A content recommendation method and system based on big data and artificial intelligence

Abstract

The application discloses a content recommendation method and system based on big data and artificial intelligence, comprising: firstly capturing the behavior of the target user in the preset period, and using the pre-trained user behavior recognition model to analyze and identify the behavior. The model is trained by a large amount of user historical behavior data, so it can accurately identify the deep intention of user behavior. According to the identification result, the system will build a precise user portrait for the user, and then provide personalized content recommendation for the user based on the portrait. Such design combines the advantages of big data analysis and artificial intelligence, can accurately capture user demand in real time, and provide more accurate content recommendation service for the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a content recommendation method and system based on big data and artificial intelligence. BACKGROUND

[0002] With the rapid development of Internet technology, users face a huge amount of information content every day. How to filter out the content of interest from these information for users has become the key to the development of technology. The traditional recommendation method is often based on simple classification or historical browsing records. This method is inefficient when facing a large amount of data, and it is difficult to accurately capture the personalized needs of users. SUMMARY

[0003] The purpose of the present application is to provide a content recommendation method and system based on big data and artificial intelligence.

[0004] In a first aspect, the embodiments of the present application provide a content recommendation method based on big data and artificial intelligence, comprising:

[0005] Obtaining at least one current user behavior of a target user in a preset monitoring period;

[0006] Inputting the at least one current user behavior into a pre-trained user behavior recognition model to obtain a user behavior recognition result corresponding to the at least one current user behavior, wherein the user behavior recognition model is trained based on a user historical behavior set as a training set;

[0007] According to the user behavior recognition result, constructing a target user portrait of the target user;

[0008] Based on the target user portrait, performing content recommendation for the target user.

[0009] In a second aspect, the embodiments of the present application provide a server system comprising a server, wherein the server is configured to execute the method of the first aspect.

[0010] Compared with the prior art, the present application provides the beneficial effects including: by using the content recommendation method and system based on big data and artificial intelligence disclosed in the present application, the behaviors of the target user in the preset period are captured, and the pre-trained user behavior recognition model is used to analyze and identify these behaviors. The model is trained by a large amount of user historical behavior data, so it can accurately identify the deep intention of user behavior. According to the recognition result, the system will construct a precise user portrait for the user, and then provide personalized content recommendation for the user based on the portrait. This design combines the advantages of big data analysis and artificial intelligence, can accurately capture the user's needs in real time, and provide more accurate content recommendation service for the user. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as limiting the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A flowchart illustrating the steps of a content recommendation method based on big data and artificial intelligence provided in an embodiment of the present invention;

[0013] Figure 2 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0015] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0016] In order to solve the technical problems mentioned in the background art Figure 1 This is a flowchart illustrating the content recommendation method based on big data and artificial intelligence provided in this embodiment of the disclosure. The following is a detailed description of this content recommendation method based on big data and artificial intelligence.

[0017] Step S201: Obtain at least one current user behavior of the target user within a preset monitoring period;

[0018] Step S202: Input the at least one current user behavior into a pre-trained user behavior recognition model to obtain the user behavior recognition result corresponding to the at least one current user behavior, wherein the user behavior recognition model is trained based on a set of user historical behaviors as the training set.

[0019] Step S203: Based on the user behavior recognition results, construct a target user profile for the target user;

[0020] Step S204: Based on the target user profile, recommend content to the target user.

[0021] In this embodiment of the invention, for example, the server records and analyzes user activities on the platform. For instance, user A watched multiple science fiction movies on a video streaming platform over the past week (a preset monitoring period) and gave several of them high ratings. Simultaneously, the server also monitored user A searching for and browsing news and forum discussions related to science fiction movies. The server inputs this behavioral data into a pre-trained user behavior recognition model. This model has been trained using a large amount of historical user behavior data and is capable of recognizing user preferences. After analysis, the model concludes that user A has a strong interest in science fiction movies, especially in themes of future technology imagination and space exploration. Based on the model's analysis results, the server constructs a user profile for user A: a male aged 30-40, a science fiction fan, enjoys exploring the unknown, is interested in new technologies and ideas, has a certain level of spending power, and is willing to pay for high-quality science fiction content. Based on this user profile, the server begins recommending content to user A. First, the server pushes a trailer for a newly released science fiction movie, which tells a story of future space exploration, perfectly matching user A's interests. Simultaneously, the system recommended science fiction books, games, and related online discussion groups to meet User A's need for in-depth exploration of the science fiction genre. Throughout the process, the server continuously monitored User A's feedback, such as whether they clicked on recommended content, watched a full movie, or read a recommended book, in order to further optimize the recommendation algorithm and improve the accuracy of the user profile.

[0022] In this embodiment of the invention, the user behavior recognition model is obtained in the following way.

[0023] Multiple historical behaviors to be classified are obtained from the user's historical behavior set, and multiple undetermined behavior categories are determined;

[0024] For each of the plurality of unclassified historical behaviors, determine a plurality of initial candidate behavior categories that are successfully associated with the target unclassified historical behavior from the plurality of undetermined behavior categories;

[0025] When the plurality of initial candidate behavior categories include a comparison behavior category, the target historical behavior to be classified is configured with behavior categories based on the plurality of initial candidate behavior categories to obtain the assigned type behavior of the target historical behavior to be classified; the comparison behavior category is obtained based on the assigned type behavior in the set of assigned type behaviors.

[0026] The set of assigned type behaviors is optimized based on the one-to-one correspondence between the multiple unclassified historical behaviors, and the unclassified historical behaviors in the set of user historical behaviors are configured with behavior categories until the behavior type classification process is completed for all unclassified historical behaviors in the set of user historical behaviors.

[0027] The user's historical behavior set, after completing the behavior type classification process, is used as the training set to train the initial user behavior recognition model, thus obtaining the trained user behavior recognition model.

[0028] In this embodiment of the invention, for example, the server first collects a large amount of historical behavior data from users (including user A), including viewing records, search records, purchase records, etc. Then, the server defines some preliminary behavior categories, such as "watching science fiction movies," "searching science fiction topics," and "purchasing books." For each historical behavior of user A, the server analyzes which preliminary behavior category it best matches. For example, for the record of user A watching a science fiction movie, the server will categorize it into the preliminary candidate behavior category of "watching science fiction movies." In some cases, a historical behavior can be associated with multiple preliminary candidate behavior categories simultaneously. For example, user A may be watching a science fiction movie while also searching for books related to that movie. In this case, the server will categorize this behavior into both "watching science fiction movies" and "searching science fiction topics" and configure the behavior categories accordingly. As more and more historical behaviors are classified and configured, the server continuously optimizes and adjusts the behavior categories. For example, the server may find a lot of overlap between the two categories of "watching science fiction movies" and "searching science fiction topics," and decide to merge these two categories into "consuming science fiction content." This process continues until all historical behaviors are accurately categorized and configured. Finally, the server uses this categorized and configured historical behavior data as a training set to train the initial user behavior recognition model. Through continuous iteration and optimization, the model gradually learns how to accurately identify and understand various user behaviors, thereby providing users with more precise content recommendation services. After these steps, the server obtains an efficient and accurate user behavior recognition model, providing strong support for subsequent content recommendation.

[0029] To more clearly describe the solutions provided in the embodiments of this application, a more detailed explanation is provided below.

[0030] In this embodiment of the invention, for example, a user history behavior set is a database or dataset containing all user behaviors within a certain time period. These behaviors may include the user's browsing history, search history, purchase history, click history, etc. This behavioral data is collected and stored for subsequent data analysis and model training. For example, on an e-commerce platform, a user history behavior set may include product pages viewed by the user, items added to the shopping cart, orders placed but not paid for, paid orders, product reviews, searched keywords, etc. All of this behavioral data is recorded in this set. Obtaining multiple unclassified historical behaviors involves extracting behavioral data from the user history behavior set that has not yet been classified. These unclassified historical behaviors are the basis for model training or data analysis and need to be classified using certain algorithms or rules. Suppose the server extracts user A's browsing history over the past week from the user history behavior set. These records include browsing different types of product pages, such as home appliances, clothing, food, etc. These browsing records are unclassified historical behaviors because the server needs to further analyze which type or category these behaviors belong to. Undetermined behavior categories refer to some possible behavior types or tags that are pre-set or automatically generated based on data characteristics. These categories are used to categorize or label users' historical behavior, helping to understand user behavior patterns and preferences. In content recommendation scenarios, pending behavior categories can include "watching science fiction movies," "watching comedy movies," "reading news," and "reading technology articles," among others. These categories are set based on the characteristics of the platform's content and users' potential interests. When a user's historical behavior matches these categories, they can be categorized into the corresponding category.

[0031] Furthermore, multiple unclassified historical behaviors refer to several user behaviors extracted from a user's historical behavior set that have not yet been categorized. These behaviors involve various operations and activities of the user at different times and in different contexts, such as browsing, searching, and purchasing. Suppose a video streaming platform records all of user A's viewing history within a month, including watching different types of movies, TV series, and short videos. These viewing records constitute multiple unclassified historical behaviors for user A. A target unclassified historical behavior refers to a specific behavior currently being processed and selected from multiple unclassified historical behaviors. This behavior will be used for association and matching with the unclassified behavior category. In user A's viewing history, suppose the target unclassified historical behavior currently being processed is that user A watched a science fiction movie titled "XXXXX". Multiple unclassified behavior categories refer to a predefined or data-based set of possible behavior categories or labels used to categorize the user's historical behavior. These categories are usually set based on the nature of the platform's content, user interests, or behavioral patterns. In the context of a video streaming platform, multiple unclassified behavior categories could include "watching science fiction movies," "watching action movies," "watching comedy movies," "watching historical dramas," etc. These categories are categorized by the platform based on the subject matter and type of content. Successfully associated initial candidate behavior categories refer to a set of initially selected behavior categories that have successfully established an association with the target historical behavior to be classified. These categories are selected from multiple pending behavior categories based on the characteristics and attributes of the target historical behavior to be classified, serving as the basis for further classification and configuration. For example, for user A's historical behavior of watching "XXXXX," the server will select "watching science fiction movies" and "watching highly-rated movies" from multiple pending behavior categories as initial candidate behavior categories for successful association, because this movie belongs to both the science fiction genre and could be a highly-rated movie.

[0032] Furthermore, multiple preliminary candidate behavior categories refer to a set of initially selected behavior categories associated with the target historical behavior to be classified. These categories are selected from multiple pending behavior categories and serve as the basis for further classification and configuration. For example, suppose user B searches for "sneakers" and browses related products on an e-commerce platform. In this case, "sneaker search" and "sneaker browsing" can be selected as preliminary candidate behavior categories for this user. Comparative behavior categories refer to some behavior categories with specific attributes or characteristics discovered through statistical analysis within the already assigned set of behavior types. These categories may have similarities to some preliminary candidate behavior categories, but also have certain differences, thus requiring special attention and differentiation. In content recommendation scenarios, if a user frequently watches science fiction movies and occasionally watches horror movies, then "science fiction movie watching" and "horror movie watching" can be identified as comparative behavior categories because although both behaviors involve watching movies, the content and user preferences can differ. Behavior category configuration refers to the process of assigning one or more final behavior categories to the target historical behavior to be classified based on preliminary candidate behavior categories and other relevant information. This process needs to consider various factors, including the characteristics of the preliminary candidate behavior categories and the existence of comparative behavior categories. For user B's search for "sneakers," the server will configure them as the "sneaker search" behavior category based on their behavioral characteristics and initial candidate behavior categories. Assigned type behaviors refer to historical behaviors that have already completed behavior category configuration. These behaviors have been assigned specific behavior categories, and these categories will be used for subsequent model training or content recommendation. During the user behavior recognition model training process, all historical behaviors that have completed behavior category configuration, such as "watching science fiction movies" and "searching for sneakers," belong to assigned type behaviors. The set of assigned type behaviors is a collection containing all assigned type behaviors. This set is dynamically updated; whenever a new historical behavior completes behavior category configuration, it is added to this set. This set is also used for statistical analysis to identify and compare behavior categories and other important features. In a content recommendation system, the set of assigned type behaviors can include all movie types watched by the user, all keywords searched, etc. This set is used not only to train the user behavior recognition model but also to optimize the recommendation algorithm and improve recommendation quality.

[0033] Furthermore, optimizing the assigned behavior set refers to the process of continuously improving and enhancing it. By constantly analyzing and adjusting the assigned behavior categories, the accuracy of classification is improved, making the set more consistent with users' actual behavior patterns and preferences. For example, the system might discover that some browsing behaviors of user C were incorrectly categorized as "browsing electronic products," when they should actually be "browsing home furnishings." Through optimization, the system can correct these misclassifications, making the assigned behavior set more accurate. Continuous classification processing is performed on all unclassified historical behaviors in the entire user history behavior set until all behaviors are assigned to specific behavior categories. This is an iterative and continuous process to ensure that each user behavior is correctly classified. In the user behavior analysis system of the e-commerce platform, the system continuously processes new user behavior data and classifies it into corresponding behavior categories. This process continues until all collected user behavior data is correctly classified. A dataset has been created where all user historical behaviors have been classified, and each behavior has been assigned a specific behavior category. In this dataset, each user behavior has been divided into a clear behavior category based on its behavioral characteristics, contextual information, etc. Suppose an online video platform collects a user's viewing history for one month and has already categorized this viewing history by behavior type, such as "watching science fiction movies" or "watching comedy shorts." Then, this categorized viewing history constitutes a set of the user's historical behavior data after the behavior type categorization process.

[0034] In this embodiment of the invention, when the plurality of primary candidate behavior categories include a comparison behavior category, the behavior category configuration of the target historical behavior to be classified is performed based on the plurality of primary candidate behavior categories to obtain the assigned type behavior of the target historical behavior to be classified. This can be implemented through the following example.

[0035] Obtain the comparison behavior category corresponding to the set of assigned type behaviors, and determine the category configuration startup requirements of the comparison behavior category;

[0036] When the comparison behavior category is included in the plurality of initial candidate behavior categories and meets the category configuration startup requirements, the behavior category is configured for the target historical behavior to be classified based on the plurality of initial candidate behavior categories, and the assigned type behavior of the target historical behavior to be classified is obtained.

[0037] In this embodiment of the invention, for example, the server first reviews and analyzes the set of assigned type behaviors. This set contains the user's past behavioral data, which has been categorized and assigned specific behavioral categories. In this step, the server pays particular attention to behaviors with obvious differences or special attributes, which may constitute contrasting behavioral categories. For example, in a video recommendation system, the server discovers that user A frequently watches science fiction movies but occasionally watches horror movies. Here, "watching science fiction movies" and "watching horror movies" can be identified as contrasting behavioral categories because, although they are both movie-watching behaviors, the content and user preferences are quite different. After determining the contrasting behavioral categories, the server further analyzes the characteristics and occurrence conditions of these categories to set the activation requirements for category configuration. These requirements may include the frequency of the behavior, the duration, and the correlation with other behaviors. Taking the video recommendation system as an example, the server might set it so that when user A frequently switches between watching science fiction movies and horror movies within a short period of time, the configuration of contrasting behavioral categories is triggered. Alternatively, if user A watches a science fiction movie and then immediately watches a horror movie, the repeated occurrence of this behavioral pattern will also satisfy the activation requirements. Once the server detects that a user's current behavior (i.e., the target historical behavior to be categorized) simultaneously meets the requirement that the initial candidate behavior category includes the comparison behavior category and satisfies the previously set category configuration activation requirements, the server will begin to perform more in-depth analysis and processing of this behavior. For example, if user A watches a science fiction movie and then immediately starts searching for information related to horror movies, the server will recognize that this behavior pattern meets the previously set activation requirements (i.e., frequently switching or continuously watching movies of different categories). After meeting the activation requirements, the server will perform more precise behavior category configuration for the target historical behavior to be categorized based on the characteristics of the initial candidate behavior categories and the user's current behavior. This process involves in-depth analysis of user behavior, such as identifying the keywords the user searches for, the content browsed, and the time spent on those pages. Taking user A as an example, the server will analyze information such as the horror movie keywords he searched, the movie detail pages he browsed, and the time he spent on those pages, thereby more accurately determining whether his current behavior belongs to the comparison behavior category of "watching horror movies." After in-depth analysis and behavior category configuration, the server will assign a final behavior category to the target historical behavior to be categorized and add it to the set of assigned type behaviors. This category will be used for subsequent content recommendation and model training. In user A's case, if the server determines that his current behavior does indeed fall under the contrasting behavior category of "watching horror movies," then this category will be assigned to this behavior and updated in the set of assigned behavior types. In the future, when user A exhibits a similar behavioral pattern again, the server will be able to more accurately recommend horror movie-related content to him.

[0038] In this embodiment of the invention, the behavior category distribution index of the assigned type behaviors configured as the comparison behavior category in the assigned type behavior set conforms to the non-routine action recognition criteria; when the comparison behavior category is included in the plurality of primary candidate behavior categories and meets the category configuration activation requirements, the behavior category is configured for the target historical behavior to be classified based on the plurality of primary candidate behavior categories to obtain the assigned type behavior of the target historical behavior to be classified, which can be implemented through the following example.

[0039] When the comparison behavior category is included in the plurality of initial candidate behavior categories, the comparison result corresponding to the comparison behavior category is obtained; the comparison result corresponding to the comparison behavior category is determined based on the category comparison between the target historical behavior to be classified and the comparison behavior category.

[0040] When the comparison result corresponding to the comparison behavior category meets the category configuration startup requirements, the target historical behavior to be classified is configured with behavior category based on the multiple initial candidate behavior categories to obtain the assigned type behavior of the target historical behavior to be classified.

[0041] In this embodiment of the invention, for example, when continuously monitoring and analyzing a set of assigned behavior types, the server discovers that the distribution of certain behavior categories differs significantly from typical user behavior patterns. These anomalous behavior distributions may indicate unconventional user actions or special interests. For example, User B typically watches educational videos at night, but suddenly starts watching horror movies frequently in the early morning. This abrupt change in behavior distribution conforms to the criteria for identifying unconventional actions, so the server marks "watching horror movies in the early morning" as a contrasting behavior category. When processing new behavior data from User B, the server's initial candidate behavior category identification module finds that User B's current behavior (such as browsing horror movies in the early morning) matches a previously marked contrasting behavior category. At this point, the server pays special attention to this behavior and considers it a potential contrasting behavior. To more accurately determine whether User B's current behavior truly belongs to a contrasting behavior category, the server performs further comparative analysis. For example, the server compares the content differences between the horror movies currently viewed by User B and the educational videos they previously watched, as well as changes in viewing time. These comparison results help the server more accurately understand the nature of user behavior. Based on the comparison results, if the server finds that user B's current behavior is significantly different from typical viewing habits, and this difference meets the previously set category configuration activation requirements (such as behavior frequency, time variation, content preferences, etc.), then the server will determine that this behavior belongs to the comparison behavior category. After confirming that user B's current behavior belongs to the comparison behavior category, the server will comprehensively configure user B's behavior based on this category and other relevant initial candidate behavior categories (such as "watching horror movies," "active user in the early morning," etc.). This configuration process may include assigning specific tags to the behavior and categorizing it into the corresponding behavior set. After completing the behavior category configuration, the server obtains the assigned type behavior for user B's current behavior, such as "unconventional user watching horror movies in the early morning." This assigned type behavior will be added to user B's behavior profile for subsequent content recommendation and personalized service provision. Simultaneously, this information can also be used to update the server's recommendation algorithm and model to improve the ability to identify and process similar behaviors.

[0042] In this embodiment of the invention, the following implementation methods are also provided.

[0043] When the comparison behavior category is not included in the plurality of initial candidate behavior categories, the conventional action recognition criteria are obtained.

[0044] When the comparison results corresponding to the multiple initial candidate behavior categories meet the conventional action recognition criteria, the target historical behavior to be classified is configured with behavior categories based on the multiple initial candidate behavior categories to obtain the assigned type behavior of the target historical behavior to be classified.

[0045] The comparison results corresponding to each of the multiple initial candidate behavior categories are determined by comparing the target historical behavior to be classified with the multiple initial candidate behavior categories.

[0046] In this embodiment of the invention, for example, when processing user C's historical behavior data, the server, through preliminary analysis, finds that user C's behavior pattern does not significantly deviate from the norm; that is, the multiple initial candidate behavior categories do not include the previously defined comparison behavior categories. For example, user C mainly watches comedy movies during the day and does not exhibit unusual behaviors such as watching horror movies in the early morning. Since user C's behavior does not show characteristics of the comparison behavior categories, the server will instead use conventional action recognition criteria to further analyze user behavior. These criteria may include conventional indicators such as the duration, frequency, and genre preference of videos watched by the user. The server compares user C's current behavior (e.g., watching a new comedy movie during the day) with multiple initial candidate behavior categories and finds that these comparison results all conform to the conventional action recognition criteria. This means that user C's behavior is stable and predictable, without any abnormalities or mutations. After confirming that user C's current behavior conforms to the conventional action recognition criteria, the server will configure user C's behavior according to these initial candidate behavior categories (e.g., "regular users watching comedy movies during the day"). The server will label this behavior as "regular comedy movie viewer," etc. After configuring the behavior categories, the server obtains the assigned behavior type for user C's current behavior, namely, "a regular user who watches comedy movies during the day." This information will be added to user C's behavior profile and used for subsequent content recommendations. For example, the server will recommend more comedy movies to user C to meet their regular viewing preferences. Through these steps, the server can accurately identify and configure users' regular behaviors, thereby providing users with more personalized and accurate content recommendation services.

[0047] In this embodiment of the invention, the step of determining multiple initial candidate behavior categories that are successfully associated with the target historical behavior to be classified from the multiple undetermined behavior categories can be implemented through the following example.

[0048] The target historical behavior to be classified is compared with the multiple undetermined behavior categories to obtain the comparison results corresponding to each of the multiple undetermined behavior categories;

[0049] Based on the comparison results corresponding to each of the multiple undetermined behavior categories, the multiple undetermined behavior categories are selected to obtain multiple initial candidate behavior categories that are successfully associated with the target unclassified historical behavior.

[0050] In this embodiment of the invention, for example, when user D watches a trailer for a science fiction movie on a video platform, this behavior is recorded and transmitted to the server. Upon receiving this new historical behavior data (i.e., the target unclassified historical behavior), the server compares it with multiple predefined unclassified behavior categories. For example, these unclassified behavior categories may include "watching a science fiction movie," "watching a horror movie," "watching a comedy movie," etc. The server assesses the correlation between the content attributes watched by user D (such as movie type, duration, and viewing time) and the characteristics of each unclassified behavior category. After the comparison is complete, the server obtains a similarity or correlation score between each unclassified behavior category and the target unclassified historical behavior. For example, the "watching a science fiction movie" category might receive a higher score because user D watched a science fiction movie trailer, while the "watching a horror movie" or "watching a comedy movie" categories might receive lower scores. Based on the comparison results, the server sets a threshold or sorts by score to select the unclassified behavior category most relevant to the target unclassified historical behavior. In this example, since user D is watching a science fiction movie trailer, the category "watching a science fiction movie" is likely to be selected as an initial candidate behavior category due to its high relevance. After the above filtering steps, the server finally determines the initial candidate behavior category most closely related to user D's current behavior (watching a science fiction movie trailer), namely "watching a science fiction movie." This category will be used for subsequent content recommendations, such as recommending more science fiction movies or science fiction-related content to user D. Through the above scenario example, we can see how the server determines the initial candidate behavior category most relevant to the user's current behavior by comparing and filtering potential behavior categories, thereby providing the user with more accurate content recommendations.

[0051] In this embodiment of the invention, the target historical behavior to be classified includes behavior data to be classified; the step of comparing the target historical behavior to be classified with the plurality of pending behavior categories to obtain the comparison results corresponding to each of the plurality of pending behavior categories can be implemented through the following example.

[0052] The behavior data to be classified is combined with the plurality of undetermined behavior categories to obtain combined behavior descriptions corresponding to each of the plurality of undetermined behavior categories;

[0053] Obtain a behavior matching model; the behavior matching model is obtained by optimizing the assigned type behaviors in the set of assigned type behaviors;

[0054] Based on the behavior matching model, behavior categories are associated with the combined behavior descriptions corresponding to each of the multiple undetermined behavior categories to obtain the comparison results corresponding to each of the multiple undetermined behavior categories.

[0055] In this embodiment of the invention, for example, user E watches an animated film on a video platform, and this behavior is recorded by the server. The server combines this unclassified behavior data (watching the animated film) with multiple undetermined behavior categories (such as "children's content consumer," "anime enthusiast," "users who occasionally watch animation," etc.). For example, combining "user E watches an animated film" with "children's content consumer" can form a combined behavior description: "User E watched an animated film as a children's content consumer." The server then obtains a behavior matching model. This model is obtained by training and optimizing a large number of assigned type behaviors in the previously assigned type behavior set. This model can analyze the correlation between user behavior and different behavior categories. The server uses the behavior matching model to analyze the correlation between each combined behavior description and the undetermined behavior category. For example, the model analyzes the degree of correlation between the combined behavior description "user E watched an animated film as a children's content consumer" and the "children's content consumer" category. Similarly, it also analyzes the degree of correlation between this behavior and categories such as "anime enthusiast" and "users who occasionally watch animation." The behavior matching model outputs a relevance score for each combined behavior description. This score reflects the degree of match between the behavior data to be classified and each potential behavior category. For example, "User E watches cartoons" may have a high relevance to "children's content consumers," a medium relevance to "anime enthusiasts," and a low relevance to "users who occasionally watch cartoons." Based on these comparisons, the server can determine which category User E's cartoon-watching behavior best fits. If the relevance to "children's content consumers" is the highest, the server will categorize this behavior into that category and recommend more children-related content to User E accordingly. Through this process, the server can accurately assign users' historical behaviors to the most appropriate behavior categories using the behavior matching model, thereby improving the accuracy of content recommendations and user satisfaction.

[0056] In this embodiment of the invention, the acquisition of the behavior matching model can be implemented through the following examples.

[0057] Obtain the behavior matching model to be optimized and the set of assigned type behaviors; the assigned type behaviors in the set of assigned type behaviors include classified behavior data and category attributes for the classified behavior data;

[0058] Based on the behavior matching model to be optimized, behavior category association is performed on the classified behavior data and the category attributes corresponding to the classified behavior data to obtain the comparison results of the classified behavior data.

[0059] The model parameters of the behavior matching model to be optimized are optimized based on the comparison results of the classified behavior data to obtain the behavior matching model.

[0060] In this embodiment of the invention, exemplarily, the server first acquires an initial or optimized behavior matching model. This model can be a machine learning-based classifier used to predict the category to which a user's behavior belongs. Simultaneously, the server also acquires a set of assigned type behaviors, containing a large amount of correctly classified user behavior data and the corresponding category attributes. For example, this set can contain information such as the type of video a user watches, viewing time, and user feedback, and this information has been labeled with category attributes such as "science fiction movie enthusiast" or "nighttime active user." The server uses the behavior matching model to be optimized to perform association analysis on the classified behavior data in the set of assigned type behaviors and their corresponding category attributes. This process is equivalent to letting the model "learn" how to correctly match user behavior with category attributes. For example, the model might try to predict whether a user belongs to the "science fiction movie enthusiast" category based on their historical data of watching science fiction movies. The comparison results reflect the accuracy of the model's predictions, i.e., the degree of match between the model's predicted category and the actual category attribute. Based on the comparison results, the server evaluates the performance of the behavior matching model. If the model's prediction accuracy is found to be insufficient, the server adjusts the model's parameters to improve its predictive ability. For example, if the model frequently makes errors in identifying the category "science fiction movie enthusiast," the server will adjust the model's weights, thresholds, or other relevant parameters to enable the model to more accurately identify this category in the next prediction. After multiple iterations and optimizations, the server will eventually obtain a performance-improved behavior matching model. This optimized model can more accurately predict the category attributes of users based on their behavior, thus providing stronger support for subsequent content recommendations. Through these steps, the server can continuously improve the accuracy of the behavior matching model, thereby enhancing the overall performance of the content recommendation system and providing users with more personalized and accurate content recommendation services.

[0061] In this embodiment of the invention, the step of selecting the plurality of undetermined behavior categories based on the comparison results corresponding to each of the plurality of undetermined behavior categories to obtain a plurality of initial candidate behavior categories that are successfully associated with the target unclassified historical behavior can be implemented through the following example.

[0062] Criteria for obtaining comparison results;

[0063] Based on the comparison results corresponding to each of the multiple undetermined behavior categories, multiple selected behavior categories that meet the selection criteria of the comparison results are selected from the multiple undetermined behavior categories;

[0064] Based on the selected behavior categories, multiple initial candidate behavior categories are obtained that are successfully associated with the target historical behavior to be classified.

[0065] In this embodiment of the invention, exemplarily, the server first obtains selection criteria for the comparison results. These criteria can be a series of specific thresholds, proportions, or other quantitative indicators used to determine which comparison results meet the requirements. For example, the criteria may stipulate that the similarity score of the comparison results must be higher than a certain value, or select the top N pending behavior categories, etc. The server applies these selection criteria to filter based on the comparison results corresponding to the multiple pending behavior categories obtained previously. For example, assuming user F watched a historical documentary on a video platform, the server will compare this behavior with pending behavior categories such as "history enthusiast," "documentary viewer," and "educational content consumer." The comparison results can be a series of similarity scores, and the server will filter out the selected behavior categories that meet the conditions based on these scores and selection criteria (such as a score higher than 0.8). After filtering, the server obtains multiple selected behavior categories that meet the selection criteria for the comparison results. These categories are the initial candidate behavior categories that are successfully associated with the target pending historical behavior (user F watching a historical documentary). For example, "history enthusiast" and "documentary viewer" may be selected as initial candidate behavior categories because of their high similarity scores. The server can now provide more accurate content recommendations to user F based on these initial candidate behavior categories. For example, based on the categories of "history enthusiast" and "documentary viewer," the server can recommend more history documentaries or related books to user F. Such recommendations not only match user F's interests but also enhance the user experience. Through these steps, the server can use comparison results selection criteria to filter out the initial candidate behavior categories most relevant to the user's historical behavior, thereby improving the accuracy of content recommendations and user satisfaction.

[0066] In this embodiment of the invention, the step of configuring the target historical behavior to be classified based on the plurality of initial candidate behavior categories to obtain the assigned type behavior of the target historical behavior to be classified can be implemented through the following example.

[0067] Obtain behavior classification models and behavior classification guidelines;

[0068] Based on the behavior classification model, the target historical behavior to be classified is configured with behavior categories according to the behavior classification guidelines and the multiple initial candidate behavior categories, and the behavior classification output of the target historical behavior to be classified is obtained.

[0069] Based on the behavior classification output, the assigned type of behavior of the target historical behavior to be classified is obtained.

[0070] In this embodiment of the invention, for example, the server first acquires a trained behavior classification model. This model can be built based on machine learning algorithms and is used to accurately classify user behavior. Simultaneously, the server also acquires behavior classification guidance, which can be a series of rules, standards, or prior knowledge to assist the model in making more accurate classifications. For example, classification guidance may include rules such as "If a user frequently watches science fiction movies, they are more likely to be classified as a 'science fiction movie enthusiast'." The server then uses the behavior classification model, combined with the behavior classification guidance and several previously acquired preliminary candidate behavior categories, to configure the behavior category for the target historical behavior to be classified. For example, assuming that the target historical behavior to be classified for user G is watching a science fiction movie on a video platform, the server will comprehensively determine the behavior category of user G based on the behavior classification model, classification guidance, and preliminary candidate behavior categories (such as "science fiction movie viewer," "movie enthusiast," etc.). After processing user G's behavior of watching a science fiction movie, the behavior classification model will output one or more possible behavior categories as the classification result. This output is a comprehensive judgment based on the model's learning results and classification guidance. For example, the model may output "science fiction movie viewer" as the primary behavior category for user G. Based on the output of the behavioral classification model, the server determines that user G's act of watching science fiction movies belongs to the category of "science fiction movie viewer." This assigned type of behavior will be recorded and used for subsequent content recommendations. For example, based on this classification result, the server will recommend more science fiction movies or science fiction-related content to user G. Through the above steps, the server can accurately classify users' historical behaviors using behavioral classification models and behavioral classification guidelines, thereby obtaining assigned type behaviors and providing strong support for subsequent personalized recommendations.

[0071] In this embodiment of the invention, the acquisition of the behavior classification model and behavior classification guidance can be implemented through the following examples.

[0072] Obtain sample user behavior and determine the initial behavior classification model and reference classification guidelines;

[0073] Based on the initial behavior classification model, and according to the reference classification guidelines and the behavior category configuration for the sample user behavior, the sample behavior classification output of the sample user behavior is obtained;

[0074] When the sample behavior classification output matches the category target value of the sample user behavior, a behavior classification model is obtained based on the initial behavior classification model, and behavior classification guidance is obtained based on the reference classification guidance.

[0075] In this embodiment of the invention, exemplarily, the server first collects a large amount of sample user behavior data. This data may include the types of videos watched, the content of web pages browsed, the types of goods purchased, etc. Simultaneously, the server determines an initial behavior classification model, which can be a basic machine learning classifier, such as a decision tree or support vector machine. Furthermore, the server develops a set of reference classification guidelines, which are preliminary rules or standards used to guide the model on how to initially classify user behavior. Next, the server uses the initial behavior classification model, combined with the reference classification guidelines, to configure behavior categories for the collected sample user behavior. For example, for user H's viewing history (including watching a large number of comedy movies), the model attempts to predict user H's behavior category based on these behaviors and the reference classification guidelines. This process produces a sample behavior classification output, i.e., the model's prediction of user H's behavior. The server compares the model's predicted sample behavior classification output with the actual target value of the user behavior's category. The target value is pre-labeled and represents the true category of the user behavior. For example, if user H is actually a "comedy movie enthusiast," then this is their target value. The server compares whether the model's prediction (sample behavior classification output) matches this target value. If the model's predictions match the target category value, the initial behavior classification model and the reference classification guidance are considered effective. In this case, the server will adopt this validated initial behavior classification model as the final behavior classification model and the reference classification guidance as the final behavior classification guidance. These models and guidance will be used for subsequent user behavior classification and content recommendation. Through these steps, the server can validate and optimize the behavior classification model and guidance based on actual user behavior data, thereby improving its ability to accurately classify user behavior and ultimately enhancing the accuracy of content recommendation and user satisfaction.

[0076] In this embodiment of the invention, when the sample behavior classification output matches the category target value of the sample user behavior, a behavior classification model is obtained based on the initial behavior classification model, and a behavior classification guide is obtained based on the reference classification guide. This can be implemented through the following example.

[0077] Obtain the category target value of the sample user behavior; the category target value includes multiple candidate target values ​​obtained by configuring behavior categories based on different behavior classification strategies for the sample user behavior;

[0078] Determine the first classification deviation between the sample behavior classification output and the plurality of candidate target values, and the second classification deviation between the plurality of candidate target values;

[0079] When the first classification bias matches the second classification bias, the initial behavior classification model is determined as the behavior classification model, and the reference classification guidance is determined as the behavior classification guidance.

[0080] In this embodiment of the invention, exemplarily, the server first obtains the category target values ​​of sample user behavior. These category target values ​​are multiple candidate target values ​​obtained by configuring the sample user behavior into categories using different behavior classification strategies. For example, for user I's shopping behavior, there are two classification strategies: one is based on shopping time (e.g., daytime shopping or nighttime shopping), and the other is based on shopping type (e.g., purchasing electronics or purchasing household goods). These two strategies will generate different candidate target values ​​for the same shopping behavior of user I. The server then calculates the first classification bias between the sample behavior classification output and the multiple candidate target values. This reflects the difference between the model prediction and the actual category target. At the same time, the server also calculates the second classification bias between the multiple candidate target values, which reflects the differences between different classification strategies. Taking user I as an example, the first classification bias may be the deviation between the model predicting user I as a "daytime shopper" and the actual candidate target value being "nighttime shopper"; the second classification bias may be the difference between the two candidate target values ​​of "daytime shopper" and "purchaser of electronics". The server compares the first classification bias and the second classification bias to determine whether the model's prediction bias is within an acceptable range, i.e., whether it "matches". If the first classification bias is close to or less than a certain preset threshold compared to the second classification bias, it indicates that the model's predictive ability is comparable to the actual classification strategy, and it can be considered effective. When the first classification bias matches the second classification bias, the server determines the initial behavior classification model as the final behavior classification model and the reference classification guidance as the final behavior classification guidance. This means that the model and guidance have passed validation based on actual user behavior and can be used for subsequent user behavior classification and content recommendation. For example, if the model's prediction bias for user I's shopping behavior matches the differences between different classification strategies, then the server will use this model and guidance for subsequent user behavior classification. Through these steps, the server can validate and optimize the accuracy of the behavior classification model based on actual user behavior data and multiple classification strategies, thereby improving the accuracy of content recommendation and user satisfaction.

[0081] In this embodiment of the invention, the process of obtaining the assigned type behavior of the target historical behavior to be classified based on the behavior classification output can be implemented through the following example.

[0082] Obtain the behavior classification output update criteria;

[0083] The behavior classification output is optimized based on the behavior classification output update criterion to obtain the optimized behavior classification output.

[0084] Based on the optimized behavior classification output, the assigned type behavior of the target historical behavior to be classified is obtained.

[0085] In this embodiment of the invention, exemplarily, the server first obtains update criteria for the behavior classification output. These criteria relate to requirements such as the accuracy, stability, and real-time performance of the classification output. For example, the criteria may stipulate that the behavior classification output needs to be updated when new user behavior data appears, or when the uncertainty of the model prediction exceeds a certain threshold. Based on these update criteria, the server optimizes the initial behavior classification output. For example, suppose user J recently watched a series of suspense movies, and the initial behavior classification output categorizes him / her as a "movie enthusiast." However, according to the update criteria, if the user has recently shown a strong preference for a certain type of content, his / her behavior should be classified more precisely. Therefore, the server optimizes user J's behavior classification output based on his / her recent viewing history, classifying him / her more accurately as a "suspense movie enthusiast." After optimization, the server obtains a more accurate and detailed behavior classification output. In this example, "suspense movie enthusiast" is the optimized behavior classification output, which more accurately reflects user J's current interests and preferences. Finally, based on the optimized behavior classification output, the server assigns the assigned type behavior, "suspense movie enthusiast," to user J's target unclassified historical behavior (watching suspense movies). This classification will be recorded and used for subsequent content recommendations, providing users with more accurate and personalized recommendations. Through these steps, the server can continuously optimize and improve user behavior classification by using the behavioral classification output update criteria, thereby more accurately capturing changes in user interests and preferences, and improving the accuracy of content recommendations and user satisfaction.

[0086] In this embodiment of the invention, the process of optimizing the set of assigned type behaviors based on the one-to-one correspondence between the plurality of unclassified historical behaviors and configuring the behavior categories of the unclassified historical behaviors in the set of user historical behaviors until the behavior type classification process is completed for all unclassified historical behaviors in the set of user historical behaviors can be implemented through the following example.

[0087] Add the assigned type behaviors corresponding to the multiple unclassified historical behaviors to the set of assigned type behaviors, and repeat the step of obtaining multiple unclassified historical behaviors from the set of user historical behaviors until the behavior type classification process is completed for all unclassified historical behaviors in the set of user historical behaviors.

[0088] In this embodiment of the invention, for example, the server first creates an empty "assigned type behavior set," which will be used to store categorized behaviors from the user's historical behavior. The server retrieves a batch of uncategorized historical behaviors from the "user historical behavior set." For example, these behaviors may include multiple product pages recently viewed by user K, the types of videos watched, etc. For each uncategorized historical behavior, the server assigns a type to it based on a previously trained behavior classification model and behavior classification guidelines. For example, if user K viewed multiple product pages for sneakers, the server would categorize this behavior as "sneaker enthusiast." The server adds the categorized uncategorized historical behaviors (i.e., assigned type behaviors) to the "assigned type behavior set." In this example, the behavior type "sneaker enthusiast" will be added to the set. The server repeats the above steps, continuously retrieving new uncategorized historical behaviors from the "user historical behavior set," assigning them types, and then updating the "assigned type behavior set." This process continues until all uncategorized historical behaviors in the "user historical behavior set" have been categorized. Once all unclassified historical behaviors have been categorized and added to the "Assigned Type Behavior Set," the server has completed the categorization of all of user K's historical behaviors. At this point, the "Assigned Type Behavior Set" contains various types of user K's historical behaviors, such as "sneaker enthusiast" and "electronics consumer," which will provide important reference for subsequent content recommendations. Through these steps, the server can systematically process and categorize users' historical behaviors, thereby building a comprehensive and accurate behavioral profile for each user, laying a solid foundation for providing personalized recommendation services.

[0089] This invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned content recommendation method based on big data and artificial intelligence. Figure 2 As shown, Figure 2 This is a structural block diagram of a computer device 100 provided in an embodiment of the present invention. The computer device 100 includes a memory 111, a processor 112, and a communication unit 113. To enable data transmission or interaction, the memory 111, processor 112, and communication unit 113 are electrically connected to each other directly or indirectly. For example, these components can be electrically connected to each other through one or more communication buses or signal lines.

[0090] For illustrative purposes, the foregoing description has been made with reference to specific embodiments. However, the foregoing illustrative discussions are not intended to be exhaustive or to limit the present disclosure to the precise forms disclosed. Numerous modifications and variations are possible in accordance with the foregoing teachings. These embodiments were chosen and described in order to best illustrate the principles of the present disclosure and its practical application, thereby enabling those skilled in the art to best utilize the disclosure and to employ various embodiments with different modifications to suit a particular intended application.

Claims

1. A content recommendation method based on big data and artificial intelligence, characterized in that, include: Obtain at least one current user behavior of the target user within a preset monitoring period; The at least one current user behavior is input into a pre-trained user behavior recognition model to obtain the user behavior recognition result corresponding to the at least one current user behavior, wherein the user behavior recognition model is trained based on a set of user historical behaviors as the training set. Based on the user behavior recognition results, a target user profile of the target user is constructed; Based on the target user profile, content recommendations are made for the target user; The user behavior recognition model is obtained through the following methods: Multiple historical behaviors to be classified are obtained from the user's historical behavior set, and multiple undetermined behavior categories are determined; For each of the multiple unclassified historical behaviors, the unclassified behavior data is combined with the multiple undetermined behavior categories to obtain a combined behavior description corresponding to each of the multiple undetermined behavior categories. Obtain the behavior matching model to be optimized and the set of assigned type behaviors; the assigned type behaviors in the set of assigned type behaviors include classified behavior data and category attributes for the classified behavior data; Based on the behavior matching model to be optimized, behavior category association is performed on the classified behavior data and the category attributes corresponding to the classified behavior data to obtain the comparison results of the classified behavior data. The model parameters of the behavior matching model to be optimized are optimized based on the comparison results of the classified behavior data to obtain the behavior matching model; the behavior matching model is obtained by optimizing the assigned type behaviors in the set of assigned type behaviors. Based on the behavior matching model, behavior categories are associated with the combined behavior descriptions corresponding to each of the multiple undetermined behavior categories to obtain the comparison results corresponding to each of the multiple undetermined behavior categories; Based on the comparison results corresponding to each of the multiple undetermined behavior categories, the multiple undetermined behavior categories are selected to obtain multiple initial candidate behavior categories that are successfully associated with the target undetermined historical behavior; the target undetermined historical behavior includes the undetermined behavior data; Obtain the comparison behavior category corresponding to the set of assigned type behaviors, and determine the category configuration startup requirements of the comparison behavior category; the distribution index of the assigned type behaviors configured as the comparison behavior category in the set of assigned type behaviors conforms to the non-routine action recognition criteria; When the comparison behavior category is included in the plurality of initial candidate behavior categories, the comparison result corresponding to the comparison behavior category is obtained; the comparison result corresponding to the comparison behavior category is determined based on the category comparison between the target historical behavior to be classified and the comparison behavior category. When the comparison result corresponding to the comparison behavior category meets the category configuration startup requirements, the target historical behavior to be classified is configured with behavior categories based on the multiple initial candidate behavior categories to obtain the assigned type behavior of the target historical behavior to be classified; the comparison behavior category is obtained based on the assigned type behaviors in the set of assigned type behaviors. The set of assigned type behaviors is optimized based on the one-to-one correspondence between the multiple unclassified historical behaviors, and the unclassified historical behaviors in the set of user historical behaviors are configured with behavior categories until the behavior type classification process is completed for all unclassified historical behaviors in the set of user historical behaviors. The user's historical behavior set, after completing the behavior type classification process, is used as the training set to train the initial user behavior recognition model, thus obtaining the trained user behavior recognition model.

2. The method according to claim 1, characterized in that, The method further includes: When the comparison behavior category is not included in the plurality of initial candidate behavior categories, the conventional action recognition criteria are obtained. When the comparison results corresponding to the multiple initial candidate behavior categories meet the conventional action recognition criteria, the target historical behavior to be classified is configured with behavior categories based on the multiple initial candidate behavior categories to obtain the assigned type behavior of the target historical behavior to be classified. The comparison results corresponding to each of the multiple initial candidate behavior categories are determined by comparing the target historical behavior to be classified with the multiple initial candidate behavior categories.

3. The method according to claim 1, characterized in that, The step of selecting multiple undetermined behavior categories based on the comparison results corresponding to each category, and obtaining multiple initial candidate behavior categories successfully associated with the target historical behavior to be classified, includes: Criteria for obtaining comparison results; Based on the comparison results corresponding to each of the multiple undetermined behavior categories, multiple selected behavior categories that meet the selection criteria of the comparison results are selected from the multiple undetermined behavior categories; Based on the selected behavior categories, multiple initial candidate behavior categories are obtained that are successfully associated with the target historical behavior to be classified.

4. The method according to claim 1, characterized in that, The step of configuring behavior categories for the target historical behavior to be classified based on the multiple initial candidate behavior categories, and obtaining the assigned type behavior of the target historical behavior to be classified, includes: Obtain behavior classification models and behavior classification guidelines; Based on the behavior classification model, the target historical behavior to be classified is configured with behavior categories according to the behavior classification guidelines and the multiple initial candidate behavior categories, and the behavior classification output of the target historical behavior to be classified is obtained. Based on the behavior classification output, the assigned type of behavior of the target historical behavior to be classified is obtained.

5. The method according to claim 4, characterized in that, The acquisition of the behavior classification model and behavior classification guidance includes: Obtain sample user behavior and determine the initial behavior classification model and reference classification guidelines; Based on the initial behavior classification model, and according to the reference classification guidelines and the behavior category configuration for the sample user behavior, the sample behavior classification output of the sample user behavior is obtained; When the sample behavior classification output matches the category target value of the sample user behavior, a behavior classification model is obtained based on the initial behavior classification model, and behavior classification guidance is obtained based on the reference classification guidance.

6. The method according to claim 5, characterized in that, When the sample behavior classification output matches the category target value of the sample user behavior, a behavior classification model is obtained based on the initial behavior classification model, and behavior classification guidance is obtained based on the reference classification guidance, including: Obtain the category target value of the sample user behavior; the category target value includes multiple candidate target values ​​obtained by configuring behavior categories based on different behavior classification strategies for the sample user behavior; Determine the first classification deviation between the sample behavior classification output and the plurality of candidate target values, and the second classification deviation between the plurality of candidate target values; When the first classification bias matches the second classification bias, the initial behavior classification model is determined as the behavior classification model, and the reference classification guidance is determined as the behavior classification guidance.

7. The method according to claim 6, characterized in that, The process of obtaining the assigned type of behavior of the target historical behavior to be classified based on the behavior classification output includes: Obtain the behavior classification output update criteria; The behavior classification output is optimized based on the behavior classification output update criterion to obtain the optimized behavior classification output. Based on the optimized behavior classification output, the assigned type behavior of the target historical behavior to be classified is obtained.

8. The method according to claim 1, characterized in that, The process involves optimizing the set of assigned type behaviors based on the one-to-one correspondence between the multiple unclassified historical behaviors, and configuring behavior categories for the unclassified historical behaviors in the user's historical behavior set, until behavior type classification is completed for all unclassified historical behaviors in the user's historical behavior set. This includes: Add the assigned type behaviors corresponding to the multiple unclassified historical behaviors to the set of assigned type behaviors, and repeat the step of obtaining multiple unclassified historical behaviors from the set of user historical behaviors until the behavior type classification process is completed for all unclassified historical behaviors in the set of user historical behaviors.

9. A server system, characterized in that, Includes a server, the server being used to perform the method according to any one of claims 1-8.

Images