An AIGC-driven automatic personalized content push system and method
Through the AIGC-driven personalized push system, user behavior data is analyzed and related analysis and sorting of recommended contents, the problem of existing recommendation systems ignoring content relevance is solved, and more rich and flexible personalized recommendations are achieved, improving the effect of improving User satisfaction and recommendation accuracy.
Patent Information
- Application Number
- CN202411277910.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-12
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-12
AI Technical Summary
When existing personalized recommendation systems process large-scale data, they often ignore the correlation between recommended content, resulting in too single recommendation results and lack flexibility.
AIGC-driven content automatic personalized push system is adopted to analyze user behavior data, generate user preference tags, sort according to browsing information, conduct correlation analysis of recommended content, reorder push list, and skip the association recommendations related to the current browsing content Content and optimize the push list based on user feedback.
Improve user satisfaction and recommendation accuracy, ensure the relevance and diversity of recommended content, avoid user fatigue, and achieve efficient personalized content push.
Smart Images

Figure CN118827774B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of data push, and in particular relates to an AIGC-driven automatic personalized content push system and method. Background Art
[0002] With the rapid development of Internet technology, the way people obtain information has shifted from traditional media to digital and networked information platforms. This shift has brought about an explosive growth of information. Users often feel overwhelmed when faced with massive amounts of information and find it difficult to quickly find content that they are really interested in. Therefore, how to filter out valuable information from huge amounts of data and effectively push it to users has become an important topic in the current field of information technology. In order to solve this problem, personalized recommendation systems have emerged. Personalized recommendation systems recommend content that users may be interested in by analyzing users' historical behaviors, preferences, and social network information.
[0003] However, the existing personalized recommendation systems still have some shortcomings. Most of the current recommendation methods are based on user preferences to generate recommendation lists. However, most recommendation algorithms tend to ignore the correlation between recommended content when processing large-scale data. The results are too procedural, resulting in recommendation results that only consider user preferences. But in fact, the recommendation list contains more content, and there is a certain correlation between some of the content. Therefore, if these correlations are not considered, the recommendation results will often be too single and lack corresponding flexibility. Based on this, this solution provides an AIGC-driven content automatic personalized push method to solve the above problems. Summary of the invention
[0004] The purpose of the present invention is to provide an AIGC-driven automatic personalized content push system and method, which can provide richer and more flexible personalized content push during user browsing, thereby improving user satisfaction and the accuracy of system recommendations.
[0005] The technical solution adopted by the present invention is as follows:
[0006] An AIGC-driven automatic personalized content push method, comprising:
[0007] Based on the acquired user behavior data, classification processing is performed to obtain multiple user preference tags;
[0008] Collecting browsing information of users under each of the user preference tags, and sorting the user preference tags according to the browsing information to obtain an initialization push list;
[0009] Matching recommended content according to the initialization push list, and performing correlation analysis on each of the recommended content to obtain associated recommended content and non-associated recommended content;
[0010] Reordering the initial push list according to the related recommended content and the non-related recommended content to obtain a personalized push list, wherein when pushing according to the personalized push list, it is possible to choose to skip the related recommended content related to the currently browsed content;
[0011] Based on the personalized push list, personalized push is performed to the user, and according to the user feedback information under the personalized push, the personalized push list is optimized to obtain an updated push list, and personalized push is continued to be performed according to the updated push list.
[0012] In a preferred solution, the step of performing classification processing based on the acquired user behavior data to obtain multiple user preference tags includes:
[0013] Establishing a historical monitoring period and collecting user behavior data within the historical monitoring period, wherein the user behavior data includes historical browsing content and historical search records;
[0014] Preprocessing the historical browsing content and historical search records, wherein the preprocessing includes removing invalid data and abnormal values;
[0015] Extracting feature information of the preprocessed historical browsing content and historical search records, and assigning user preference tags to the historical browsing content and historical search records under the same feature information;
[0016] Among them, all the user preference tags are unique.
[0017] In a preferred solution, the step of collecting browsing information of users under each of the user preference tags, and sorting the user preference tags according to the browsing information to obtain an initial push list includes:
[0018] Obtaining browsing information under the user preference tag during the historical monitoring period, wherein the browsing information includes the user's browsing completion degree and browsing times for historical browsing content;
[0019] Performing a weighted score on each user preference tag according to the browsing completion degree and browsing times to obtain a push score for each user preference tag;
[0020] The push scores are sorted in descending order, and the push order of each user preference tag is output one by one according to the sorting result of the push scores, and simultaneously summarized into an initialization push list.
[0021] In a preferred solution, the step of weighting and scoring each user preference tag according to the browsing completion degree and the browsing times to obtain the push score of each user preference tag includes:
[0022] Setting a plurality of sample periods with equal time intervals within the historical monitoring period;
[0023] Collecting the page views of all the historical browsing contents and the page view time of each of the historical browsing contents within each of the sample time periods;
[0024] Determine the distribution weights of the browsing completion degree and the number of browsing times according to the browsing volume and browsing time of the historical browsing content;
[0025] A weighted function is obtained, and the browsing completion degree and the number of browsing times, as well as the corresponding allocation weights are input into the weighted function for weighted calculation to obtain a push score for each of the user preference tags.
[0026] In a preferred solution, the step of determining the distribution weights of the browsing completion degree and the number of browsing times according to the browsing volume and browsing time of the historical browsing content includes:
[0027] Calculate the unit page views in each sample period according to the page views and page views time of the historical page views in each sample period;
[0028] Arrange the unit page views in each sample period according to the chronological order of occurrence;
[0029] Obtaining a standard offset, and performing an offset process on each of the unit views according to the standard offset to obtain a plurality of clustering intervals;
[0030] Counting the number of unit views in each clustering interval and recording it as a screening condition parameter, and marking the clustering interval under the screening condition parameter with the largest value as a standard interval;
[0031] The unit page views in the standard interval are summed and averaged to obtain the standard unit page views in the historical monitoring period;
[0032] A weight allocation function is obtained, and the standard unit browsing volume is input into the weight allocation function, and the output result of the weight allocation function is recorded as the allocation weight of the browsing completion degree and the browsing number.
[0033] In a preferred solution, the step of performing correlation analysis on each of the recommended contents to obtain related recommended contents and non-related recommended contents includes:
[0034] Acquire all the recommended contents, and perform feature extraction to obtain key features of each of the recommended contents;
[0035] Performing vector conversion on each of the key features to obtain a key feature vector of each of the recommended contents;
[0036] Matching the recommended contents one by one, inputting the key feature vectors of the recommended contents after the pairwise matching into a preset association calculation function, and recording the output result of the association calculation function as an association score;
[0037] Obtaining a correlation evaluation threshold, and comparing the correlation score with the correlation evaluation threshold;
[0038] When the correlation score is greater than the correlation evaluation threshold, it indicates that there is a correlation between the recommended contents after pairwise matching, and they are simultaneously recorded as related recommended contents;
[0039] When the correlation score is less than or equal to the correlation evaluation threshold, it indicates that there is no correlation between the recommended contents after pairwise matching, and they are simultaneously recorded as non-correlated recommended contents.
[0040] In a preferred solution, the step of reordering the initialized push list according to the associated recommended content and the non-associated recommended content to obtain a personalized push list includes:
[0041] Acquire the initialization push list, and record adjacent recommended contents in the initialization push list as pre-processed data and post-processed data respectively;
[0042] Collecting the correlation score between the associated recommended content of the pre-processed data and the pre-processed data, converting the correlation score to the same dimension as the push score, and recording it as a condition parameter;
[0043] Performing weighted sum processing on the condition parameter and the push score of the associated recommended content of the previous data to be processed to obtain the associated push score;
[0044] Compare the associated push score with the push score of the subsequent data to be processed, and when the associated push score is greater than the push score of the subsequent data to be processed, adjust the associated recommended content corresponding to the associated push score to the front of the subsequent data to be processed, otherwise, continue to maintain the original order;
[0045] Repeat the processing of all adjacent recommended contents in the initialization push list until the entire initialization push list is reordered and output as a personalized push list.
[0046] In a preferred solution, the step of optimizing the personalized push list according to user feedback information under personalized push to obtain an updated push list includes:
[0047] Obtain user feedback and satisfaction with each recommended content under personalized push;
[0048] Convert the feedback satisfaction into an optimized weight of a push score corresponding to each recommended content;
[0049] Obtain an optimization function, and input the push score and the corresponding optimization weight corresponding to each recommended content into the optimization function for calculation to obtain an updated push score;
[0050] The personalized push list is reordered according to the update push score to obtain an update push list, and the update push list is applied to execute push processing of each recommended content.
[0051] The present invention also provides an AIGC-driven automatic personalized content push system, using the above-mentioned AIGC-driven automatic personalized content push method, including:
[0052] A data classification module, wherein the data classification module is used to perform classification processing based on the acquired user behavior data to obtain multiple user preference tags;
[0053] An initialization module, the initialization module is used to collect browsing information of users under each of the user preference tags, and sort the user preference tags according to the browsing information to obtain an initialization push list;
[0054] A correlation analysis module, the correlation analysis module is used to match the recommended content according to the initialization push list, and perform correlation analysis on each of the recommended content to obtain related recommended content and non-related recommended content;
[0055] A personalized push module, the personalized push module is used to reorder the initialized push list according to the associated recommended content and the non-associated recommended content to obtain a personalized push list;
[0056] A list update module, wherein the list update module is used to perform personalized push to users based on the personalized push list, and optimize the personalized push list according to user feedback information under the personalized push to obtain an updated push list, and continue to perform personalized push according to the updated push list.
[0057] And, an electronic device, the electronic device comprising:
[0058] at least one processor;
[0059] and a memory communicatively coupled to the at least one processor;
[0060] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the above-mentioned AIGC-driven automatic personalized content push method.
[0061] The technical effects achieved by the present invention are:
[0062] The present invention realizes personalized content push and improves user satisfaction and participation by analyzing user behavior data. Specifically, through correlation analysis, it can ensure the relevance and diversity of recommended content, enrich users' browsing content, avoid user fatigue, and improve user satisfaction with pushed content. At the same time, according to user feedback, the personalized push list can be dynamically adjusted accordingly, the push effect of recommended content can be continuously improved, and automatic learning and adaptation to changes in user preferences can be achieved to realize efficient personalized content push. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] Figure 1 It is a schematic flow chart of the method of the present invention;
[0064] Figure 2 It is a schematic diagram of the system module of the present invention;
[0065] Figure 3 It is a schematic diagram of the structure of an electronic device of the present invention. DETAILED DESCRIPTION
[0066] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.
[0067] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0068] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" that appears in different places in this specification does not refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.
[0069] See also Figure 1 As shown, the present invention provides an AIGC-driven automatic personalized content push method, comprising:
[0070] S1. Based on the acquired user behavior data, classification processing is performed to obtain multiple user preference tags;
[0071] S2. Collect browsing information of users under each user preference tag, and sort the user preference tags according to the browsing information to obtain an initial push list;
[0072] S3, matching recommended content according to the initialization push list, and performing correlation analysis on each recommended content to obtain related recommended content and non-related recommended content;
[0073] S4, reordering the initial push list according to the associated recommended content and the non-associated recommended content to obtain a personalized push list, wherein when pushing according to the personalized push list, it is possible to choose to skip the associated recommended content related to the currently browsed content;
[0074] S5. Execute personalized push to the user based on the personalized push list, optimize the personalized push list according to the user feedback information under the personalized push, obtain an updated push list, and continue to execute personalized push according to the updated push list.
[0075] As described in the above steps S1-S5, with the development of big data and artificial intelligence technology, personalized push can more accurately meet the needs of users and improve user experience and satisfaction. In this embodiment, the user's online behavior data will be collected and analyzed first. By classifying the user behavior data, multiple preference tags of the user can be identified to reflect the user's interests and needs. Next, the browsing information of the corresponding user will be collected for each user preference tag, and the user preference tags will be sorted according to the browsing information to generate an initialization push list. The initialization push list is preliminarily sorted according to the user's historical behavior and preferences in order to better meet the user's initial needs. Then, the recommended content will be matched according to the initialization push list. The recommended content can be in various forms such as articles, videos, pictures, etc., and then each push list will be sorted. The recommended content is analyzed for relevance, so that the related recommended content and the non-related recommended content can be distinguished. After that, the initialized push list will be reordered according to the related recommended content and the non-related recommended content. In this way, the related recommended content can be displayed to the user first, thereby improving the user's satisfaction with the pushed content and the click-through rate. The personalized push list obtained after the reordering can better meet the user's personalized needs. Finally, personalized push will be performed to the user based on the personalized push list. After receiving the pushed content, the user will give feedback information (such as a rating system) according to his or her interests and needs. The personalized push list can be optimized based on the feedback information to obtain an updated push list. The updated push list will be continuously optimized to ensure that the user's personalized needs can be met and the content can be accurately matched with the user's needs.
[0076] In a preferred embodiment, the step of performing classification processing based on the acquired user behavior data to obtain multiple user preference tags includes:
[0077] S101, constructing a historical monitoring period, and collecting user behavior data within the historical monitoring period, where the user behavior data includes historical browsing content and historical search records;
[0078] S102, preprocessing the historical browsing content and historical search records, the preprocessing including removing invalid data and abnormal values;
[0079] S103, extracting characteristic information of the preprocessed historical browsing content and historical search records, and assigning user preference tags to the historical browsing content and historical search records under the same characteristic information;
[0080] Among them, all user preference tags are unique.
[0081] As in the above steps S101-S103, when outputting the user preference label, a historical monitoring period is first constructed. The historical monitoring period is a time period connecting the current moment and can reflect the user's short-term behavior. During the historical monitoring period, user behavior data will be collected. The user behavior data mainly includes the user's historical browsing content and historical search records. Through the user's historical browsing content and historical search records, the user's interests and needs can be understood. Then the historical browsing content and historical search records will be preprocessed. The purpose of preprocessing is to ensure the quality of the data in order to conduct a more accurate analysis. The preprocessing step includes removing invalid data and outliers to exclude irrelevant or erroneous information that may interfere with the analysis results, which can ensure the accuracy and reliability of subsequent analysis. After the preprocessing is completed, the feature information in the historical browsing content and historical search records will be extracted. These feature information are key indicators of user behavior and can reflect the user's preferences and interests. In order to better integrate the feature information, the historical browsing content and historical search records under the same feature information are assigned corresponding user preference labels. At the same time, it is also necessary to ensure the uniqueness of all user preference labels, so that each user preference label represents a unique preference to avoid duplication or confusion, and ensure the accuracy of classification and the effectiveness of subsequent analysis.
[0082] In a preferred embodiment, the steps of collecting browsing information of users under each user preference tag, sorting the user preference tags according to the browsing information, and obtaining an initial push list include:
[0083] S201, obtaining browsing information under a user preference tag during a historical monitoring period, where the browsing information includes the browsing completion degree and browsing times of the user's historical browsing content;
[0084] S202, weighted scoring is performed on each user preference tag according to the browsing completion degree and the browsing times, so as to obtain a push score for each user preference tag;
[0085] S203: sorting the push scores in descending order, and outputting the push order of each user preference tag one by one according to the sorting result of the push scores, and synchronously summarizing them into an initialization push list.
[0086] As described in S201-S203 above, in order to better meet the needs of users, after the user preference tags are determined, it is first necessary to collect the browsing information of each user under different user preference tags. The browsing information includes the browsing completion and browsing times of various pushed contents by the user during the historical monitoring period, so as to reflect the user's interest and preference for different contents. Then, each user preference tag is weighted and scored according to the browsing completion and browsing times of the historical browsing contents by the user, so as to evaluate the push score of each user preference tag. After the push score is output, the push score of each user preference tag will be sorted. The sorting is based on the size of the push score value, specifically, it is arranged in order from large to small. Then, according to the sorting result of the push score, the push order of each user preference tag is output one by one, and synchronously summarized into an initialized push list. Provide a corresponding basis for the subsequent generation of personalized push lists.
[0087] In a preferred embodiment, the step of weighting and scoring each user preference tag according to the browsing completion degree and the browsing times to obtain the push score of each user preference tag includes:
[0088] Set up multiple sample periods with equal time intervals within the historical monitoring period;
[0089] Collect the page views of all historical browsing contents and the browsing time of each historical browsing content in each sample period;
[0090] Determine the distribution weight of browsing completion and browsing times based on the browsing volume and browsing time of historical browsing content;
[0091] A weighted function is obtained, and the browsing completion degree and the number of browsing times, as well as the corresponding distribution weights are input into the weighted function for weighted calculation to obtain the push score of each user preference tag.
[0092] As mentioned above, when calculating the push score, firstly, multiple sample periods with equal time intervals are set within the historical monitoring period, so as to better capture the browsing behavior of users in different time periods. Next, the page views of all historical browsing contents and the browsing time of each historical browsing content within the sample period will be collected. The page views can reflect the popularity of a certain content, and the browsing time can reveal the user's depth of the content. Then, the distribution weights of browsing completion and number of views will be determined based on the page views and browsing time of the historical browsing content. For example, if a user spends more time on a certain historical browsing content, then the browsing completion of this content will be given a higher weight. Similarly, the number of views of historical browsing content with a large number of views will also be given a higher weight. Finally, a preset weighting function is introduced, and the browsing completion and number of views, as well as the corresponding distribution weights are input into the weighting function for weighted calculation, so that the push score of each user preference tag can be obtained. The expression of the weighting function is:
[0093] ;
[0094] In the formula, Indicates push rating. and They represent the distribution weights of browsing completion and browsing times, Indicates browsing completion. Indicates the number of views. After the push score is output, the push list can be initialized.
[0095] In a preferred embodiment, the step of determining the distribution weights of browsing completion and browsing times according to the browsing volume and browsing time of historical browsing content includes:
[0096] Step 1. Calculate the unit page views in each sample period based on the page views and page views time of the historical page views in each sample period.
[0097] Step 2: Arrange the unit page views in each sample period according to the chronological order of occurrence;
[0098] Step 3, obtain the standard offset, and perform offset processing on each unit of browsing volume according to the standard offset to obtain multiple clustering intervals;
[0099] Step 4. Count the number of unit views in each clustering interval and record it as the filtering condition parameter, and mark the clustering interval under the filtering condition parameter with the largest value as the standard interval;
[0100] Step 5, sum and average the unit page views in the standard interval to obtain the standard unit page views in the historical monitoring period;
[0101] Step 6. Obtain a weight distribution function, input the standard unit page views into the weight distribution function, and record the output of the weight distribution function as the distribution weights of the page view completion and page view count.
[0102] As described in the above steps Step 1-Step 6, in order to determine the distribution weights of browsing completion and browsing times, it is necessary to perform corresponding analysis based on the browsing volume and browsing time of historical browsing content. First, it is necessary to collect the browsing volume and browsing time of historical browsing content in each sample period, and then calculate the unit browsing volume in each sample period. The unit browsing volume refers to the ratio of the browsing volume to the corresponding browsing volume in the browsing time, which can reflect the user's preference for the historical browsing content in the sample period. Then, the unit browsing volume in each sample period is arranged in the order of occurrence, and a preset standard offset is introduced. The standard offset is determined based on the statistical characteristics of historical data and is used to offset each unit browsing volume. Specifically, the unit browsing volume can be The page views are moved to the right or left by a certain value to form multiple clustering intervals. After obtaining multiple clustering intervals, it is necessary to count the number of unit page views in each clustering interval. The number of unit page views in each clustering interval will be recorded as the filtering condition parameter, and then the filtering condition parameter with the largest value is found, and the corresponding clustering interval is calibrated as the standard interval. The standard interval represents the user's habitual browsing behavior, and then the unit page views in the standard interval are summed and averaged. By summing and averaging the unit page views in the standard interval, the standard unit page views in the historical monitoring period can be obtained, and then the weight distribution function is introduced, and the standard unit page views are input into the weight distribution function, where the expression of the weight distribution function is:
[0103] ;
[0104] In the formula, Indicates the distribution weight of browsing completion and browsing times, =1,2, , Represents constant parameters related to browsing completion. , Represents constant parameters related to the number of views, It represents the standard unit of page views. By calculating the weight distribution function, the distribution weights of page view completion and page view times can be obtained.
[0105] In a preferred embodiment, the step of performing correlation analysis on each recommended content to obtain related recommended content and non-related recommended content includes:
[0106] S301, obtaining all recommended contents, and performing feature extraction to obtain key features of each recommended content;
[0107] S302, performing vector conversion on each key feature to obtain a key feature vector of each recommended content;
[0108] S303, matching the recommended contents one by one, and inputting the key feature vectors of the recommended contents after the pairwise matching into a preset association calculation function, and recording the output result of the association calculation function as an association score;
[0109] S304, obtaining a correlation evaluation threshold, and comparing the correlation score with the correlation evaluation threshold;
[0110] When the correlation score is greater than the correlation evaluation threshold, it indicates that there is a correlation between the recommended contents after pairwise matching, and they are simultaneously recorded as related recommended contents;
[0111] When the correlation score is less than or equal to the correlation evaluation threshold, it indicates that there is no correlation between the recommended contents after pairwise matching, and they are simultaneously recorded as non-correlated recommended contents.
[0112] As described in the above steps S301-S304, when determining the associated recommended content and the non-associated recommended content, it is first necessary to obtain all the recommended content and perform corresponding feature extraction to identify the core features of each recommended content, that is, the key features, which can be achieved through technical means such as text mining and natural language processing. Then, the key features are converted into vectors to facilitate subsequent mathematical calculations. The vector conversion can use methods such as TF-IDF and Word2Vec. Next, the recommended contents are matched one by one, and the similarities and differences between different recommended contents are compared. Specifically, the key feature vectors of each pair of matched recommended contents are input into the preset association measurement function. The association measurement function will calculate an association score based on the similarity of the feature vectors. The expression of the association measurement function is:
[0113] ;
[0114] In the formula, represents the association score, , are constant parameters, represents the number of feature points of the key feature vector, , Represents the key feature vector of the recommended content after pairwise matching, Indicates the number of times the recommended content appears at the same time after pairwise matching. Represents the total number of occurrences of the recommended content. After the correlation score is output, a preset correlation evaluation threshold is introduced. The correlation evaluation threshold is a pre-set standard used to determine whether the correlation score is high enough to indicate that there is a correlation between the two recommended contents. Specifically, when the correlation score is greater than or equal to the correlation evaluation threshold, it indicates that there is a certain correlation between the recommended contents after the two are matched. In this case, the pair of recommended contents will be recorded as related recommended contents. On the contrary, if the correlation score is less than or equal to the correlation evaluation threshold, it is considered that there is no significant correlation between the recommended contents after the two are matched. In this case, the pair of recommended contents will be recorded as unrelated recommended contents.
[0115] In a preferred embodiment, the step of reordering the initial push list according to the associated recommended content and the non-associated recommended content to obtain a personalized push list includes:
[0116] S401, obtaining an initialization push list, and recording adjacent recommended contents in the initialization push list as pre-processed data and post-processed data respectively;
[0117] S402, collecting the correlation score between the related recommended content of the previous data to be processed and the previous data to be processed, converting the correlation score to the same dimension as the push score, and recording it as a condition parameter;
[0118] S403, performing weighted sum processing on the condition parameter and the push score of the associated recommended content of the previous data to be processed to obtain the associated push score;
[0119] S404, comparing the associated push score with the push score of the subsequent data to be processed, and when the associated push score is greater than the push score of the subsequent data to be processed, adjusting the associated recommended content corresponding to the associated push score to the front of the subsequent data to be processed, otherwise, continue to maintain the original order;
[0120] S405: Repeat processing of all adjacent recommended contents in the initialization push list until the entire initialization push list is reordered and output as a personalized push list.
[0121] As described in the above steps S401-S405, after the associated recommended content and the non-associated recommended content are output, the adjacent recommended content in the initialization push list is first marked as the preceding data to be processed and the following data to be processed, respectively, and then the associated recommended content of the preceding data to be processed and the associated score between the associated recommended content of the preceding data to be processed and the preceding data to be processed are collected. At the same time, in order to ensure the consistency of the associated score and the push score, the associated score needs to be converted to the same dimension as the push score and recorded as a conditional parameter for subsequent processing, and then the conditional parameter and the push score of the associated recommended content of the preceding data to be processed are weighted and summed. Processing, so as to obtain a comprehensive correlation push score, and then compare the correlation push score with the push score of the post-processed data. If the correlation push score is higher than the push score of the post-processed data, then the correlation recommended content corresponding to the correlation push score is adjusted to the front of the post-processed data. On the contrary, if the correlation push score is lower, the original order will be kept unchanged. Repeat the above steps to process all adjacent recommended contents in the initialization push list until the entire list is re-sorted. After sorting, the final personalized push list is output, thereby providing users with more accurate and personalized recommendations. When users view recommended content, they can have a certain degree of continuity. For example, when the recommended content is information of category a, information related to information of category a can be recommended immediately. For example, when browsing videos, a certain video is recommended according to user preferences. Then, when browsing the video, the subsequent content with correlation with the video will be recommended first without additional search. Since the recommended content itself is pushed by user preferences, this method makes the continuity of recommended content better.
[0122] Among them, when pushing according to the personalized push list, you can choose to skip the related recommended content related to the current browsing content. The personalized push list re-determined based on the related push score, in addition to ensuring the continuity of the recommended content, this method should also have a report skip function. Through this skip function, when the current browsing content pushed according to the personalized push list does not meet or has subjective viewing singleness, you can also choose to directly skip the recommended content associated with the current browsing content through the skip function. This method makes the personalized push list meet the user's needs.
[0123] In a preferred embodiment, the personalized push list is optimized according to the user feedback information under the personalized push, and the step of obtaining the updated push list includes:
[0124] S501, obtaining user feedback satisfaction for each recommended content under personalized push;
[0125] S502, converting the feedback satisfaction into an optimization weight of a push score corresponding to each recommended content;
[0126] S503, obtaining an optimization function, and inputting the push score and the corresponding optimization weight corresponding to each recommended content into the optimization function for calculation to obtain an updated push score;
[0127] S504: re-sort the personalized push list according to the update push score to obtain an update push list, and apply the update push list to execute push processing of each recommended content.
[0128] As described in the above steps S501-S504, in order to further improve the user experience, it is necessary to optimize the push list according to the user's feedback information in the personalized push. First, it is necessary to collect the user's feedback satisfaction data on each recommended content in the personalized push. This embodiment is achieved through user rating, so as to understand the user's satisfaction with each recommended content. Secondly, the collected feedback satisfaction data is converted into the optimization weight of the push score corresponding to each recommended content. The purpose is to convert the user's subjective satisfaction into a quantifiable optimization indicator. For example, a scoring standard can be set to divide the user's satisfaction into several levels, and different weight values are assigned to each level. Then, an optimization function is introduced, and the push score corresponding to each recommended content and the corresponding optimization weight are input into the optimization function for calculation. The expression of the optimization function is: update push score = optimization weight × push score. Finally, the personalized push list is re-sorted according to the update push score, and the recommended content is arranged from high to low according to the user satisfaction. In this way, an update push list can be obtained, and the update push list is applied to the actual personalized push processing to ensure that the recommended content received by the user is more in line with their interests and needs.
[0129] See also Figure 2 , an AIGC-driven automatic personalized content push system, using the above-mentioned AIGC-driven automatic personalized content push method, comprising:
[0130] A data classification module is used to perform classification processing based on the acquired user behavior data to obtain multiple user preference labels;
[0131] Initialization module, which is used to collect browsing information of users under each user preference tag, and sort the user preference tags according to the browsing information to obtain an initialization push list;
[0132] A correlation analysis module is used to match recommended content according to the initial push list, and to perform correlation analysis on each recommended content to obtain related recommended content and non-related recommended content;
[0133] A personalized push module, which is used to reorder the initial push list according to the associated recommended content and the non-associated recommended content to obtain a personalized push list;
[0134] The list update module is used to perform personalized push to users based on the personalized push list, and optimize the personalized push list according to user feedback information under the personalized push, obtain an updated push list, and continue to perform personalized push according to the updated push list.
[0135] In the above, the system includes a data classification module, an initialization module, a correlation analysis module, a personalized push module and a list update module. The data classification module is responsible for performing detailed classification processing based on the collected user behavior data. Through this process, the system can identify and extract multiple user preference tags, which reflect the user's interests and preferences in different fields. The function of the initialization module is to collect the browsing information of users under each user preference tag. Through in-depth analysis and processing of these browsing information, the system can effectively sort the user preference tags, thereby generating an initialization push list, which lays the foundation for subsequent personalized push. The correlation analysis module matches the recommended content based on the initialization push list, and performs detailed correlation analysis on each recommended content. Through this process The system can distinguish between related recommended content and non-related recommended content. The personalized push module uses related recommended content and non-related recommended content to reorder the initialized push list, thereby generating a more accurate personalized push list. The personalized push list can better meet the personalized needs of users and improve user satisfaction. Finally, the list update module is responsible for applying the personalized push list to the actual user push process. It will analyze the user's feedback on the personalized push content and optimize the personalized push list to generate an updated push list. The system will continue to use this updated push list for personalized push, forming a virtuous circle, continuously improving the quality of push content and user satisfaction, and bringing users a more accurate and considerate service experience.
[0136] See also Figure 3 , an electronic device, the electronic device comprising:
[0137] at least one processor;
[0138] and a memory communicatively coupled to the at least one processor;
[0139] The memory stores a computer program that can be executed by at least one processor, and the computer program is executed by at least one processor so that the at least one processor can implement the above-mentioned AIGC-driven automatic personalized content push method.
[0140] The processor of the electronic device can be of various types, such as a central processing unit (CPU), a graphics processing unit (GPU) or a dedicated AI processor. These processors can efficiently execute complex algorithms and data processing tasks to realize the functions of the personalized push system. The memory of the electronic device may include a random access memory (RAM), a read-only memory (ROM) and a non-volatile memory, such as a solid-state drive (SSD) or a hard disk drive (HDD). The computer program stored in the memory includes codes of a data classification module, an initialization module, a correlation analysis module, a personalized push module and a list update module. These codes can realize the entire personalized push process under the execution of the processor. In addition, the electronic device The device may also include input and output devices and an operator, such as a touch screen, keyboard, mouse and display, so that users can interact with the device. The operator is responsible for performing more complex mathematical operations and data processing tasks to ensure the efficient operation of the personalized push system. The device may also have a network connection function, such as a wireless local area network (WLAN) or a cellular network module to achieve data exchange with a server or other devices. In actual applications, the electronic device may be a smart phone, tablet computer, personal computer or server. The server can be deployed in the cloud to provide personalized push services to multiple users at the same time. Through distributed computing and big data processing technology, the server can process massive user data and achieve efficient and real-time personalized push.
[0141] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.
[0142] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.
Claims
1. An AIGC-driven automatic personalized content push method, characterized by: include: Based on the acquired user behavior data, classification processing is performed to obtain multiple user preference tags; Collecting browsing information of users under each of the user preference tags, and sorting the user preference tags according to the browsing information to obtain an initialization push list; Matching recommended content according to the initialization push list, and performing correlation analysis on each of the recommended content to obtain associated recommended content and non-associated recommended content; Reordering the initial push list according to the related recommended content and the non-related recommended content to obtain a personalized push list, wherein when pushing according to the personalized push list, it is possible to choose to skip the related recommended content related to the currently browsed content; Executing personalized push to the user based on the personalized push list, optimizing the personalized push list according to user feedback information under the personalized push, obtaining an updated push list, and continuing to execute personalized push according to the updated push list; The step of performing correlation analysis on each of the recommended contents to obtain related recommended contents and non-related recommended contents includes: Acquire all the recommended contents, and perform feature extraction to obtain key features of each of the recommended contents; Performing vector conversion on each of the key features to obtain a key feature vector of each of the recommended contents; Matching the recommended contents one by one, inputting the key feature vectors of the recommended contents after the pairwise matching into a preset association calculation function, and recording the output result of the association calculation function as an association score; Obtaining a correlation evaluation threshold, and comparing the correlation score with the correlation evaluation threshold; When the correlation score is greater than the correlation evaluation threshold, it indicates that there is a correlation between the recommended contents after pairwise matching, and they are simultaneously recorded as related recommended contents; When the correlation score is less than or equal to the correlation evaluation threshold, it indicates that there is no correlation between the recommended contents after pairwise matching, and they are simultaneously recorded as non-correlated recommended contents; The step of reordering the initialization push list according to the associated recommended content and the non-associated recommended content to obtain a personalized push list includes: Acquire the initialization push list, and record adjacent recommended contents in the initialization push list as pre-processed data and post-processed data respectively; Collecting the correlation score between the associated recommended content of the pre-processed data and the pre-processed data, converting the correlation score to the same dimension as the push score, and recording it as a condition parameter; Performing weighted sum processing on the condition parameter and the push score of the associated recommended content of the previous data to be processed to obtain the associated push score; Compare the associated push score with the push score of the subsequent data to be processed, and when the associated push score is greater than the push score of the subsequent data to be processed, adjust the associated recommended content corresponding to the associated push score to the front of the subsequent data to be processed, otherwise, continue to maintain the original order; Repeating the processing of all adjacent recommended contents in the initialization push list until the entire initialization push list is reordered and output as a personalized push list; When users view recommended content, they can have a certain degree of continuity. When the current browsing content pushed according to the personalized push list does not conform to or has subjective viewing singleness, the user can directly skip the recommended content associated with the current browsing content through the skip function.
2. The method for automatically personalizing content push driven by AIGC according to claim 1, characterized in that: The step of performing classification processing based on the acquired user behavior data to obtain multiple user preference tags includes: Establishing a historical monitoring period and collecting user behavior data within the historical monitoring period, wherein the user behavior data includes historical browsing content and historical search records; Preprocessing the historical browsing content and historical search records, wherein the preprocessing includes removing invalid data and abnormal values; Extracting feature information of the preprocessed historical browsing content and historical search records, and assigning user preference tags to the historical browsing content and historical search records under the same feature information; Among them, all the user preference tags are unique.
3. The method for automatically personalizing content push driven by AIGC according to claim 2, characterized in that: The step of collecting browsing information of users under each of the user preference tags, and sorting the user preference tags according to the browsing information to obtain an initial push list includes: Obtaining browsing information under the user preference tag during the historical monitoring period, wherein the browsing information includes the user's browsing completion degree and browsing times for historical browsing content; Performing a weighted score on each user preference tag according to the browsing completion degree and browsing times to obtain a push score for each user preference tag; The push scores are sorted in descending order, and the push order of each user preference tag is output one by one according to the sorting result of the push scores, and simultaneously summarized into an initialization push list.
4. The method for automatically personalizing content push driven by AIGC according to claim 3, characterized in that: The step of weighting and scoring each user preference tag according to the browsing completion degree and the browsing times to obtain a push score for each user preference tag includes: Setting a plurality of sample periods with equal time intervals within the historical monitoring period; Collecting the page views of all the historical browsing contents and the page view time of each of the historical browsing contents within each of the sample time periods; Determine the distribution weights of the browsing completion degree and the number of browsing times according to the browsing volume and browsing time of the historical browsing content; A weighted function is obtained, and the browsing completion degree and the number of browsing times, as well as the corresponding allocation weights are input into the weighted function for weighted calculation to obtain a push score for each of the user preference tags.
5. The method for automatically personalizing content push driven by AIGC according to claim 4, characterized in that: The step of determining the distribution weights of the browsing completion degree and the number of browsing times according to the browsing amount and browsing time of the historical browsing content includes: Calculate the unit page views in each sample period according to the page views and page views time of the historical page views in each sample period; Arrange the unit page views in each sample period according to the chronological order of occurrence; Obtaining a standard offset, and performing an offset process on each of the unit views according to the standard offset to obtain a plurality of clustering intervals; Counting the number of unit views in each clustering interval and recording it as a screening condition parameter, and marking the clustering interval under the screening condition parameter with the largest value as a standard interval; The unit page views in the standard interval are summed and averaged to obtain the standard unit page views in the historical monitoring period; A weight allocation function is obtained, and the standard unit browsing volume is input into the weight allocation function, and the output result of the weight allocation function is recorded as the allocation weight of the browsing completion degree and the browsing number.
6. The method for automatically personalizing content push driven by AIGC according to claim 1, characterized in that: The step of optimizing the personalized push list according to user feedback information under personalized push to obtain an updated push list includes: Obtain user feedback and satisfaction with each recommended content under personalized push; Convert the feedback satisfaction into an optimized weight of a push score corresponding to each recommended content; Obtain an optimization function, and input the push score and the corresponding optimization weight corresponding to each recommended content into the optimization function for calculation to obtain an updated push score; The personalized push list is reordered according to the update push score to obtain an update push list, and the update push list is applied to execute push processing of each recommended content.
7. An AIGC-driven automatic personalized content push system, characterized by: The method for automatically personalizing content push driven by AIGC according to any one of claims 1 to 6 comprises: A data classification module, wherein the data classification module is used to perform classification processing based on the acquired user behavior data to obtain multiple user preference tags; An initialization module, the initialization module is used to collect browsing information of users under each of the user preference tags, and sort the user preference tags according to the browsing information to obtain an initialization push list; A correlation analysis module, the correlation analysis module is used to match the recommended content according to the initialization push list, and perform correlation analysis on each of the recommended content to obtain related recommended content and non-related recommended content; A personalized push module, the personalized push module is used to reorder the initialized push list according to the associated recommended content and the non-associated recommended content to obtain a personalized push list; A list update module, wherein the list update module is used to perform personalized push to users based on the personalized push list, and optimize the personalized push list according to user feedback information under the personalized push to obtain an updated push list, and continue to perform personalized push according to the updated push list.
8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the AIGC-driven automatic personalized content push method described in any one of claims 1 to 6.
Citation Information
Patent Citations
Recommendation list determination method and server
CN115455280A
Visual website construction promotion method and promotion system
CN117972257A
