Content recommendation method and device, computer device, readable storage medium and program product

Abstract

This application relates to a content recommendation method, apparatus, computer device, computer-readable storage medium, and computer program product. The method includes: responding to a target trigger operation on a target interface; obtaining a target behavior identifier corresponding to the target trigger operation and an original content list corresponding to the target interface, the original content list containing at least one candidate content; determining a semantic description text of the corresponding target historical behavior based on the target behavior identifier and a user identifier of the target user performing the target trigger operation, and generating a prompt message based on the semantic description text and the original content list; inputting the prompt message into a large language model to obtain a user preference degree for each candidate content; and generating a target content list based on the user preference degree for each candidate content. This method accurately matches user needs, effectively improving the accuracy and adaptability of content recommendations.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, apparatus, computer device, computer-readable storage medium, and computer program product. Background Technology

[0002] With the development of information technology, the application of recommendation scenarios has become increasingly widespread, covering multiple fields such as product recommendations on e-commerce platforms, entertainment ranking information push, and financial channel financial product recommendations.

[0003] In related technologies, recommendation systems typically present homogeneous content lists to all users, lacking specific consideration for individual user characteristics, behavioral preferences, or profile attributes. Taking financial product recommendations as an example, the system often displays the same product series to different users, which not only affects the user experience but also reduces the accuracy and conversion efficiency of product recommendations. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer device, computer-readable storage medium, and computer program product that can effectively improve the accuracy and suitability of content recommendations, addressing the aforementioned technical problems.

[0005] Firstly, this application provides a method for recommending content, including:

[0006] In response to a target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content;

[0007] Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate prompt information based on the semantic description text and the original content list;

[0008] The prompt information is input into a large language model to obtain the user preference level for each candidate content;

[0009] A list of target content is generated based on the user preference for each candidate content.

[0010] In one embodiment, determining the semantic description text of the corresponding target historical behavior based on the target behavior identifier and the user identifier of the target user performing the target triggering operation includes:

[0011] Query the database for the semantic description text of the target's historical behavior corresponding to the target behavior identifier;

[0012] Filter semantic description text associated with the user identifier from the semantic description text of the target's historical behavior.

[0013] In one embodiment, querying the database for the semantic description text of the target historical behavior corresponding to the target behavior identifier includes:

[0014] Based on the mapping relationship between behavior identifiers and behavior categories, the target behavior category to which the target behavior identifier belongs is determined;

[0015] Query the database for the semantic description text set corresponding to the target behavior category;

[0016] Select semantic description texts that correspond to the target behavior identifier from the set of semantic description texts.

[0017] In one embodiment, before determining the semantic description text of the corresponding target historical behavior based on the target behavior identifier and the user identifier of the target user performing the target triggering operation, the method further includes:

[0018] In response to a trigger operation on any interface, obtain the behavior data corresponding to the trigger operation, wherein the behavior data includes user identifier, behavior identifier, behavior parameters and timestamp;

[0019] The behavioral data is processed using natural language processing and intent recognition to generate corresponding semantic description text.

[0020] The behavioral data and corresponding semantic description text are stored in the database according to their respective behavioral categories.

[0021] In one embodiment, the step of performing natural language processing and intent recognition processing on the behavioral data to generate corresponding semantic description text includes:

[0022] The behavioral data is subjected to natural language processing to generate a text describing the behavioral facts;

[0023] The behavioral data is processed for intent recognition to generate an intent inference description text;

[0024] The behavioral fact description text and the intent inference description text are fused together to generate the corresponding semantic description text.

[0025] In one embodiment, generating prompt information based on the semantic description text and the original content list includes:

[0026] Each candidate content is processed by content feature extraction to obtain the content features of each candidate content, and a semantic description text of the candidate content is generated based on the content features;

[0027] The semantic description text of the target's historical behavior is integrated with the semantic description text of the candidate content to generate a prompt message.

[0028] In one embodiment, generating a target content list based on the user preference for each candidate content includes:

[0029] In response to the existence of at least two candidate contents with the same user preference, the candidate contents are arranged according to a preset priority rule;

[0030] After reordering each candidate content according to user preference and the preset priority rules, a target content list is generated.

[0031] Secondly, this application also provides a content recommendation device, comprising:

[0032] The acquisition module is used to respond to a target trigger operation in the target interface, acquire the target behavior identifier corresponding to the target trigger operation, and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content;

[0033] The determination module is used to determine the semantic description text of the corresponding target historical behavior based on the target behavior identifier and the user identifier of the target user who performs the target trigger operation, and to generate prompt information based on the semantic description text and the original content list;

[0034] The input module is used to input the prompt information into the large language model to obtain the user preference degree of each candidate content;

[0035] The generation module is used to generate a list of target content based on the user preference for each candidate content.

[0036] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0037] In response to a target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content;

[0038] Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate prompt information based on the semantic description text and the original content list;

[0039] The prompt information is input into a large language model to obtain the user preference level for each candidate content;

[0040] A list of target content is generated based on the user preference for each candidate content.

[0041] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0042] In response to a target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content;

[0043] Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate prompt information based on the semantic description text and the original content list;

[0044] The prompt information is input into a large language model to obtain the user preference level for each candidate content;

[0045] A list of target content is generated based on the user preference for each candidate content.

[0046] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0047] In response to a target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content;

[0048] Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate prompt information based on the semantic description text and the original content list;

[0049] The prompt information is input into a large language model to obtain the user preference level for each candidate content;

[0050] A list of target content is generated based on the user preference for each candidate content.

[0051] The aforementioned recommendation methods, apparatus, computer devices, computer-readable storage media, and computer program products, in response to a target trigger operation on a target interface, acquire a target behavior identifier corresponding to the target trigger operation and a list of original content corresponding to the target interface. The original content list contains at least one candidate content. Then, based on the target behavior identifier and the user identifier of the target user executing the target trigger operation, the semantic description text of the corresponding target historical behavior is determined. Based on the semantic description text and the original content list, a prompt message is generated. This prompt message is then input into a large language model to obtain the user preference degree for each candidate content. Finally, a target content list is generated based on the user preference degree of each candidate content. Therefore, in the content recommendation process, the semantic description text of the target historical behavior determined by the user's real-time trigger behavior can fully explore user behavior facts and preference intentions. Combined with the deep semantic analysis capabilities of a large language model, it can accurately match user needs, achieving personalized and differentiated content recommendations, thereby effectively improving the accuracy and adaptability of content recommendations. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart illustrating a content recommendation method in one embodiment;

[0054] Figure 2 This is a flowchart illustrating the process of generating prompt information in one embodiment;

[0055] Figure 3 This is a flowchart illustrating a content recommendation method in another embodiment;

[0056] Figure 4 This is a flowchart illustrating the content recommendation method in yet another embodiment;

[0057] Figure 5 This is a structural block diagram of a device for recommending content in one embodiment;

[0058] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0061] In one embodiment, such as Figure 1 As shown, a content recommendation method is provided. This embodiment illustrates the method applied to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0062] Step 102: In response to the target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface. The original content list contains at least one candidate content.

[0063] The target interface can be understood as a specific functional interface on the terminal that is triggered by user interaction, such as the product list page of an e-commerce platform, the news page of an entertainment information platform, or the financial product section of a financial application. This application does not limit this.

[0064] In addition, the target trigger operation can be the interactive behavior performed by the user on the target interface, which is collected and reported by the terminal device, such as clicking, refreshing, etc. This application does not limit this.

[0065] Furthermore, the target behavior identifier can correspond to the target triggering operation and be used to characterize the specific interactive action. Different target behavior identifiers can be associated with different sorting or recommendation optimization goals. For example, when the target triggering operation is "refreshing the financial product page", the corresponding target behavior identifier can be "REFRESH_FINANCE_01", and the associated recommendation optimization goal can be "promoting clicks on financial products"; when the target triggering operation is "the product details page has finished loading", the corresponding target behavior identifier can be "LOAD_PROD_DETAIL_02", and the associated recommendation optimization goal can be "promoting user purchase behavior", etc. This application does not limit this.

[0066] Furthermore, the original content list can be understood as an unsorted list of candidate content, which may include at least one candidate item, and the candidate items may correspond to the target interface. For example, on an e-commerce platform page, the original content list may be a list of candidate products; on a financial product page, the original content list may be a list of candidate financial products, and so on. This application does not limit this.

[0067] Step 104: Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate a prompt message based on the semantic description text and the original content list.

[0068] The user identifier can be used to represent the target user who performs the target-triggered operation. Its presentation format can be diverse, such as numbers or strings, and this application does not limit it.

[0069] Furthermore, semantic descriptive text can be understood as natural language descriptive text generated after processing users' historical behavioral data, which can reflect users' behavioral facts and preferences. For example, for the behavior of "clicking on high-risk financial products," the semantic descriptive text could be "The user actively clicked on financial products marked as high-risk. This behavior indicates that the user may be interested in high-yield investment opportunities and has a certain tendency to explore risk tolerance," etc. This application does not limit this.

[0070] Optionally, there are several ways to determine the semantic description text of the target historical behavior. For example, you can query the semantic description text of the target historical behavior corresponding to the target behavior identifier in the database, and then filter the semantic description text associated with the user identifier from the semantic description text of the target historical behavior.

[0071] The database can store semantic description text of the historical behavior of all users within the platform or system. This semantic description text can be classified and stored according to the behavior identifier dimension or the user identifier dimension, and the storage format can be a data table, a document, etc. This application does not limit this.

[0072] Therefore, in this embodiment, a search can be performed in the database based on the target behavior identifier to obtain the semantic description text of the target historical behavior of all users corresponding to the target behavior identifier. Then, the semantic description text associated with the current user identifier can be further filtered out. Thus, by using both behavior identifier and user identifier for dual retrieval, the efficiency and accuracy of data retrieval are effectively improved.

[0073] Optionally, one can first determine the target behavior category to which the target behavior identifier belongs based on the mapping relationship between behavior identifiers and behavior categories, then query the semantic description text set corresponding to the target behavior category in the database, and then filter the semantic description text corresponding to the target behavior identifier from the semantic description text set.

[0074] To facilitate data classification, storage, and rapid retrieval, semantic description text can be stored in the database according to behavioral categories. Furthermore, the mapping relationship between behavioral identifiers and behavioral categories can be pre-defined. Within the same behavioral category, semantic description text corresponding to the same type of historical behavior generated at different times can be stored.

[0075] Therefore, in this embodiment, the target behavior identifier can first be matched and searched within a preset mapping relationship between behavior identifiers and behavior categories to determine the target behavior category to which the target behavior identifier belongs. Then, the corresponding semantic description text set can be retrieved from the database using the target behavior category as an index, and finally, the semantic description text corresponding to the target behavior identifier can be filtered out. Thus, by using partitioned retrieval, a complete traversal of the database is avoided, saving retrieval time and effectively improving query performance.

[0076] Therefore, in this embodiment of the application, the semantic description text of the target's historical behavior can be determined in multiple ways, which enriches the methods for determining the semantic description text, saves processing time, and effectively improves the efficiency of obtaining the semantic description text.

[0077] Furthermore, after determining the semantic description text, it can be processed with the original content list to generate a prompt message. For example, the semantic description text can be concatenated with the original content list to obtain the prompt message; or the original content list can be presented in document form, and the semantic description text can be integrated with the document to obtain the prompt message, etc. This application does not limit this approach.

[0078] It is understandable that since semantic description text can reflect a user's historical behavior and intention preferences, and the original content list can represent candidate content information on the target page, the prompt information generated based on the semantic description text and the original content list can effectively represent the user's preference description and contain the characteristics of the candidate content. Therefore, using the prompt information for subsequent processing can effectively improve the accuracy and reliability of the processing.

[0079] Step 106: Input the prompt information into the large language model to obtain the user preference for each candidate content.

[0080] Among them, the Large Language Model (LLM) can be any model with natural language understanding, semantic analysis and reasoning capabilities, which can effectively identify the user's true potential intent. This application does not limit the specific structure, parameter scale and training method of the large language model.

[0081] Optionally, the large language model can be deployed on the local side of the mobile device or on the server side, and can be called through the server-side LLM API, etc. This application does not limit this.

[0082] In addition, user preference score can characterize the degree of matching between candidate content and user preferences. The higher the user preference score, the more the candidate content meets the user's needs. The lower the user preference score, the greater the deviation between the candidate content and the user's needs.

[0083] Therefore, in this embodiment of the application, the prompt information can be input into the large language model. After semantic analysis and reasoning processing by the large language model, the user preference degree of each candidate content can be obtained, which can provide a basis for the re-ranking of subsequent candidate content and effectively improve the matching degree of recommendation results and user satisfaction.

[0084] Step 108: Generate a list of target content based on the user preference for each candidate content.

[0085] Specifically, all candidate content in the original content list can be rearranged according to user preference from highest to lowest to generate a target content list for display. This ensures that the presentation order of the target content in the target content list effectively matches user preferences, better aligning with the user's current needs and interests. Consequently, it can gain priority in attracting user attention during display, improving the accuracy of content recommendations and ultimately achieving efficient personalized recommendations.

[0086] It is understandable that in the process of generating a list of target content based on the user preference of each candidate content, there may be multiple candidate content with the same user preference. In this case, priority rules can be used to further sort them.

[0087] Optionally, in response to the existence of at least two candidate contents with the same user preference, the candidate contents can be arranged according to a preset priority rule. Then, after reordering each candidate contents according to user preference and the preset priority rule, a target content list can be generated.

[0088] The preset priority rules may include time priority, click rate priority, etc., and this application does not limit them.

[0089] For example, if candidate content 1 and candidate content 2 have the same user preference, but candidate content 1 was published earlier and candidate content 2 was published more recently, then it can be determined that candidate content 2 has a higher priority than candidate content 1, that is, candidate content 2 comes first and candidate content 1 comes later, and so on. This application does not limit this.

[0090] In the aforementioned recommendation method, in response to a target trigger operation on the target interface, the method obtains the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface. The original content list contains at least one candidate content. Then, based on the target behavior identifier and the user identifier of the target user executing the target trigger operation, the semantic description text of the corresponding target historical behavior is determined. Based on the semantic description text and the original content list, a prompt message is generated. This prompt message is then input into a large language model to obtain the user preference score for each candidate content. Finally, a target content list is generated based on the user preference score for each candidate content. Therefore, in the content recommendation process, the semantic description text of the target historical behavior determined by the user's real-time trigger behavior can fully explore user behavior facts and preference intentions. Combined with the deep semantic analysis capabilities of the large language model, it can accurately match user needs, achieving personalized and differentiated content recommendations, thereby effectively improving the accuracy and adaptability of content recommendations.

[0091] In one exemplary embodiment, such as Figure 2 As shown, the steps for generating prompt information based on semantic description text and the original content list include steps 202 to 204. Among them:

[0092] Step 202: Extract content features from each candidate content to obtain the content features of each candidate content, and generate semantic description text of the candidate content based on the content features.

[0093] Content feature extraction can be understood as identifying and extracting textual information that can represent the core content from the raw data of candidate content. Furthermore, there are various methods for content feature extraction, such as extraction based on preset rules or templates, or text feature extraction using natural language processing models, etc. This application does not limit these methods.

[0094] It is understandable that after obtaining the content features of each candidate content, the content features of each candidate content can be fused to generate a semantic description text covering all candidate content; or the content features of each candidate content can be processed separately using natural language to generate a semantic description text corresponding to each candidate content, etc. This application does not limit this.

[0095] Step 204: Integrate the semantic description text of the target's historical behavior with the semantic description text of the candidate content to generate a prompt message.

[0096] The semantic description text of the target user's historical behavior reflects their historical behavior and intention preferences, while the semantic description text of the candidate content represents their core attributes and key features. Therefore, by integrating the semantic description text of the target user's historical behavior with that of the candidate content, the generated prompt information can effectively represent the user's preferences and accurately reflect the characteristics of the candidate content.

[0097] In addition, there are several ways to integrate and process the information. For example, the semantic description text of the target's historical behavior can be concatenated with the semantic description text of the candidate content, and the resulting text is the prompt message. Alternatively, the two can be filled into the specified positions in the template to generate the final prompt message.

[0098] In this embodiment, by extracting content features from each candidate content, the content features of each candidate content can be obtained. Semantic description text of the candidate content is then generated based on these content features. Finally, the semantic description text of the target historical behavior is integrated with the semantic description text of the candidate content to generate a prompt message. Therefore, by integrating the semantic description text of the target historical behavior with the semantic description text of the candidate content, the generated prompt message simultaneously includes both user preference intent and candidate content features, providing complete contextual information for the large language model. This improves the accuracy and reliability of the prompt message and provides a strong foundation for ensuring the accuracy of subsequent content recommendations.

[0099] It is understandable that before determining the semantic description text of the corresponding target historical behavior based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, the user's historical behavior data can be recorded to obtain the corresponding semantic description text.

[0100] In one embodiment, in response to a trigger operation on any interface, behavioral data corresponding to the trigger operation can be obtained. This behavioral data includes a user identifier, a behavior identifier, behavior parameters, and a timestamp. The behavioral data can then undergo natural language processing and intent recognition to generate corresponding semantic description text. The behavioral data and its corresponding semantic description text are then stored in a database according to their respective behavior categories.

[0101] This can be achieved by collecting users' historical behavior data in advance, such as through client-side event tracking or log collection, to obtain and record triggered actions in real time, thereby obtaining corresponding behavioral data. Triggered actions can be varied, including clicks, browsing, searching, favorites, purchases, comments, etc., and this application does not limit the scope of these actions.

[0102] In addition, Action ID can uniquely identify a specific action, and ActionParameters can record the context information when the action occurs, that is, record the specific object associated with the operation and additional information, such as product ID, price, page dwell time, etc.

[0103] Additionally, timestamps can be used to record when an operation occurred. This application does not limit this use.

[0104] The behavioral data can then be processed using natural language to obtain basic text. This basic text can then be processed for intent recognition to infer the user's potential intent in performing the triggered action.

[0105] In this process, natural language templates can be used to fill in and supplement user data to generate basic text, which can correspond to behavioral identifiers. Alternatively, a text generation model can be used, taking behavioral data as input and processing it to obtain a more complete basic text, etc. This application does not limit the scope of the methods described.

[0106] Furthermore, during the intent recognition process, intent recognition can be based on rules or dictionaries, or an intent classification model can be used. This model uses basic text combined with user identifiers, behavior identifiers, behavior parameters, timestamps, etc., to predict the user's possible intent. This basic text can then be fused with the user's possible intent to generate a corresponding semantic description text, which not only includes the user's behavioral facts but also reflects the user's potential intent.

[0107] In addition, a behavior category can contain multiple related behavior identifiers. Classifying and storing behavior data and its semantic description text based on behavior categories can provide strong data support for quickly retrieving the semantic description text of target historical behaviors, thereby greatly saving data processing time.

[0108] Optionally, when performing natural language processing and intent recognition on behavioral data to generate corresponding semantic description text, one can also perform natural language processing on behavioral data to generate behavioral fact description text, and perform intent recognition on behavioral data to generate intent inference description text. Then, the behavioral fact description text and the intent inference description text are fused together to generate the corresponding semantic description text.

[0109] Among them, behavioral fact description text can be understood as text that objectively describes the user's actual operational behavior. It is only used to record the user's historical operation data and does not include subjective intent inference. It can truthfully and accurately represent the user's objective behavioral information. Intent inference description text can be understood as text obtained after analyzing user behavior data through intent recognition. It can be used to represent the potential needs and preferences behind the user's historical operations. Semantic description text can be text formed by combining behavioral fact description text and intent inference description text, thus including both objective behavior records and subjective intent representations.

[0110] Optionally, behavioral fact description text can be obtained by sorting fields and concatenating sentences in behavioral data, or by filling fields with preset natural language templates, or by other natural language processing methods such as text generation models. This application does not limit the specific methods used.

[0111] Alternatively, intent inference description text can be generated through preset matching rules. For example, a mapping relationship between behavior identifiers, behavior parameters, and user intent can be established, thereby determining the corresponding intent inference description text based on user data. Alternatively, an intent classification model can be used to process the input behavior data to obtain intent labels and confidence levels, and then the intent labels can be converted into natural language descriptions to obtain the intent inference description text. Alternatively, large language models can be used to process behavior data to generate intent inference description text, etc., and this application does not limit this approach.

[0112] In addition, when merging the behavioral fact description text and the intention inference description text to generate the corresponding semantic description text, the two can be directly concatenated in the order of behavioral fact first and intention inference last, and the resulting text is the semantic description text. Alternatively, the behavioral fact description text and the intention inference description text can be processed to smooth the sentences and optimize the sentences to generate the semantic description text, etc. This application does not limit this.

[0113] For example, if the current behavioral data includes: user ID USER001, behavior ID CLICK_FINANCE_01, annualized return of 3.2%, risk level R1, product type P001, page type financial recommendation page, and timestamp X-XX-XXX 08:00, then natural language processing based on this behavioral data can generate the following behavioral fact description text: User USER001 clicked on product type P001 on the financial recommendation page at 08:00 on X-XX-XXX, with an annualized return of 3.2% and a risk level of R1. Further intent recognition processing can be performed on this behavioral data. R1 typically corresponds to low risk, and an annualized return of 3.2% is considered a stable return in the market. Combined with the user's click on product type P001 on the financial recommendation page, it can be assumed that the user may be interested in low-risk, stable-return financial products. The generated intent inference description text could be: The user tends to choose low-risk, stable-return financial products. The semantic description text obtained by merging the above text can be: User USER001 clicked on a product of type P001 on the wealth management recommendation page at 08:00 on XX / XX / XXXX, with an annualized rate of return of 3.2% and a rating of R1. The user tends to choose low-risk, stable-return wealth management products.

[0114] It should be noted that the above examples are merely illustrative and should not be construed as limiting the behavioral fact description text, intent inference description text, semantic description text, etc., in the embodiments of this application. Optionally, when performing natural language processing and intent recognition processing on behavioral data to generate corresponding semantic description text, natural language processing can also be performed on the behavioral data to generate behavioral fact description text, and intent recognition processing can be performed on the behavioral data and behavioral fact description text to generate intent inference description text, and then the corresponding semantic description text can be generated, etc. This application does not limit this approach.

[0115] Therefore, in this embodiment of the application, by performing natural language processing and intent recognition processing on behavioral data, behavioral fact description text and intent inference description text can be generated. Then, by fusing the behavioral fact description text and the intent inference description text, a corresponding semantic description text can be generated. This semantic description text contains both the user's objective operational facts and potential preference intents, realizing a two-layer comprehensive representation of the user's historical behavior. This effectively improves the accuracy and reliability of the semantic description text, thereby providing an accurate and reliable data foundation for subsequent content recommendation based on the semantic description text.

[0116] The content recommendation method provided in this application can be applied to any scenario that requires content or product recommendations. For example, in news or video applications, this method can adjust the order of information streams based on users' real-time reading interests; or in e-commerce shopping guide scenarios, this method can dynamically optimize the priority of product display based on users' real-time browsing and search behavior; or in social platforms, it can personalize the sorting of users' friends' updates or advertising content, etc. This application does not limit this application.

[0117] The following is combined Figure 3 A brief explanation of the recommended methods for the content provided in this application is provided.

[0118] Step 302: In response to a trigger operation on any interface, obtain the behavior data corresponding to the trigger operation, wherein the behavior data includes user identifier, behavior identifier, behavior parameters and timestamp.

[0119] First, multi-dimensional user behavior data collection and standardization processing can be performed. For example, a standardized behavior data collection interface can be pre-built into the client software development kit (SDK). This interface can define the core data fields required for each behavior record, including behavior type, behavior ID, and behavior parameters.

[0120] The behavior category can be a broad category that identifies the behavior, such as product browsing (product_view), purchase click (purchase_click), content sharing (content_share), etc. This application does not limit this.

[0121] In addition, a behavior ID can uniquely identify a specific behavior, such as click_financial_product_A, etc., but this application does not limit it.

[0122] In addition, Action Parameters can be used to record contextual information when an action occurs, such as product ID, price, and page dwell time, but this application does not limit this.

[0123] Optionally, when a user triggers a predefined action (such as clicking, browsing, or adding to favorites) within the application, the business code can call the SDK's trackEvent method, passing in the corresponding action ID and action parameters.

[0124] Step 304: Perform natural language processing and intent recognition on the behavioral data to generate corresponding semantic description text.

[0125] The semantic description text can be configured with a corresponding natural language description for each behavior ID, used to explain the potential intent and user interest of the behavior to the large language model. For example, for the behavior of "clicking on a high-risk financial product," the corresponding semantic description text set could be: The user actively clicked on a financial product labeled "high-risk." This behavior indicates that the user may be interested in high-yield investment opportunities and has a certain tendency to explore risk tolerance, etc. This application does not impose any limitations on this.

[0126] Step 306: Store the behavioral data and corresponding semantic description text in the database according to their respective behavioral categories.

[0127] The collected behavioral data, automatically collected device information, and timestamps can be packaged together and accompanied by corresponding semantic description text to form a complete historical record. According to the behavioral category, the record can be transmitted to local storage or the server to build a continuously updated user behavior profile database.

[0128] Step 308: In response to the target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface. The original content list contains at least one candidate content.

[0129] Optionally, the SDK can provide a prediction interface called `getPersonalizedRanking`. Developers can call this interface and pass in parameters when recommendations are needed, such as when opening a new product list page. This allows them to obtain the target behavior identifier corresponding to the triggered action, as well as the original content list of the target interface.

[0130] The target behavior identifier, also known as the target behavior ID, can indicate the target that this ranking aims to optimize, such as promoting clicks on financial products, etc. This application does not limit this.

[0131] Step 310: Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, query the database for the semantic description text of the target historical behavior corresponding to the target behavior identifier.

[0132] Step 312: Filter semantic description text associated with user identifier from the semantic description text of the target's historical behavior.

[0133] Step 314: Extract content features from each candidate content to obtain the content features of each candidate content, and generate semantic description text of the candidate content based on the content features.

[0134] Step 316: Integrate the semantic description text of the target's historical behavior with the semantic description text of the candidate content to generate a prompt message.

[0135] Step 318: Input the prompt information into the large language model to obtain the user preference degree for each candidate content.

[0136] Among them, since the prompt information is rich in user profile and contextual information, the large language model can perform a deep understanding of the prompt information in terms of natural language and user intent, analyze the long-term preferences and real-time interests reflected by the user's historical behavior, and predict the user's preference probability for each candidate in the candidate list, i.e., the user's preference degree.

[0137] Step 320: Generate a list of target content based on the user preference for each candidate content.

[0138] Optionally, in this application embodiment, flexible deployment methods can be supported. It can utilize the local model on the device to ensure real-time performance and privacy, or call a more powerful model in the cloud through the HTTP API to obtain stronger inference capabilities. This application does not limit this.

[0139] It should be noted that the specific content and implementation of steps 310 to 320 can be referred to the description of the various embodiments of this application, and will not be repeated here.

[0140] In this embodiment, during the content recommendation process, the semantic description text of the target historical behavior determined by the user's real-time triggered behavior can be used to fully explore the user's behavioral facts and preference intentions. Then, combined with the deep semantic analysis capabilities of the large language model, user needs can be accurately matched to achieve personalized and differentiated content recommendations, thereby effectively improving the accuracy and adaptability of content recommendations.

[0141] The following is based on Figure 4 For example, this application provides a brief explanation of the recommended methods for the content provided.

[0142] Step 402: In response to a trigger operation on any interface, obtain the corresponding behavior data through the SDK collection interface. The behavior data includes user identifier, behavior identifier, behavior parameters and timestamp.

[0143] The SDK can include a pre-defined standardized behavior data collection interface. By using this SDK collection interface, the behavior data corresponding to the triggered operation can be obtained.

[0144] Step 404: Perform natural language processing and intent recognition on the behavioral data to generate corresponding semantic description text.

[0145] For ease of understanding, semantic description text may also be referred to as semantic behavior description, etc., and this application does not limit it in this way.

[0146] Step 406: Store the behavioral data and corresponding semantic description text in the database according to their respective behavioral categories.

[0147] In the process of storing behavioral data and corresponding semantic description text, a user's historical behavior profile was also established through a database. This profile allows for a relatively clear understanding of the user's historical behavioral facts and preferences.

[0148] Step 408: In response to the target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface. The original content list contains at least one candidate content.

[0149] The original content list corresponding to the target interface can also be called the product list to be sorted, etc., and this application does not limit it in this way.

[0150] Step 410: Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate a prompt message based on the semantic description text and the original content list.

[0151] Specifically, based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, a query can be performed in the user's historical behavior archive to determine the semantic description text of the corresponding target historical behavior. Then, based on the semantic description text and the original content list, a prompt message can be generated.

[0152] Step 412: In the case of local priority, input the prompt information into the local LLM on the mobile device to obtain the user preference of each candidate content.

[0153] Step 414: In the case of server-side invocation, the server-side LLM API is invoked, and the prompt information is input into the large language model to obtain the user preference for each candidate content.

[0154] In the process of reasoning about prompts using a large language model, it is possible to first determine whether to use a local large language model or call a server-side large language model, and then process the prompts in combination with the specific large language model.

[0155] Optionally, the local LLM on the mobile device can be used first, or the server-side LLM API interface can be called to invoke the server-side LLM, etc. This application does not limit this.

[0156] Step 416: Generate a list of target content based on the user preference for each candidate content.

[0157] The target content list is generated based on the user's preference for each candidate content, which can better meet the user's personalized needs, and therefore can also be called a personalized ranking result.

[0158] Step 418: Update and render the interface according to the target content list.

[0159] During the dynamic updating and rendering of the user interface, the application front-end receives a list of target content that has been intelligently sorted from the SDK prediction interface and will no longer use the original default sorting. Interface components such as list views and grid views will be dynamically rendered based on the new sorting results, placing items that the user is most interested in, such as the most likely financial products to click, at the top of the screen. This significantly improves the efficiency of content distribution and the relevance of the user experience, achieving targeted and personalized interface display.

[0160] Therefore, this solution, by introducing semantic description text of target behavior identifiers, can transform any user behavior into contextual information that the model can understand. This allows the system to automatically discover and comprehensively utilize massive amounts of unstructured behavioral features for decision-making, eliminating the need to redevelop rules for each new dimension and greatly expanding the system's application boundaries and scenario adaptability. Simultaneously, through deep semantic analysis and reasoning of prompts using a large language model, it can accurately discern the user's true intent and potential preferences, thereby achieving intelligent ranking of recommended content, significantly enhancing the exposure of products of user interest, and improving click-through rates and conversion rates. Furthermore, since the recommendation process of this solution is triggered in real time, the latest user behavior can be promptly incorporated into the analysis and updated ranking results can be generated, ensuring timeliness and guaranteeing that users always receive content that best matches their current interests and context, improving the smoothness and immediacy of the experience. In addition, this solution also supports reasoning using localized models on mobile devices, so that the analysis and ranking of all user behavior data can be completed directly on the terminal, without uploading sensitive raw data to the cloud. This effectively protects user privacy while providing accurate and personalized services and meeting increasingly stringent data compliance requirements.

[0161] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0162] Based on the same inventive concept, this application also provides a content recommendation apparatus for implementing the content recommendation method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations of the content recommendation apparatus embodiment provided below can be found in the limitations of the content recommendation method described above, and will not be repeated here.

[0163] In one exemplary embodiment, such as Figure 5 As shown, a content recommendation device 500 is provided, including: an acquisition module 510, a determination module 520, an input module 530, and a generation module 540, wherein:

[0164] The acquisition module 510 is used to respond to a target trigger operation in the target interface, acquire the target behavior identifier corresponding to the target trigger operation, and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content.

[0165] The determination module 520 is used to determine the semantic description text of the corresponding target historical behavior based on the target behavior identifier and the user identifier of the target user who performs the target trigger operation, and to generate prompt information based on the semantic description text and the original content list.

[0166] The input module 530 is used to input the prompt information into the large language model to obtain the user preference degree of each candidate content.

[0167] The generation module 540 is used to generate a list of target content based on the user preference of each candidate content.

[0168] In one embodiment, the determining module 520 includes:

[0169] The query unit is used to query the database for semantic description text of the target historical behavior corresponding to the target behavior identifier;

[0170] The filtering unit is used to filter semantic description text associated with the user identifier from the semantic description text of the target historical behavior.

[0171] In one embodiment, the query unit is specifically used for:

[0172] Based on the mapping relationship between behavior identifiers and behavior categories, the target behavior category to which the target behavior identifier belongs is determined;

[0173] Query the database for the semantic description text set corresponding to the target behavior category;

[0174] Select semantic description texts that correspond to the target behavior identifier from the set of semantic description texts.

[0175] In one embodiment, the determining module 520 includes:

[0176] The acquisition unit is used to acquire behavioral data corresponding to any trigger operation in response to any interface, wherein the behavioral data includes user identifier, behavior identifier, behavior parameters and timestamp;

[0177] The generation unit is used to perform natural language processing and intent recognition processing on the behavioral data to generate corresponding semantic description text.

[0178] The storage unit is used to store the behavioral data and the corresponding semantic description text in the database according to the behavioral category to which they belong.

[0179] In one embodiment, the generating unit is specifically used for:

[0180] The behavioral data is subjected to natural language processing to generate a text describing the behavioral facts;

[0181] The behavioral data is processed for intent recognition to generate an intent inference description text;

[0182] The behavioral fact description text and the intent inference description text are fused together to generate the corresponding semantic description text.

[0183] In one embodiment, the determining module 520 is specifically used for:

[0184] Each candidate content is processed by content feature extraction to obtain the content features of each candidate content, and a semantic description text of the candidate content is generated based on the content features;

[0185] The semantic description text of the target's historical behavior is integrated with the semantic description text of the candidate content to generate a prompt message.

[0186] In one embodiment, the generation module 540 is specifically used for:

[0187] In response to the existence of at least two candidate contents with the same user preference, the candidate contents are arranged according to a preset priority rule;

[0188] After reordering each candidate content according to user preference and the preset priority rules, a target content list is generated.

[0189] The modules in the recommended device described above can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0190] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores behavioral data and semantic descriptive text data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a content recommendation method.

[0191] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0192] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0193] In response to a target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content;

[0194] Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate prompt information based on the semantic description text and the original content list;

[0195] The prompt information is input into a large language model to obtain the user preference level for each candidate content;

[0196] A list of target content is generated based on the user preference for each candidate content.

[0197] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0198] Query the database for semantic description text of the target historical behavior corresponding to the target behavior identifier; filter the semantic description text associated with the user identifier from the semantic description text of the target historical behavior.

[0199] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0200] Based on the mapping relationship between behavior identifiers and behavior categories, determine the target behavior category to which the target behavior identifier belongs; query the database for the semantic description text set corresponding to the target behavior category; and filter the semantic description texts corresponding to the target behavior identifier from the semantic description text set.

[0201] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0202] In response to a trigger operation on any interface, the system acquires the behavioral data corresponding to the trigger operation, wherein the behavioral data includes a user identifier, a behavior identifier, behavior parameters, and a timestamp; the system performs natural language processing and intent recognition processing on the behavioral data to generate corresponding semantic description text; and the system stores the behavioral data and the corresponding semantic description text in the database according to their respective behavior categories.

[0203] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0204] The behavioral data is subjected to natural language processing to generate a behavioral fact description text; the behavioral data is subjected to intent recognition processing to generate an intent inference description text; the behavioral fact description text and the intent inference description text are fused to generate a corresponding semantic description text.

[0205] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0206] Each candidate content is processed by content feature extraction to obtain the content features of each candidate content, and a semantic description text of the candidate content is generated based on the content features; the semantic description text of the target historical behavior is integrated with the semantic description text of the candidate content to generate a prompt message.

[0207] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0208] In response to the existence of at least two candidate contents with the same user preference, the candidate contents are arranged according to a preset priority rule; after reordering each candidate content according to user preference and the preset priority rule, a target content list is generated.

[0209] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0210] In response to a target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content;

[0211] Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate prompt information based on the semantic description text and the original content list;

[0212] The prompt information is input into a large language model to obtain the user preference level for each candidate content;

[0213] A list of target content is generated based on the user preference for each candidate content.

[0214] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0215] Query the database for semantic description text of the target historical behavior corresponding to the target behavior identifier; filter the semantic description text associated with the user identifier from the semantic description text of the target historical behavior.

[0216] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0217] Based on the mapping relationship between behavior identifiers and behavior categories, determine the target behavior category to which the target behavior identifier belongs; query the database for the semantic description text set corresponding to the target behavior category; and filter the semantic description texts corresponding to the target behavior identifier from the semantic description text set.

[0218] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0219] In response to a trigger operation on any interface, the system acquires the behavioral data corresponding to the trigger operation, wherein the behavioral data includes a user identifier, a behavior identifier, behavior parameters, and a timestamp; the system performs natural language processing and intent recognition processing on the behavioral data to generate corresponding semantic description text; and the system stores the behavioral data and the corresponding semantic description text in the database according to their respective behavior categories.

[0220] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0221] The behavioral data is subjected to natural language processing to generate a behavioral fact description text; the behavioral data is subjected to intent recognition processing to generate an intent inference description text; the behavioral fact description text and the intent inference description text are fused to generate a corresponding semantic description text.

[0222] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0223] Each candidate content is processed by content feature extraction to obtain the content features of each candidate content, and a semantic description text of the candidate content is generated based on the content features; the semantic description text of the target historical behavior is integrated with the semantic description text of the candidate content to generate a prompt message.

[0224] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0225] In response to the existence of at least two candidate contents with the same user preference, the candidate contents are arranged according to a preset priority rule; after reordering each candidate content according to user preference and the preset priority rule, a target content list is generated.

[0226] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0227] In response to a target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content;

[0228] Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate prompt information based on the semantic description text and the original content list;

[0229] The prompt information is input into a large language model to obtain the user preference level for each candidate content;

[0230] A list of target content is generated based on the user preference for each candidate content.

[0231] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0232] Query the database for semantic description text of the target historical behavior corresponding to the target behavior identifier; filter the semantic description text associated with the user identifier from the semantic description text of the target historical behavior.

[0233] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0234] Based on the mapping relationship between behavior identifiers and behavior categories, determine the target behavior category to which the target behavior identifier belongs; query the database for the semantic description text set corresponding to the target behavior category; and filter the semantic description texts corresponding to the target behavior identifier from the semantic description text set.

[0235] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0236] In response to a trigger operation on any interface, the system acquires the behavioral data corresponding to the trigger operation, wherein the behavioral data includes a user identifier, a behavior identifier, behavior parameters, and a timestamp; the system performs natural language processing and intent recognition processing on the behavioral data to generate corresponding semantic description text; and the system stores the behavioral data and the corresponding semantic description text in the database according to their respective behavior categories.

[0237] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0238] The behavioral data is subjected to natural language processing to generate a behavioral fact description text; the behavioral data is subjected to intent recognition processing to generate an intent inference description text; the behavioral fact description text and the intent inference description text are fused to generate a corresponding semantic description text.

[0239] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0240] Each candidate content is processed by content feature extraction to obtain the content features of each candidate content, and a semantic description text of the candidate content is generated based on the content features; the semantic description text of the target historical behavior is integrated with the semantic description text of the candidate content to generate a prompt message.

[0241] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0242] In response to the existence of at least two candidate contents with the same user preference, the candidate contents are arranged according to a preset priority rule; after reordering each candidate content according to user preference and the preset priority rule, a target content list is generated.

[0243] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0244] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0245] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0246] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A content recommendation method, characterized in that, The method includes: In response to a target trigger operation in the target interface, obtain the target behavior identifier corresponding to the target trigger operation and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content; Based on the target behavior identifier and the user identifier of the target user who performed the target trigger operation, determine the semantic description text of the corresponding target historical behavior, and generate prompt information based on the semantic description text and the original content list; The prompt information is input into a large language model to obtain the user preference level for each candidate content; A list of target content is generated based on the user preference for each candidate content.

2. The method according to claim 1, characterized in that, The step of determining the semantic description text of the corresponding target historical behavior based on the target behavior identifier and the user identifier of the target user executing the target triggering operation includes: Query the database for the semantic description text of the target's historical behavior corresponding to the target behavior identifier; Filter semantic description text associated with the user identifier from the semantic description text of the target's historical behavior.

3. The method according to claim 2, characterized in that, The step of querying the database for the semantic description text of the target historical behavior corresponding to the target behavior identifier includes: Based on the mapping relationship between behavior identifiers and behavior categories, the target behavior category to which the target behavior identifier belongs is determined; Query the database for the semantic description text set corresponding to the target behavior category; Select semantic description texts that correspond to the target behavior identifier from the set of semantic description texts.

4. The method according to claim 1, characterized in that, Before determining the semantic description text of the corresponding target historical behavior based on the target behavior identifier and the user identifier of the target user performing the target triggering operation, the method further includes: In response to a trigger operation on any interface, obtain the behavior data corresponding to the trigger operation, wherein the behavior data includes user identifier, behavior identifier, behavior parameters and timestamp; The behavioral data is processed using natural language processing and intent recognition to generate corresponding semantic description text. The behavioral data and corresponding semantic description text are stored in the database according to their respective behavioral categories.

5. The method according to claim 4, characterized in that, The step of performing natural language processing and intent recognition on the behavioral data to generate corresponding semantic description text includes: The behavioral data is subjected to natural language processing to generate a text describing the behavioral facts; The behavioral data is processed for intent recognition to generate an intent inference description text; The behavioral fact description text and the intent inference description text are fused together to generate the corresponding semantic description text.

6. The method according to claim 1, characterized in that, The step of generating prompt information based on the semantic description text and the original content list includes: Each candidate content is processed by content feature extraction to obtain the content features of each candidate content, and a semantic description text of the candidate content is generated based on the content features; The semantic description text of the target's historical behavior is integrated with the semantic description text of the candidate content to generate a prompt message.

7. The method according to claim 1, characterized in that, The step of generating a target content list based on the user preference for each candidate content includes: In response to the existence of at least two candidate contents with the same user preference, the candidate contents are arranged according to a preset priority rule; After reordering each candidate content according to user preference and the preset priority rules, a target content list is generated.

8. A content recommendation device, characterized in that, The device includes: The acquisition module is used to respond to a target trigger operation in the target interface, acquire the target behavior identifier corresponding to the target trigger operation, and the original content list corresponding to the target interface, wherein the original content list contains at least one candidate content; The determination module is used to determine the semantic description text of the corresponding target historical behavior based on the target behavior identifier and the user identifier of the target user who performs the target trigger operation, and to generate prompt information based on the semantic description text and the original content list; The input module is used to input the prompt information into the large language model to obtain the user preference degree of each candidate content; The generation module is used to generate a list of target content based on the user preference for each candidate content.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images