Virtual creation content personalized pushing method and system based on user portrait

Abstract

The application relates to the technical field of content pushing, in particular to a virtual creation content personalized pushing method and system based on a user portrait, which comprises the following steps: acquiring each virtual creation content in user historical browsing information; determining a candidate term set of each creation content; screening a keyword set of a user portrait; acquiring a first characteristic value, a second characteristic value and a third characteristic value of each keyword, obtaining an interest heat weight of each keyword; performing upper extension on the keyword, performing hierarchical text classification on all the extended keywords; determining the interest heat weight of each extended keyword according to the hierarchical relationship between the classified keywords and the distribution of the known interest heat weight; and constructing a user portrait based on the interest heat weight of all the keywords, so as to push virtual creation content to the user. The pushing accuracy of the virtual creation content is improved.

Description

Technical Field

[0001] This application relates to the field of content delivery technology, specifically to a method and system for personalized delivery of virtual creative content based on user profiles. Background Technology

[0002] With the rapid development of internet technology, the way we obtain information has shifted from traditional media to digital online information platforms. This has led to the emergence of the challenge of filtering valuable virtual content from massive amounts of data and personalizing it for specific user groups. By utilizing user browsing history, preferences, and other data to build user profiles, and then accurately pushing virtual content that users are interested in, the traffic needs of information platforms and creators are matched with the personalized needs of users.

[0003] User profiles are the foundation for personalized content delivery and are crucial for delivering information that users are interested in. Existing methods, when constructing user profiles, determine user interests and preferences based on historical browsing data and simple behavioral statistics over a long period, without considering the timeliness and validity of user preference data, making it difficult to capture dynamic changes in user interests. Furthermore, relying solely on keywords from user browsing history to construct user profiles fails to deeply mine user preferences, resulting in low accuracy in user profiling and impacting the precision and targeting of virtual content delivery. Summary of the Invention

[0004] To address the aforementioned technical issues, the purpose of this application is to provide a method and system for personalized push notifications of virtual creative content based on user profiles. The specific technical solution adopted is as follows: In a first aspect, embodiments of this application provide a method for personalized push of virtual creation content based on user profiles, the method comprising the following steps: Obtain each piece of virtual creation content from the user's browsing history; Based on the visual, textual, and keyword information of each piece of content, a set of candidate keywords for each piece of content is determined; based on the browsing time percentage of each piece of content, the browsing effectiveness of each piece of content is determined, which is used to filter and obtain a set of keyword keywords for user profile based on the set of candidate keywords. The first feature value of each keyword is obtained by analyzing the browsing effectiveness of the content corresponding to each keyword in the keyword set; the second feature value of each keyword is determined by analyzing the interaction between the user and the content corresponding to each keyword; and the third feature value of each keyword is determined based on the changes of each keyword over time in the user's browsing history. By fusing the first feature value, the second feature value, and the third feature value, the interest popularity weight of each keyword entry is obtained; the keyword entries are expanded at a higher level, and hierarchical text classification is performed on all expanded keyword entries; based on the hierarchical relationship between the classified entries and the known distribution of interest popularity weights, the interest popularity weight of each expanded keyword entry is determined. User profiles are built based on the interest and popularity weights of all keywords, which are then used to push virtual creation content to users.

[0005] In one embodiment, determining the set of candidate terms for each piece of created content includes: Visual tag word set is extracted for each piece of content using visual recognition technology, topic word set is extracted for the text information of each piece of content using LDA topic model, and keyword set is obtained when each piece of content is published; Each term in the visual tag term set, theme term set, and keyword set is assigned a weight. The weight of any term in the visual tag term set, theme term set, and keyword set is the sum of the weights of its respective set. All terms are sorted in descending order of weight, and the terms corresponding to the first preset number of weights are used to form a candidate term set for each piece of content.

[0006] In one embodiment, the browsing validity is the proportion of browsing time for each piece of content to the total duration of the content.

[0007] In one embodiment, the filtering of the keyword set for obtaining the user profile includes: Content with browsing validity greater than a preset threshold is considered valid browsing content. After deduplicating the candidate terms from all valid browsing content, a set of keyword terms for the user profile is formed.

[0008] In one embodiment, the first feature value is the average browsing effectiveness of all created content corresponding to each keyword.

[0009] In one embodiment, determining the second feature value for each keyword entry includes: User interactions with the content created for each keyword include clicking, liking, commenting, and sharing, with interaction levels increasing sequentially. Each interaction is assigned a value in ascending order of interaction level as its importance, and the importance of the interaction is positively correlated with the interaction level. The maximum interaction importance of a user for a single piece of content is taken as the interaction importance of the corresponding content. The second feature value is the average interaction importance of all content corresponding to each keyword.

[0010] In one embodiment, determining the third feature value for each keyword entry includes: Suppose that the candidate term set of the content viewed d days ago contains the keyword entries. The number of creative contents is ,in, D represents the preset number of days prior to the current date, and the keyword entry. The third eigenvalue The expression is: In the formula, exp() is an exponential function with the natural constant as the base.

[0011] In one embodiment, determining the interest popularity weight of each expanded keyword entry includes: For any extended keyword, count the number of interval levels between the keyword and any of its subordinate keyword, obtain the calculation result of an exponential function with a preset distance decay factor as the base and the number of interval levels as the exponent, calculate the product of the calculation result and the interest popularity weight of the subordinate keyword, and calculate the sum of the products of the keyword and all its subordinate keyword. The interest popularity weight of any keyword is obtained by combining the dispersion of the interest popularity weights of all subordinate keyword entries of any keyword entry with the normalized value of the sum.

[0012] In one embodiment, constructing a user profile based on the interest popularity weights of all keyword entries includes: The user's keywords are used as input to a word cloud function, and the word frequency statistics in the word cloud function are replaced with the interest popularity weight of each keyword to output a user profile.

[0013] Secondly, embodiments of this application also provide a personalized push system for virtual creative content based on user profiles, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of any of the methods described above.

[0014] This application has at least the following beneficial effects: This application improves the accuracy and comprehensiveness of extracted keywords by analyzing users' historical browsing information and combining the visual, textual, and keyword features of each piece of content, enabling a more accurate capture of user interests and preferences. By analyzing the browsing time of each piece of content, the effectiveness of browsing is determined, and a set of keyword entries for user profiles is obtained, improving the accuracy and refinement of user profiling and helping to make the pushed content more aligned with user interests. By analyzing user interactions with content, the application can more accurately assess which content has a real impact on users, enhancing the utilization value of interaction data and preventing overly simplistic click or browsing data from being misinterpreted as the sole standard of user interest. Based on the changes in keyword entries over time in users' historical browsing information, the application can dynamically update user... By analyzing changes in user interests, the push algorithm can adapt to these changes, solving the problem that traditional push methods cannot track changes in user interests. This ensures that recommended content always reflects the latest user preferences, improving the continuity and effectiveness of the user experience. By integrating the first, second, and third feature values ​​of keyword entries, the algorithm comprehensively evaluates the multi-dimensional characteristics of user interests, avoiding the limitations of traditional user profiling that relies too heavily on historical behavior. Through continuous learning and adjustment, it can more accurately identify and adapt to changes in user interests, solving the problem that traditional user profiling cannot adapt to rapidly changing interests. By expanding keyword entries and performing hierarchical text classification, the push system can handle and explore a wider range of interest dimensions, avoiding the problem of being limited to a narrow set of keywords. This builds a more robust user profile and improves the accuracy and diversity of subsequent personalized pushes. The introduction of hierarchical relationships allows for a better grasp of diverse user needs, preventing the over-concentration of recommended content on a particular field or keywords from overlooking other potential interests, thus ensuring the diversity and comprehensiveness of recommended content. The interest popularity weight of keyword entries not only reflects the recommendation priority of mainstream popular content, but also effectively mines users' long-tail interests through hierarchical expansion, representing the user's true interest in keyword entries, avoiding the invalidation of keyword entries, improving the authenticity and effectiveness of user profiles while maintaining timeliness. By expanding keywords related to user interests and refining their classification, potentially attractive content can be pushed, broadening the content coverage of recommended information, increasing the diversity of creative content users encounter, improving content exposure and user stickiness. Through accurate interest matching and continuously updated user profiles, users can receive virtual creative content that better matches their needs and interests, improving the accuracy of content push. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating the steps of a method for personalized push notification of virtual creative content based on user profiles, provided in one embodiment of this application. Figure 2 Build a flowchart for user profiles. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by this application to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the personalized push method and system for virtual creation content based on user profiles proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, 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 application pertains.

[0019] The following, in conjunction with the accompanying drawings, details the specific scheme of the personalized push method and system for virtual creative content based on user profiles provided in this application.

[0020] Please see Figure 1 The diagram illustrates a flowchart of a method for personalized push notifications of virtual creative content based on user profiles, according to an embodiment of this application. The method includes the following steps: S1 retrieves each piece of virtual creation content from the user's browsing history.

[0021] This embodiment acquires users' historical browsing information and behavioral information data on a virtual content creation and publishing platform. The historical browsing information data includes browsing history, search history, and browsing time for each piece of virtual content; the behavioral information data includes likes, comments, shares, and favorites of the content. The time span of the acquired historical browsing information and behavioral information data is from the present to the past. Data within a day, time span of days The implementer can set the parameters according to the implementation scenario without any special restrictions. In this embodiment, the parameters are as follows: The size is set to 10.

[0022] The acquired user browsing history and behavior data are cleaned according to the timestamp attribute to avoid data duplication and affect subsequent big data statistical analysis.

[0023] S2, based on the visual information, text information, and keyword information of each piece of created content, determine the candidate term set for each piece of created content; determine the browsing validity of each piece of created content based on the browsing time percentage of each piece of created content, which is used to filter and obtain the keyword set for user profile based on the candidate term set.

[0024] Existing methods for building user profiles by collecting user browsing and preference data typically rely on simple statistical information such as clicks and views. They add keywords related to the content being created to the user profile and then push content directly or indirectly similar to that content within a short period. This approach lacks consideration of whether clicks were accidental. Furthermore, relying on historical browsing records can lead to the continued pushing of repeatedly viewed, outdated content, resulting in delayed content delivery and a lack of authenticity and timeliness in the user profile, leading to erroneous content delivery. Moreover, building user profiles solely based on user browsing history and keyword entries is insufficient for in-depth analysis of user preferences. To achieve higher-quality personalized push notifications, more robust user profiles are needed.

[0025] The keywords in the user profile are derived from the user's basic information data and the keywords associated with the content they browse. The basic information data includes keywords such as the user's age group, gender, and region of origin. These basic keywords are used to push content that matches the user's basic information during the cold start phase.

[0026] Furthermore, for each piece of virtual content created by a user obtained in this embodiment, a set of keywords for the user profile is obtained based on the user's preferences for virtual content. Specifically: Before each virtual creation can be publicly released on the platform, the creator needs to assign keywords, and the content must undergo multi-faceted review by the platform. For video content, visual and audio information is extracted from the video media file during the review stage. Then, AI visual recognition technology is used to identify the types of visual scenes and objects in the video, and outputs visual tag terms contained in the video, such as animal or human information. All visual tag terms for each virtual creation are combined into a visual tag term set. For the audio information of the video, speech recognition technology is used to convert the audio information into text information. Then, LDA topic model is used to extract the text topic terms of the video content from the text data, forming a topic term set for each virtual creation. LDA topic model is a well-known existing technology, and its specific process will not be elaborated here.

[0027] For the obtained user browsing history data Virtual creation content ,in, Indicates the first Virtual creation content, virtual creation content The set of keywords is denoted as It includes multiple virtual creations. Keywords provided during publication; virtual creation content The set of visual tag entries is denoted as It includes several visual tag entries; virtual creation content The set of keywords is denoted as It contains several textual keywords.

[0028] So, virtual creation content Historical browsing term collection Represented as: in, , , To separately , , Each of the three term sets is assigned a score weight to a new set. , , They are respectively , , The score weight of each term is used to represent the contribution of the three factors to the set of historically viewed terms, satisfying the following conditions: Its size can be set by the implementer according to the implementation scenario, without special restrictions. In this embodiment, , , The values ​​were set to 0.4, 0.3, and 0.3 respectively. This is the union operator.

[0029] The final collection of historical browsing entries for , , The union of terms. Each term in the set has a score weight. Assume the same term appears in all three sets of terms. So, the entry The greater the weight of the final score, the more numerically it is expressed as Explanation of the term It perfectly matches the creative content. Topics, entries As creative content The probability of a keyword entry is higher; if the entry If a term appears only in one set of terms, its final score weight may be relatively small, as it is considered creative content. The lower the probability of a keyword appearing in a search query.

[0030] Collection of browsing history entries The final scores of all entries are sorted in descending order, and the entries corresponding to the top K final scores are used as the content to be created. The candidate terms are used to form a candidate term set. ,in, This represents the j-th candidate term. The size can be set by the implementer according to the implementation scenario, without special restrictions. In this embodiment... .

[0031] To prevent accidental clicks from corrupting user profiles through browsing history, the first step is to define the validity of browsing for each piece of created content. For the first item in the browsing history... Each creative content The total duration is defined as In this embodiment, the total duration of video media content is the total video duration; the total duration of text content is the recommended optimal reading duration. In this embodiment, the recommended optimal reading duration for text is defined according to a reading speed of 400 words per minute. The reading speed standard can be set by the implementer, and this embodiment does not impose any special restrictions. The created content... The browsing time is recorded as Then the content of creation The calculation method for browsing validity is as follows: In the formula, For the first Each creative content The validity of the browsing.

[0032] For all content created in a user's browsing history, content with browsing validity exceeding a preset threshold is considered valid browsing content. The candidate keywords from all valid browsing content are deduplicated to form a keyword set for the user profile, used to construct the user profile. In this embodiment, the preset threshold is set to 0.2; implementers can set it according to actual circumstances, and this embodiment does not impose any restrictions on it.

[0033] S3. Obtain the first feature value of each keyword by analyzing the browsing effectiveness of the content corresponding to each keyword in the keyword set; determine the second feature value of each keyword by analyzing the interaction between the user and the content corresponding to each keyword; and determine the third feature value of each keyword based on the changes of each keyword in the user's browsing history over time.

[0034] For the keyword set of user profiles, the first Keyword When creating content The candidate term set contains keyword terms. At that time, calculate the content created. The effectiveness of browsing is determined by the number of keywords appearing in the candidate term set. All created content was analyzed, and the average browsing effectiveness of this created content was calculated, denoted as the _th_. Keyword First eigenvalue First eigenvalue The closer a value is to 1, the more positive the user's reaction to the keyword. The higher the interest, the more complete the browsing experience can be of related creative content each time.

[0035] Furthermore, when creating content The candidate term set contains keyword terms. At the same time, analyze users' content creation. The interactive behaviors in this embodiment include clicking, liking, commenting, and sharing, with the interaction level increasing sequentially. Each interactive behavior is assigned a value in ascending order of interaction level as its importance, where the importance is positively correlated with the interaction level. In this embodiment, the interaction importance of clicking, liking, commenting, and sharing is assigned values ​​of 1, 2, 3, and 4 respectively. Implementers can set the values ​​of interactive behaviors according to actual circumstances; this embodiment does not impose any restrictions on this.

[0036] When users interact with content creators When an interactive action occurs, the interaction with the highest importance level is selected as the content creation factor. The importance of interaction, for example, when users interact with content creators. If interactive behaviors include liking and sharing, then the importance of the interaction corresponding to the sharing will be considered as the content creation value. The importance of interaction. Statistically, the number of keywords appearing in the candidate keyword set. All created content, and calculate the mean of the interaction importance of this created content, as the first... Keyword The second eigenvalue The larger the second feature value, the more interested the user is in the keyword.

[0037] Furthermore, suppose that the candidate term set of the content created by the user d days ago contains keyword terms. The number of creative contents is ,in, D represents the preset number of days prior to the current date. Keyword The third eigenvalue The expression is: In the formula, exp() is an exponential function with the natural constant as the base. Wherein, This represents the freshness weight, which gradually decreases as the interval between browsing records increases. It is used to characterize the gradual cooling of interest; the longer the time interval between browsing and the statistical day, the less clearly the user's level of interest is reflected. This embodiment constructs the third feature value. This can reflect users' preferences for timely topics of interest. For a trending topic, users may browse it frequently on a particular day, but as time goes on, they may lose interest in the topic, resulting in more keywords from several days ago being viewed more frequently. The more frequently a user browses a content, the lower its contribution to the third characteristic value; conversely, the more recently browsed the keywords in the content, the higher their freshness, and the better they reflect the user's level of interest.

[0038] S4, integrate the first feature value, the second feature value, and the third feature value to obtain the interest popularity weight of each keyword; perform hierarchical expansion on the keyword, and perform hierarchical text classification on all expanded keyword entries; determine the interest popularity weight of each expanded keyword based on the hierarchical relationship between the classified entries and the known distribution of interest popularity weights.

[0039] By integrating the first, second, and third feature values ​​of the x-th keyword, a new keyword in the keyword set is constructed. Keyword Interest popularity weight It should be noted that fusion refers to combining multiple variables, which can be done through addition, multiplication, or a combination of addition and multiplication. In this embodiment, the first... Keyword Interest popularity weight The specific expression is: In the formula, For the first keyword in the keyword collection Keyword Interest popularity weighting For the first Keyword The first eigenvalue, For the first Keyword The second eigenvalue, For the first Keyword The third eigenvalue, For normalization, this embodiment uses the Sigmoid function.

[0040] The calculation of interest popularity weight combines users' browsing and interaction behavior with certain creative content over a period of time. This can avoid invalid clicks from polluting user profiles, ensure the authenticity and accuracy of keyword entries, and evaluate the timeliness of keyword entries based on changes in users' browsing frequency over a period of time, ensuring the timeliness of keyword entries in user profiles.

[0041] Furthermore, to avoid the user profile being too niche and isolated, resulting in a sparse user profile in subsequent personalized recommendations, this embodiment considers expanding the user profile based on the keywords in the user keyword set to a higher domain. This ensures that the constructed user profile is more robust and can be used to discover content that users are interested in, thereby achieving better personalized recommendations.

[0042] Therefore, this embodiment utilizes Wikipedia, an open-source knowledge base in the field of natural language processing, to expand the keywords in the keyword set. For example, if the keyword in the keyword set is the name of a basketball star, it can be expanded to include: the name of the basketball star's team / club → basketball → sports, etc. Let the keyword set contain X keywords, and the expanded set contain Y keywords, then Y ≥ X. Implementers can use other existing open-source knowledge bases as needed; this embodiment does not impose any restrictions on this.

[0043] Furthermore, for the expanded set of keywords, this embodiment performs hierarchical text classification on all keywords to obtain various tree structures. For example, if "news" is the parent category, and the keyword set includes not only "news" but also "international news," "economic news," and "global stock market," then the constructed tree structure is "news → international news → economic news → global stock market."

[0044] Based on the interest popularity weights of the original X keyword entries in the keyword entry set, the interest popularity weights of each expanded keyword entry are calculated. It should be understood that in each tree structure, at least the last child node has an interest popularity weight. This embodiment uses a bottom-up traversal strategy to traverse the entire tree structure. The node corresponding to the first expanded keyword entry without an interest popularity weight is taken as the parent node. All its child nodes are searched, and the interest popularity weights of its child nodes are used to perform distance decay to obtain the parent node's interest popularity weight, until all nodes in the entire tree structure are assigned interest popularity weights.

[0045] Specifically, for any parent node of a tree structure Its interest popularity weight The expression is: In the formula, The suppression coefficient, parent node The set of child node keywords, Indicates the parent node The set of child node keywords Keyword entries for each sub-node Distance attenuation coefficient, Indicates the parent node The set of child node keywords Keyword entries for each sub-node Interest popularity weighting , representing the distance attenuation factor, is used to control the attenuation intensity of the distance attenuation coefficient. Its magnitude can be set by the implementer according to the implementation scenario. In this embodiment, The size is 0.6. Indicates the parent node To child nodes The number of hierarchical levels between them, i.e., the parent node With child nodes The greater the hierarchical difference between them, the more layers they are separated by, and the more child nodes they have. Interest popularity weight for parent nodes The smaller the impact.

[0046] The interest popularity weight of higher-level extended keywords decreases as their distance from child nodes increases, and is normalized according to the interest popularity weight of child nodes. This prevents the interest popularity weight of higher-level extended keywords from exceeding that of their child nodes, thus avoiding excessive expansion of higher-level extended keywords and negatively impacting the keywords that users are truly interested in based on their profiles. Furthermore, an inhibition coefficient is used... Judging the balance of interests under the parent term based on the distribution of interest popularity weights in the child nodes, and then performing secondary adjustment, can more robustly represent user preferences. The expression for the inhibition coefficient is: In the formula, Indicates the parent node The information entropy of the interest popularity weights of all child nodes in the set of child node keywords reflects the degree of dispersion of the interest popularity weight distribution. This represents its theoretical maximum entropy. Parent node The number of child nodes in the set of child keyword entries, where log is the logarithm function to the base 2. When the parent node When a parent node contains many child node keywords and the distribution of interest popularity weights among these child node keywords varies significantly, it indicates that users are less interested in the parent node. Interest arises from a preference for a particular term, at which point the inhibition coefficient... Smaller, suppressing the weight of interest popularity The expansion of the parent node; conversely, it indicates the user's perception of the parent node. I am quite interested in all the entries below, and I also want to expand on the related terms. Content that users are interested in is more likely to be included, thus avoiding excessive suppression of interest-based popularity weighting. The expansion.

[0047] S5 constructs user profiles based on the interest and popularity weights of all keyword entries, which are then used to push virtual creation content to users.

[0048] Repeat the above steps to obtain the interest popularity weights of all higher-level expanded keyword entries. Use all of the user's keyword entries as input to the WordCloud word cloud function, and replace the word frequency statistics in the WordCloud word cloud function with the interest popularity weight of each keyword entry for keyword importance assessment. A higher interest popularity weight indicates higher keyword importance. Output the user profile. The calculation of the WordCloud word cloud function is a well-known technology, and the specific process will not be elaborated here. The user profile construction flowchart is as follows: Figure 2 As shown.

[0049] The user profile constructed in this embodiment involves a wide range of keywords, which facilitates the search for similar content for subsequent personalized push notifications. This avoids excessive sparsity in the co-occurrence matrix of users and creative content, and the interest popularity weight of the extended keywords will not exceed that of the original keywords, thus avoiding excessive mining of creative content from other related domains for push notifications.

[0050] The user profile constructed using this embodiment is updated with a set update cycle, which is set to D in this embodiment, the same as the number of days of the acquired historical data. The user profile is updated every update cycle.

[0051] Furthermore, information about the content created on the publishing platform is obtained, and its content tags are acquired based on this information. Target users matching these content tags are then identified from the user profile database. Content matching these tags is then recommended to the user, achieving personalized content delivery. Pushing content based on user profiles is a well-known existing technology, and this embodiment will not elaborate on it in detail.

[0052] Based on the same inventive concept as the above methods, this application also provides a personalized push system for virtual creative content based on user profiles, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above methods for personalized push of virtual creative content based on user profiles.

[0053] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0054] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0055] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A method for personalized push of virtual creation content based on user profiles, characterized in that, The method includes the following steps: Obtain each piece of virtual creation content from the user's browsing history; Based on the visual, textual, and keyword information of each piece of content, a set of candidate keywords for each piece of content is determined. The browsing validity of each piece of content is determined based on its browsing time percentage, and is used to filter and obtain a set of keyword entries for the user profile based on the candidate keyword set. The browsing validity is the proportion of each piece of content's browsing time to the total browsing time of the content. Content with browsing validity greater than a preset threshold is considered valid browsing content. The keyword entry set for the user profile is then formed by deduplicating the keywords from the candidate keyword set of all valid browsing content. The first feature value of each keyword is obtained by analyzing the browsing effectiveness of the corresponding content in the keyword set; the second feature value of each keyword is determined by analyzing the user's interaction with the corresponding content; and the third feature value of each keyword is determined based on the changes of each keyword over time in the user's browsing history, including: assuming that the candidate keyword set of the content viewed d days ago contains the keyword. The number of creative contents is ,in, D represents the preset number of days prior to the current date, and the keyword entry. The third eigenvalue The expression is: In the formula, exp() is an exponential function with the natural constant as the base; By fusing the first feature value, the second feature value, and the third feature value, the interest popularity weight of each keyword is obtained; the keyword is expanded to a higher level, and hierarchical text classification is performed on all expanded keyword entries; based on the hierarchical relationship between the classified entries and the known distribution of interest popularity weights, the interest popularity weight of each expanded keyword is determined, including: for any expanded keyword, counting the number of interval levels between the keyword and any of its subordinate keyword entries, obtaining the calculation result of an exponential function with a preset distance decay factor as the base and the number of interval levels as the exponent, calculating the product of the calculation result and the interest popularity weight of the subordinate keyword, and calculating the sum of the products of the keyword and all its subordinate keyword entries; combining the dispersion of the interest popularity weights of all subordinate keyword entries of the keyword and the normalized value of the sum, the interest popularity weight of the keyword is obtained; User profiles are built based on the interest and popularity weights of all keywords, which are then used to push virtual creation content to users.

2. The personalized push method for virtual creative content based on user profiles as described in claim 1, characterized in that, The process of determining the candidate term set for each piece of creative content includes: Visual tag word set is extracted for each piece of content using visual recognition technology, topic word set is extracted for the text information of each piece of content using LDA topic model, and keyword set is obtained when each piece of content is published; Each term in the visual tag term set, theme term set, and keyword set is assigned a weight. The weight of any term in the visual tag term set, theme term set, and keyword set is the sum of the weights of its respective set. All terms are sorted in descending order of weight, and the terms corresponding to the first preset number of weights are used to form a candidate term set for each piece of content.

3. The personalized push method for virtual creative content based on user profiles as described in claim 1, characterized in that, The first feature value is the average browsing effectiveness of all created content corresponding to each keyword.

4. The method for personalized push of virtual creation content based on user profiles as described in claim 1, characterized in that, Determining the second feature value for each keyword entry includes: User interactions with the content created for each keyword include clicking, liking, commenting, and sharing, with interaction levels increasing sequentially. Each interaction is assigned a value in ascending order of interaction level as its importance, and the importance of the interaction is positively correlated with the interaction level. The maximum interaction importance of a user for a single piece of content is taken as the interaction importance of the corresponding content. The second feature value is the average interaction importance of all content corresponding to each keyword.

5. The personalized push method for virtual creation content based on user profiles as described in claim 1, characterized in that, The process of constructing user profiles based on the interest popularity weights of all keyword entries includes: The user's keywords are used as input to a word cloud function, and the word frequency statistics in the word cloud function are replaced with the interest popularity weight of each keyword to output a user profile.

6. A personalized virtual content push system based on user profiles, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-5.

Images