag generation module calculates a user behavior tag score by defining the user behavior tag and combining gender, age and interest weights; User interest analysis module, which treats user behavior data as a document collection, calculates the user interest score TF-IDF, compares it with the interest score threshold, and labels the user accordingly; A push level module, which divides push levels into level one, level two, and level three according to user behavior tag scores; A personalized recommendation module, which calculates the engagement score by monitoring the user's engagement status with the video, and adjusts the video push strategy based on the engagement score to give priority to pushing videos with high engagement scores; A dynamic adjustment module, wherein the dynamic adjustment module dynamically adjusts the engagement score through a decay rate β, and adjusts the video distribution and push sequence according to the adjusted engagement score.

2. According to claim 1, the user management system based on AI big data is characterized by: Obtain the user's gender and age, and record them as the first feature data; Obtain the user's browsing history, search keywords, and favorites in the application and record them as the second feature data; Encode the first characteristic data, convert gender into binary code, male is recorded as 1, female is recorded as 2, under 20 years old is recorded as 3, 21 to 40 years old is recorded as 4, 41 to 60 years old is recorded as 5, and over 61 years old is recorded as 6; For the second feature data, the user's text data set is regarded as a document set, where each record is regarded as a document, and the number of documents in which each keyword t appears in the user document d is counted.

3. The AI ​​big data user management system according to claim 2 is characterized by: The user's first characteristic data and second characteristic data are integrated and set as the user behavior mark. Tag, and generate a user behavior tag mapping table based on each feature data.

4. The AI ​​big data user management system according to claim 3 is characterized by: By formula Calculate word frequency ; in, represents the number of times word t appears in document d, Represents the total number of occurrences of all words in document d; TF reflects the frequency of a word in a document; By formula Calculate the inverse document frequency IDF; Among them, N represents the total number of documents, dft represents the number of documents containing word t, and IDF represents the rarity of the word; The user interest score TF-IDF is calculated by the formula TF-IDF=TF*IDF; Compare the interest score TF-IDF with the interest score threshold to get the tags that the user is interested in.

5. According to claim 4, the user management system based on AI big data is characterized in that: Assign gender weight to Wgender, age weight to Wage, interest weight to Winterest, and satisfy Wgender + Wage + Winterest = 1. The formula S = (Wgender × GenderCode) + (Wage × AgeCode) + (Winterest × TF−IDFm×[TF−IDFm>Tlabel]) to calculate the user behavior label score S, Compare the user behavior label score S with the preset user behavior label score threshold and add a corresponding label. Where [TF−IDFm>Tlabel] is a Boolean expression, TF−IDFm represents the interest score of the mth user, m=1, 2, n; it is 1 if the TF-IDF value is greater than the threshold, otherwise it is 0.

6. The AI ​​big data user management system according to claim 5 is characterized in that The m users collected are respectively given corresponding label scores and labeled accordingly. The sports label score Sa, the e-sports label score Sb, and the food label score Sc are divided into the first, second, and third levels from large to small.

7. The AI ​​big data user management system according to claim 6 is characterized by: Taking short video push as an example, assume that all m users collected meet the requirements of three tags; Where m≥1; real-time monitoring of users’ viewing of short videos during the period; obtaining the completion rate, number of likes, number of shares, number of comments, viewing time, and number of views of the pushed videos; The weight of the completion rate is assigned as W, the weight of the number of likes is assigned as Lk, the weight of the number of shares is assigned as Sh, and the weight of the number of comments is assigned as Cm, and W+Lk+Sh+Cm=1; Calculating the Level 1 Engagement Score , Secondary Engagement Score and level 3 engagement scores By comparing the values ​​of , the tags with the highest engagement scores are pushed first.

8. The AI ​​big data user management system according to claim 7 is characterized by: Introduce an exponential decay function to adjust the engagement score to dynamically adjust the push of future videos; Assuming that the user's interest decays exponentially over time, the decay rate is controlled by β, and the formula δn= Calculate the attenuation coefficient δn of the nth video; According to the decay coefficient, the engagement score is adjusted by the formula Calculate the adjusted engagement score ; Where i represents the push level, i=1, 2 and 3; Yi is the number of videos in the i-th level, Ti[n], Lki[n], Shi[n] and Cmi[n] are the viewing time, number of likes, number of shares and number of comments of the n-th video respectively.

9. The AI ​​big data user management system according to claim 8, characterized in that: The decay rate β is the speed at which user interest decays over time, and the acquisition process is: Through the equation PIi(t) = PIi(0)* To describe the time decay rate β changes with time; PIi(t) is the observed engagement score at time t; PIi(0) is the initial participation score value, i.e., the value at t=0; Collect a series of observations (ti,PIi(t)), where ti is a time point and PIi(t) is the corresponding observation; Get the total duration of a series of observations, calculate the ratio of the total duration to the number of time nodes i, and get the average duration ty; Get the sum of a series of observations and calculate the ratio with the number of time nodes i to get the average observation value PIiy at the time point; By formula The decay rate β is calculated; Among them, i=1, 2,...n; PIi is the participation score of the corresponding time node.

10. The AI ​​big data user management system according to claim 8, characterized in that: Based on the calculated engagement score ,Adjust push strategy; Assume that it is expected that N videos will be provided; N videos are allocated according to the proportion of the engagement scores, along with the adjusted first-level engagement score PIS1, the adjusted second-level engagement score PIS2, and the adjusted third-level engagement score PIS3, and are pushed and sorted in descending order according to the engagement scores.

Images