5G message personalized pushing method and system based on AI model

Abstract

The invention relates to the technical field of wireless communication, in particular to a 5G message personalized pushing method and system based on an AI model, and the method comprises the steps: obtaining a behavior record of a user in a system use process; according to the behavior record of the user, analyzing a browsing content preference condition of the user after breaking the information cocoon room; and according to the browsing content preference condition of the user after the information cocoon room is broken, personalized message pushing is carried out to the user. According to the invention, the behavior record of the user is analyzed based on the AI model, and personalized message pushing is carried out to the 5G terminal user through the 5G network. According to the method, the browse content preference condition of the user after breaking the information cocoon room is analyzed, and the new preference type possibly displayed after the user breaks the information cocoon room is predicted and analyzed, so that the personalized pushing strategy of the system to the user preference is optimized.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and in particular to a 5G message personalized push method and system based on an AI model. Background Art

[0002] Message push has become a key strategy for various application platforms to attract users and increase usage frequency. The system aims to optimize the user experience by deeply analyzing the user's daily use of the application.

[0003] However, in the process of users using forums and other systems to obtain information, there is an information cocoon phenomenon. This may be due to the user's long-term attention and preference for a certain type of information, and the spontaneous construction of the information cocoon; it may also be due to the improper system push mechanism, that is, the system over-focuses on a certain type of information to push to users, thereby guiding users to gradually prefer to receive this type of information during use, thus forming an externally constructed information cocoon. If the information pushed by the system deviates from the user's real preferences due to the "information cocoon" effect, in the long run, it is likely to cause users' disgust and dissatisfaction, which is not conducive to the use and promotion of the application. Summary of the invention

[0004] In order to solve the technical problem that the pushed information deviates from the real preferences of users due to the "information cocoon" effect, the purpose of the present invention is to provide a 5G message personalized push method and system based on an AI model. The technical solution adopted is as follows:

[0005] The first aspect of the present invention is a personalized 5G message push method based on an AI model, and the push method includes: step S1, obtaining the user's behavior record during the use of the system; the behavior record includes the data record when the user operates on data information of different content types; step S2, analyzing the user's browsing content preference after breaking the information cocoon according to the user's behavior record; obtaining the user's preference sequence for different content types according to the data record when the user operates on data information of different content types; analyzing the changing trend of the user's preferences according to the user's preference sequence for different content types, and screening out user groups with similar preferences; from the user groups with similar preferences, screening out user groups with similar preferences and consistent preferences; from the user groups with similar preferences and consistent preferences, obtaining the user's browsing content preference after breaking the information cocoon; step S3, according to the user's browsing content preference after breaking the information cocoon, personalized message push is performed to the user.

[0006] Furthermore, based on the preference sequences of users for different types of content they browse, the changing trends of user preferences are analyzed, and user groups with similar preferences are screened out, specifically including: obtaining changes in user preference preferences based on changes in the preference sequences of users on adjacent dates; analyzing changes in user preference preferences within a certain period of time to obtain a new preference sequence for users; and obtaining user groups with similar preferences according to the new preference sequence for users.

[0007] Furthermore, based on the data records when the user operates on the data information of different content types, a sequence of the user's preference for the different content types browsed is obtained, specifically including: the data records include: browsing time, number of browsing times and browsed content; based on the browsing time, number of browsing times and browsed content when the user operates on the different content types, the user's preference coefficient for each content type is obtained; based on the size of the preference coefficient, the content types browsed by the user are sorted to obtain a sequence of the user's preference for the different content types browsed.

[0008] Furthermore, based on the changes between the preference sequences of the user on adjacent dates, the changes in the user's preferences are obtained, specifically including: generating the user's preference content for the previous day based on the user's preferences that ranked in the top 95% in the preference sequence of the previous day; pushing today's content to the user; the today's content includes the previous day's preference content and related content, the related content is different from the previous day's preference content, and the related content accounts for 5% of the today's content; based on the difference between the user's today's preference sequence and the previous day's preference sequence, obtaining a similarity parameter between the preference sequences of two adjacent days; and using the similarity parameter between the preference sequences of two adjacent days to represent the changes in the user's preferences.

[0009] Furthermore, the changes in the user's preferences include the similarity of the browsed content being the first similarity level and the similarity of the browsed content being the second similarity level; when the similarity parameter between the preference sequences of two adjacent days is greater than or equal to the first threshold, it indicates that the similarity of the browsed content of the user on two adjacent days is the first similarity level; when the similarity parameter between the preference sequences of two adjacent days is less than the first threshold, it indicates that the similarity of the browsed content of the user on two adjacent days is the second similarity level.

[0010] Furthermore, the changes in users' preferences within a certain period of time are analyzed to obtain the users' new preference sequence; according to the users' new preference sequence, a user group with similar and consistent preferences is obtained, specifically including: when the similarity of the browsing content of a user on two consecutive days is the second similarity level, according to the difference between the user's preference sequence and the preference sequences of other users, a similarity parameter between the preference sequences of the user and other users is obtained; other users whose similarity parameters between the preference sequences of the user and other users are greater than the second threshold are identified as a user group with similar preferences.

[0011] Furthermore, from the user groups with similar preferences, user groups with similar preferences and consistent preferences are screened out, specifically including: analyzing the preference degree sequence of different content types browsed by each user in the user groups with similar preferences within the current preset number of days, and the similarity parameters between the preference degree sequences of each user on two adjacent days, to obtain the difference index between the browsing record sequences of each user on two adjacent days; users whose difference index between the browsing record sequences of two adjacent days is less than or equal to a third threshold are identified as the user group with similar preferences and consistent preferences.

[0012] Furthermore, from the user group with similar and consistent preferences, the browsing content preferences of users after breaking out of the information cocoon are obtained, specifically including: analyzing the data records of the user group with similar and consistent preferences when operating data information of different content types, to obtain the total number of views and the total browsing time of each content type; according to the total number of views and the total browsing time of each content type, the popularity coefficient of each content type in the user group with similar and consistent preferences is obtained, and then the browsing content preferences of users after breaking out of the information cocoon are obtained.

[0013] Furthermore, personalized message push is performed to the user based on the user's browsing content preference after breaking out of the information cocoon, specifically including: sorting by size according to the popularity coefficient of each content type among the user group with similar and consistent preferences to form a popularity coefficient table; when the user uses the system next time, replacing the related content with the content in the popularity coefficient table whose popularity coefficient is greater than a fourth threshold.

[0014] The second aspect of the present invention is a 5G message personalized push system based on an AI model, the push system comprising: a behavior recording module, for obtaining the behavior records of users in the process of using the system; the behavior records include data records when users operate on data information of different content types; a preference analysis module, for analyzing the user's browsing content preferences after breaking out of the information cocoon based on the user's behavior records; for obtaining a sequence of the user's preference for different content types browsed based on the data records when the user operates on data information of different content types; according to the sequence of the user's preference for different content types browsed, analyzing the changing trend of the user's preferences, and screening out user groups with similar preferences; from the user groups with similar preferences, screening out user groups with similar preferences and consistent preferences; from the user groups with similar preferences and consistent preferences, obtaining the user's browsing content preferences after breaking out of the information cocoon; a message push module, for pushing personalized messages to users based on the user's browsing content preferences after breaking out of the information cocoon.

[0015] The present invention has the following beneficial effects:

[0016] The present invention aims to solve the problem that due to the information cocoon phenomenon, the information obtained by users when using information exchange platforms such as forums is single or one-sided, which makes it difficult to accurately obtain the real preferences of users, and thus it is difficult to provide the best personalized message push method. The present invention analyzes the browsing content preferences of users after breaking out of the information cocoon, such as the significant change in the proportion of original content types and new content types, and then predicts and analyzes the new preference types that users may show after breaking out of the information cocoon based on the actual behavior records of users and the preference characteristics of user groups with similar preferences, thereby optimizing the system's personalized push strategy for user preferences.

[0017] The present invention introduces a small amount of related content in the message push process, records and analyzes the user's operation behavior, such as the number of times and duration of user browsing, which produces a significant change in the proportion of the original content type and the new content type. The present invention records and analyzes the user's behavior after breaking out of the information cocoon, obtains the user's browsing time and frequency of new content types, and evaluates the user's preference for new content. Then, based on the user information with similar preferences, the personalized push strategy is optimized to achieve the best user preference recommendation push, thereby improving the system's push effect and continuously attracting users. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A 5G message personalized push method based on an AI model is provided in one embodiment of the present invention;

[0020] Figure 2 A 5G message personalized push system based on an AI model is provided as an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a 5G message personalized push method and system based on an AI model proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0022] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0023] The following is a detailed description of a specific solution of a 5G message personalized push method based on an AI model provided by the present invention in conjunction with the accompanying drawings. Figure 1 , which shows a flow chart of a personalized 5G message push method based on an AI model provided by an embodiment of the present invention. The personalized 5G message push method based on an AI model in an embodiment of the present invention comprises the following steps:

[0024] Step S1, obtain the user's behavior record in the process of using the system; the behavior record includes the data record when the user operates on data information of different content types. The system here is an application platform or self-media operation platform on various mobile terminals, and any similar platform with information promotion functions such as recommending content to users through historical data analysis. The user's behavior record in the process of using the system refers to the record traces left after the user enters the system through the user ID (Identification, identity identification number) to perform related operations, including but not limited to browsing content, browsing times, browsing time, likes, collections, forwarding, attention, etc. Content type refers to the category of the browsed content, such as popular science or variety shows. According to the needs of the actual scene, it can also be a finer category, such as popular science can be divided into natural science, technology invention, health and health, environmental protection, social science, culture and art, education and teaching, and agricultural technology.

[0025] Step S2, analyzing the user's browsing content preferences after breaking out of the information cocoon based on the user's behavior records.

[0026] According to the data records when the user operates on the data information of different content types, a sequence of the user's preference degree for the different content types browsed is obtained.

[0027] According to the browsing time, browsing times and browsing contents of users on different content types, the preference coefficient of users for each content type is obtained; according to the size of the preference coefficient, the content types browsed by users are sorted to obtain the preference degree sequence of users for the different content types browsed.

[0028] According to the preference sequence of different types of content browsed by users, the changing trend of user preferences is analyzed, and user groups with similar preferences are screened out. According to the changes between the preference sequences of users on adjacent dates, the changes in user preferences are obtained. According to the preferences of users ranked in the top 95% in the preference sequence of the previous day, the user's preference content of the previous day is generated.

[0029] Push the content of the day to the user; the content of the day includes the preferred content of the previous day and related content, the related content is different from the preferred content of the previous day, and the related content accounts for 5% of the content of the day. The 95% and 5% can be changed accordingly according to the needs of the actual scenario.

[0030] According to the difference between the user's preference sequence of the day and the preference sequence of the previous day, the similarity parameter between the preference sequences of two adjacent days is obtained; the similarity parameter between the preference sequences of two adjacent days represents the change of the user's preference. The above-mentioned related content in the implementation of the present invention refers to the content that is not the user's direct preference or weak preference information.

[0031] The changes in user preferences include the browsing content similarity being the first similarity level and the browsing content similarity being the second similarity level; when the similarity parameter between the preference degree sequences of two consecutive days is greater than or equal to the first threshold, it means that the browsing content similarity of the user on two consecutive days is the first similarity level; when the similarity parameter between the preference degree sequences of two consecutive days is less than the first threshold, it means that the browsing content similarity of the user on two consecutive days is the second similarity level. The first similarity level represents a higher browsing content similarity, and the second similarity level represents a lower browsing content similarity.

[0032] Analyze the changes in users' preferences within a certain period of time to obtain the users' new preference sequence; according to the users' new preference sequence, obtain user groups with similar preferences.

[0033] When the similarity of the browsing contents of a certain user on two consecutive days is the second similarity level; according to the difference between the preference degree sequence of the user and the preference degree sequence of other users, the similarity degree parameter between the preference degree sequence of the user and other users is obtained; other users whose similarity degree parameter between the preference degree sequence of the user and other users is greater than the second threshold are identified as a user group with similar preferences; from the user group with similar preferences, a user group with similar preferences and consistent preferences is obtained. In the embodiment of the present invention, the first threshold and the second threshold may be the same or different, preferably the same, and the value is 0.8.

[0034] Analyze the preference degree sequence of different content types browsed by each user in the current preset days and the similarity degree parameters between the preference degree sequences of each user in two adjacent days in the user group with similar preferences, and obtain the difference index between the browsing record sequences of each user in two adjacent days. Users whose difference index between the browsing record sequences of two adjacent days is less than or equal to the third threshold are identified as a user group with similar preferences and consistent preferences. In the embodiment of the present invention, the third threshold is 0.3. The current preset number of days is set to within 15 days before the current day in the embodiment of the present invention. In actual scenarios, it can be other days according to different needs and purposes.

[0035] From user groups with similar and consistent preferences, we can obtain users' browsing content preferences after breaking out of their information cocoons.

[0036] Analyze the data records of user groups with similar and consistent preferences when operating data information of different content types, and obtain the total number of views and total browsing time of each content type. According to the total number of views and total browsing time of each content type, the popularity coefficient of each content type among user groups with similar and consistent preferences is obtained, and then the browsing content preferences of users after breaking out of the information cocoon are obtained.

[0037] Step S3: Push personalized messages to users based on their browsing preferences after breaking out of the information cocoon. Sort by popularity coefficient of each content type among user groups with similar and consistent preferences to form a popularity coefficient table; when the user uses the system next time, replace the previous associated content with the content in the popularity coefficient table that has a popularity coefficient greater than the fourth threshold; that is, replace the original associated content with the content in the popularity coefficient table that has a certain percentage of popularity coefficients. When pushing the content of the day to the user, the proportion of the replaced associated content remains unchanged.

[0038] In the embodiment of the present invention, whether the user breaks out of the information cocoon is judged based on the difference between the preference degree sequence of the current day and the previous day, and whether the user's preferences are consistent is judged based on the change of the preference degree sequence in half a month. However, in some actual scenarios, this frequency time may have other possibilities, such as judging whether the user breaks out of the information cocoon based on the difference between the preference degree sequence of the current month and the previous month, and judging whether the user's preferences are consistent based on the change of the preference degree sequence in half a year; for example, judging whether the user breaks out of the information cocoon based on the difference between the preference degree sequence of the current hour and the previous hour, and judging whether the user's preferences are consistent based on the change of the preference degree sequence of the current day; for another example, judging whether the user breaks out of the information cocoon based on the difference between the preference degree sequence of the current login period and the previous login period, and judging whether the user's preferences are consistent based on the change of the preference degree sequence of the previous ten login periods; and so on.

[0039] Another possible embodiment of the present invention is to address the defect that the information obtained by users when using information exchange platforms such as forums due to the information cocoon phenomenon is single or one-sided, and it is difficult to accurately obtain the user's real preferences, thereby making it difficult to provide the best personalized message push method. The embodiment of the present invention introduces a small amount of non-user direct preference or weak preference information in the message push process, records and analyzes the user's operation behavior, such as the number of user browsing times, duration, etc., which produces a large change in the proportion of the original content type and the new content type. Then, based on the user's actual behavior record and the preference characteristics of user groups with similar preferences, predict and analyze the new preference content type that the user may show after breaking the information cocoon, thereby optimizing the system's personalized push strategy for user preferences. By recording and analyzing the user's behavior after breaking the information cocoon, the user's browsing time, frequency, etc. of the new content type are obtained, so as to evaluate the user's preference for the new content, and then compare the user information with similar preference preferences, optimize the personalized push strategy, so as to achieve the best user preference recommendation push, thereby improving the system's push effect and continuously attracting users. The present invention analyzes the user's behavior record based on the AI ​​model, and finally pushes personalized messages to 5G terminal users through the 5G network.

[0040] The embodiment of the present invention records and analyzes the behavior data of users after they break out of the information cocoon in detail, accurately obtains key indicators such as the browsing time and frequency of users for new content types, and evaluates the user's preference for new types. Subsequently, the user information with similar preferences is used for comparative analysis to optimize personalized push content, achieve accurate and efficient user preference recommendation push, improve the system's push effect, and continuously attract and retain users.

[0041] The specific scenario targeted by the embodiments of the present invention is that the system may push non-user-preferred content due to the information cocoon phenomenon, resulting in a decrease in the user's preference for the system. Therefore, by pushing a small amount of weakly related information to the user to analyze the user's preference, and further combining the preference content of the same type of users to comprehensively analyze the user's preference type, personalized push of user messages is achieved. The overall logic of the 5G message personalized push method based on the AI ​​model of the embodiment of the present invention is as follows:

[0042] 1. The system obtains the user's behavior records when using the system.

[0043] The forum system is a platform for users to obtain information and communicate. After logging into the system using a 5G terminal, users can obtain news, hot topics, and content messages that meet their preferences after AI model data processing.

[0044] After the user logs in to the system, any operation performed on various types of system messages will be automatically recorded by the system, including data such as the operation type C, operation duration T, and operation frequency N for the user ID.

[0045] The recorded data will be transmitted to the system backend and then used to create the user's personal behavior record form. Whenever the user operates the system content, the form will be updated in real time according to the user's operation.

[0046] At the same time, the system will generate a daily behavior record form based on the user's daily operations on the system content to facilitate the management and analysis of the user's behavior status.

[0047] 2. Analyze users’ browsing content preferences after breaking out of the information cocoon based on their behavior records.

[0048] a. Obtain the overall changes based on the data records when users operate on different types of data information.

[0049] When users use self-media operation platforms such as forums, mainly information promotion and similar platforms, they will pay more attention to the content they prefer. For example, if a user likes to browse news, the number of times they browse news-type content in their behavior records will be higher, and the browsing time will be relatively longer; in contrast, the number of views and duration of other types of content will be relatively small and short.

[0050] Therefore, by analyzing the browsing time and browsing content of each content type of users, their preference for each content type can be obtained.

[0051]

[0052] Where v is the user's preference coefficient for each content type, i is the number of times the user operates on the content type, N is the number of times the user operates on the module, and T i is the duration of the user's i-th operation on the module, T c is the total time of using the system platform, N c The total number of operations performed by users on all content types in the system. It is the ratio of the number of times a user operates on a content type to the total number of times a user operates on all content types in the system. It is used to describe the user's preference. The larger the value, the higher the preference, and vice versa. The total time that the user operates the module. The total time that users operate the module accounts for the total time that users operate the system, which is used to indicate the user's preference for this type of information.

[0053] The relationship between type, module and system: type refers to the similarity of several types of data, such as several videos are popular science or variety shows. Module refers to the part that users can operate when using the system. A type is a module, and types and modules can be replaced by types. System is a software application system, which includes information types and modules.

[0054] Use the above method to traverse all content types browsed by the user, obtain the user's preference for all content types, and compare the v values, sort the content types C browsed by the user in descending order, and obtain the user's preference sequence. At the same time, record the user's preference coefficient v corresponding to each content type C in the sequence.

[0055] b. Based on the user's preference sequence for different browsing content types, analyze the changing trend of user preferences and screen out user groups with similar preferences in order to deeply study and understand user preferences.

[0056] When users use forums and other platforms to obtain information, they are prone to fall into the phenomenon of information cocoons, whether due to their own preferences or influenced by platform recommendations. If the information cocoon is formed due to the user's own preferences, the content pushed by the system will usually be in line with the user's interests, increasing the number of user views and also attracting users. However, if the information cocoon is mainly caused by the platform's recommendation algorithm, users may be less interested in these push messages. In the long run, this situation may not only fail to achieve the desired effect, but may also cause users' disgust and dissatisfaction.

[0057] Therefore, in order to more accurately analyze user preferences and avoid push messages deviating from users’ real preferences due to the information cocoon phenomenon, it is necessary to re-analyze and evaluate users’ preferences.

[0058] ①According to the changes in the user's preference sequences on adjacent dates, obtain the changes in the user's preference.

[0059] For users who have the phenomenon of information cocoon, the system adds or replaces 5% of the content with low preference levels for the user's recommended content every day, and replaces it with other relevant content types. The system also detects the user's browsing of the newly added content. If the user browses this part of the content frequently or for a long time, then the user likes this part of the content more. Conversely, if the user browses this part of the content less frequently or for a shorter time, then the user has less interest in this part of the content.

[0060] Using the method in step a, the frequency and duration of the user's operations on different content types on the same day are deeply analyzed to construct the user's preference sequence for different content types. If the user shows high-frequency and long-term operation behavior on the new content type, then the content type is ranked high in the preference sequence, and the sequence on the same day will show significant changes compared with the previous day; on the contrary, if the user is not active in operating the new content type, the difference between the preference sequence on the same day and the previous day is small.

[0061] By comparing and analyzing the changes in the preference levels of each content type in the user preference sequence of two consecutive days, we can analyze whether the user is interested in new content. If the difference is small, it means that the user is not very interested in the new content; if the difference is significant, it means that the user has a high interest in the new content, and the user's information cocoon is generated by system recommendations, and the subsequent recommended content strategy needs to be adjusted accordingly.

[0062] Extract the content type information from each sequence, use cosine similarity to obtain the similarity between the user's preference sequences on two consecutive days, and combine the difference between the corresponding values ​​of the preference coefficients of the same content type, that is, the difference between the operation proportions of the user for the same content type on two consecutive days Δf = v t -v y (v t is the user's preference coefficient for content type on that day, v y is the user's preference coefficient for content type on the previous day) as a reference value, and obtain the similarity parameters between the user's operation sequences on two consecutive days:

[0063]

[0064] Where δ is the similarity parameter between the two sequences, z is the sequence number of each content type in the current sequence, n is the sequence length, and d is the cosine value between the current sequence and the historical sequence, which is used to indicate the similarity between the two sequences. The higher the cosine value, the more similar the two sequences are, and vice versa. Δf z is the change difference between the corresponding values ​​of the preference coefficients of the content types in the two sequences. The larger the value, the greater the change in the proportion of users browsing the same content type, and vice versa. It is the sum of the differences between the corresponding values ​​of the preference coefficient of the same content type in the two sequences. The larger the value, the less similar the two sequences are, and vice versa. is the mean of the difference between the preference coefficients of the same content type in the two sequences, which is used to represent the similarity parameter between the two sequences. To ensure that this value is proportional to the similarity between the two sequences, the value is minus one power operation, so that The larger the value, the higher the similarity between sequences, and vice versa. Indicates the similarity between two sequences. The value is normalized by the norm() function to make the value range [0,1].

[0065] Then, the threshold is set to 0.8. When δ≥0.8, the similarity of the browsing contents of the user on two consecutive days is the first similarity level. Otherwise, the similarity of the browsing contents of the user on two consecutive days is the second similarity level.

[0066] ② Analyze the changes in user preferences during the browsing process, obtain the user's new preference sequence, and obtain a stable user group with a similar preference sequence according to the sequence.

[0067] When the similarity of the browsing content of the user on two consecutive days is the second similarity level, that is, there is a significant difference in the browsing of different types of content in the forum system by the user on two consecutive days, in order to ensure that the messages pushed to the user in the future are closer to their personal preferences, it is necessary to re-analyze the user's preferences. In this process, the content that the user really likes may not have been detected, so it is necessary to conduct a deeper analysis of the user's actual preference type. In order to ensure that the content pushed in the future is the content that the user is interested in, the above method is implemented with the help of AI model to identify other user sequences with similar user preference content types in the system, and match other user preference sequences with the user's preference sequence, and obtain the similarity parameter δ between the user preference sequence and the other user preference sequences. If δ≥0.8, it will be classified into the category with similar preference types.

[0068] Among these users, the preference types of some users are still in the analysis stage, while the preference types of others have been analyzed. For user data that is still in the process of changing, its instability may have an adverse impact on the data results. Therefore, it is necessary to use the user data that has been analyzed and has relatively stable preference types as the main object of research.

[0069] To this end, we collected the preference sequences of each user in the past half month, and analyzed these sequences based on the differences between adjacent dates to understand the changes in user preferences between adjacent days. If the user's browsing status changes little between adjacent days, it means that their preference type is relatively stable; otherwise, it means that the user's preference is still in a state of great change.

[0070]

[0071] Where F is the difference index between the preference sequences of each user on two consecutive days, j is the number of the user's browsing days, m is the total number of browsing days, k is the number of the newly added content types detected by the user, l is the number of newly added content types, and v j,k is the user's preference coefficient for the newly added content type k on day j, is the difference between the preference coefficients of users for the same content type on two consecutive days. The closer the value is to 1, the more stable the user's preference type area is; otherwise, it is still in the analysis process; δ j+1 is the similarity parameter between the two sequences of the types of content browsed by the user on the next day and the previous day in two consecutive days. The larger the value, the greater the difference in the types browsed by the user on two consecutive days. Conversely, the smaller the difference. To ensure that the various parts in the formula are positively correlated, we obtain δ j+1 The reciprocal of .

[0072] The threshold is set to 0.3. When F≤0.3, the browsing behavior of this type of users for the newly added content type detected by the users tends to be stable, otherwise it is still in the process of change.

[0073] c. Obtain information types with a high number of views other than the newly added types from a user group with similar preferences and stable browsing behavior. When conducting analysis, it is necessary to reduce the impact of the newly added types, so the newly added types should be excluded and the original information types should be used for analysis.

[0074] Through the previous step, we can obtain a user group with similar preferences and stable browsing behavior. In order to better understand the real preferences of users who break out of the information cocoon and ensure that the pushed content is as close to the user's preferences as possible, we need to refer to the content types that are stably browsed by users with similar preferences to the users who break out of the information cocoon at this time, push information of this part of content types to users, and observe their browsing behavior.

[0075] The content types browsed by each user are obtained and sorted out, and the browsing times of each user are sorted according to the content type. If there are content types with the same browsing times, they are further sorted according to the browsing time.

[0076] At this time, the number of views and viewing duration of each content type in the sequence are obtained, and the popularity coefficient is determined by the size of the number of views and viewing duration of each content type information, and then the most popular content type is filtered out to determine the content type pushed to users to break out of the information cocoon.

[0077] P = max(a c ,b c );

[0078] Where P is the popularity coefficient of content type C among the screened users, a c , b c are the number of users who browsed content type C and the total browsing time in the sorted sequence, respectively. c ,b c) The maximum value of the product of the number of users browsing the sequence and the total browsing time is filtered out as the type of information that needs to be pushed.

[0079] 3. Push personalized messages to users based on their preferences after their behavior records change.

[0080] The system uses the function management module to transmit the operation records of users using 5G terminals to the background database using efficient and secure 5G network equipment, and saves them in the user's corresponding personal behavior record table in the database. When analyzing the user's operation behavior, the AI ​​model starts running and retrieves all recent behavior record forms of the user from the system background database through the user ID to obtain the user's behavior data.

[0081] Next, the AI ​​model analyzes the user's behavior data to detect whether there are users who have broken the information cocoon phenomenon, and analyzes the changes in the types of information that users prefer. In response to the changes in the types of information that users prefer, the AI ​​model retrieves highly similar users in the database by comparing the similarities of the preference types. After executing the operation in step 2 (b), the AI ​​model further screens out users with stable preference information types, and obtains the popularity coefficient of each information type operated by these users in their overall operation information types through the method in step (c).

[0082] Then, the AI ​​model sorts the popularity coefficients of each information type by size, creates a popularity coefficient table, and passes the table to the personalized message push module. After receiving the data table passed by the AI ​​model, the message push module pushes personalized 5G messages to users through 5G base stations based on the top five popular information types in the table, and records the user's operations on these contents.

[0083] Finally, when the user logs in next time, the system will replace the last 5% of the content in the user's favorite list with the top 5 information types in the popularity coefficient form in the recommended content, and record the user's operation.

[0084] This process will be repeated until the user's favorite content type is obtained, thereby achieving accurate push of personalized messages to the user.

[0085] Based on the same inventive concept as the above method embodiment, the embodiment of the present invention also provides a 5G message personalized push system based on an AI model, such as Figure 2 As shown, the push system includes:

[0086] The behavior record module is used to obtain the user's behavior record when using the system; the behavior record includes the data record when the user operates on data information of different content types.

[0087] The preference analysis module is used to analyze the user's browsing content preferences after breaking out of the information cocoon based on the user's behavior records; to obtain the user's preference degree sequence for different content types based on the data records when the user operates on data information of different content types; to analyze the changing trend of user preferences based on the user's preference degree sequence for different content types, and to screen out user groups with similar preferences; and to obtain the user's browsing content preferences after breaking out of the information cocoon from the user group with similar and consistent preferences.

[0088] The message push module is used to push personalized messages to users based on their browsing content preferences after breaking out of the information cocoon.

[0089] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0090] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A personalized 5G message push method based on an AI model, characterized in that: The push method includes: Step S1, obtaining a user's behavior record in the process of using the system; the behavior record includes a data record when the user operates data information of different content types; Step S2, analyzing the user's browsing content preferences after breaking out of the information cocoon based on the user's behavior records; According to the data records when the user operates on the data information of different content types, a sequence of the user's preference for the different content types browsed is obtained; according to the sequence of the user's preference for the different content types browsed, the change trend of the user's preference is analyzed, and user groups with similar preferences are screened out; from the user groups with similar preferences, user groups with similar preferences and consistent preferences are screened out; from the user groups with similar preferences and consistent preferences, the browsing content preferences of the users after breaking out of the information cocoon are obtained; Step S3: Push personalized messages to users based on their browsing content preferences after breaking out of the information cocoon.

2. The 5G message personalized push method based on the AI ​​model according to claim 1 is characterized in that: According to the sequence of users' preferences for different types of content they browse, we analyze the changing trends of users' preferences and filter out user groups with similar preferences, including: According to the changes between the preference degree sequences of the user on adjacent dates, the changes in the user's preference are obtained; Analyze the changes in users' preferences within a certain period of time to obtain the users' new preference sequence; according to the users' new preference sequence, obtain user groups with similar preferences.

3. The 5G message personalized push method based on the AI ​​model according to claim 2 is characterized in that: According to the data records when the user operates on the data information of different content types, the user's preference degree sequence for browsing different content types is obtained, which specifically includes: The data records include: browsing time, browsing times and browsing content; According to the browsing time, browsing times and browsing contents of users when operating on different content types, the user's preference coefficient for each content type is obtained; According to the size of the preference coefficient, the content types browsed by the user are sorted to obtain a sequence of the user's preference levels for the different content types browsed.

4. The 5G message personalized push method based on the AI ​​model according to claim 3 is characterized in that: According to the changes between the preference degree sequences of the user on adjacent dates, the changes in the user's preference are obtained, specifically including: Generate the user's previous day preference content based on the top 95% of the user's preferences in the previous day's preference sequence; Pushing the content of the day to the user; the content of the day includes the preferred content of the previous day and related content, the related content is different from the preferred content of the previous day, and the related content accounts for 5% of the content of the day; According to the difference between the user's preference sequence of the current day and the preference sequence of the previous day, the similarity parameter between the preference sequences of two adjacent days is obtained; The similarity parameter between the preference degree sequences of two adjacent days is used to represent the change of the user's preference.

5. The 5G message personalized push method based on the AI ​​model according to claim 4 is characterized in that: The user's preference change includes browsing content similarity being at a first similarity level and browsing content similarity being at a second similarity level; When the similarity parameter between the preference sequences of two adjacent days is greater than or equal to the first threshold, it indicates that the similarity between the browsing contents of the user on two adjacent days is at the first similarity level; when the similarity parameter between the preference sequences of two adjacent days is less than the first threshold, it indicates that the similarity between the browsing contents of the user on two adjacent days is at the second similarity level.

6. The 5G message personalized push method based on the AI ​​model according to claim 5 is characterized in that: Analyze the changes in user preferences within a certain period of time to obtain a new sequence of user preferences; according to the new sequence of user preferences, obtain user groups with similar preferences, including: When the similarity of the browsing contents of a certain user on two consecutive days is at the second similarity level, the similarity parameter between the preference sequence of the user and the preference sequence of other users is obtained according to the difference between the preference sequence of the user and the preference sequence of other users; Other users whose similarity parameters between the user and other users' preference sequences are greater than a second threshold are identified as a user group with similar preferences.

7. The 5G message personalized push method based on the AI ​​model according to claim 6 is characterized in that: From the user groups with similar preferences, user groups with similar preferences and consistent preferences are screened out, specifically including: Analyze the preference degree sequence of different content types browsed by each user in the similar preference user group within the current preset number of days, and the similarity degree parameters between the preference degree sequences of each user in two consecutive days, and obtain the difference index between the browsing record sequences of each user in two consecutive days; The users whose difference index between the browsing record sequences of two consecutive days is less than or equal to the third threshold are identified as a user group with similar and consistent preferences.

8. The 5G message personalized push method based on the AI ​​model according to any one of claims 4 to 7, characterized in that: From the user groups with similar and consistent preferences, we can get the browsing content preferences of users after breaking out of the information cocoon, including: Analyze the data records of user groups with similar and consistent preferences when operating data information of different content types, and obtain the total number of views and total browsing time of each content type; Based on the total number of views and total browsing time of each content type, the popularity coefficient of each content type among the user groups with similar and consistent preferences is obtained, and then the browsing content preferences of users after breaking out of the information cocoon are obtained.

9. The 5G message personalized push method based on the AI ​​model according to claim 8 is characterized in that: According to the user's browsing content preferences after breaking out of the information cocoon, personalized message push is sent to the user, including: According to the popularity coefficient of each content type among the user groups with similar and consistent preferences, the content types are sorted by size to form a popularity coefficient table; When the user uses the system next time, the associated content is replaced by the content in the popularity coefficient table whose popularity coefficient is greater than the fourth threshold.

10. The 5G message personalized push system based on AI model is characterized by: The push system comprises: A behavior record module is used to obtain the behavior record of the user in the process of using the system; the behavior record includes the data record when the user operates on data information of different content types; The preference analysis module is used to analyze the browsing content preferences of users after breaking out of the information cocoon based on the user's behavior records; to obtain the user's preference degree sequence for different types of content browsed based on the data records when the user operates on data information of different content types; to analyze the changing trend of user preferences based on the user's preference degree sequence for different types of content browsed, and to screen out user groups with similar preferences; to screen out user groups with similar preferences and consistent preferences from the user groups with similar preferences; to obtain the browsing content preferences of users after breaking out of the information cocoon from the user groups with similar preferences and consistent preferences; The message push module is used to push personalized messages to users based on their browsing content preferences after breaking out of the information cocoon.

Images