Article recommendation method, article recommendation device and electronic equipment for interactive platform

Abstract

The embodiment of the invention discloses an article recommendation method, an article recommendation device and electronic equipment for an interactive platform. A specific embodiment of the method comprises the following steps: analyzing and processing content data of a target interaction scene on an interaction platform, and determining context information of the target interaction scene; according to attribute information of a target user and behavior data in the target interaction scene, portrait data of the target user is determined, and the target user is a user in the target interaction scene; generating an article recommendation list according to the context information, the portrait data and a pre-trained recommendation model; and in the target interaction scene, displaying an article indicated by the article recommendation list. The implementation mode is related to a short video platform technology, commodity recommendation of the live broadcast room can be realized based on context perception and audience behavior analysis, the correlation and accuracy of commodity recommendation can be improved, and thus the commodity delivery quantity is increased.

Description

Technical Field

[0001] The embodiments of the present disclosure relate to the technical field of short video platforms, and specifically to an item recommendation method, an item recommendation device, and an electronic device for an interactive platform. Background Art

[0002] With the rapid development of short video platforms, live shopping has gradually become one of the main ways of daily consumption. Different from the product recommendation scene of online e-commerce platforms, the live broadcast scene of short video platforms is a real-time interactive scene between anchors, products and viewers. The anchor needs to adjust the next recommended products in real time based on the current live broadcast atmosphere, the products to be displayed, and the audience's behavioral feedback to maximize the sales of the products.

[0003] However, the inventors found that in the live broadcasting scenes on various short video platforms, the anchors' explanations, demonstrations and recommendations of products are mainly through the live broadcasting operation team's manual selection of recommended products based on their own experience, or the sorting of products based on set rules. This method often fails to deeply understand the personalized needs and preferences of each user, affects the accuracy of product recommendations, and thus affects the actual sales of products.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0005] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0006] Some embodiments of the present disclosure propose an item recommendation method for an interactive platform, an item recommendation device for an interactive platform, an electronic device, a computer-readable medium, and a computer program product to solve one or more of the technical problems mentioned in the above background technology section.

[0007] In a first aspect, some embodiments of the present disclosure provide an item recommendation method for an interactive platform, comprising: analyzing and processing content data of a target interactive scene on the interactive platform to determine context information of the target interactive scene; determining portrait data of the target user based on attribute information of the target user and behavior data in the target interactive scene, wherein the target user is a user in the target interactive scene; generating an item recommendation list based on the context information, portrait data and a pre-trained recommendation model; and displaying the items indicated in the item recommendation list in the target interactive scene.

[0008] In some embodiments, content data of a target interactive scene on an interactive platform is analyzed and processed to determine context information of the target interactive scene, including: using a natural language processing model to analyze and process the content data of the target interactive scene to extract text feature data, wherein the content data includes at least one of video data, audio data, and text data; extracting key data from the text feature data, and determining context information of the target interactive scene based on the key data, wherein the key data includes at least one of the following: content theme, keywords, and emotional information.

[0009] In some embodiments, key data is extracted from text feature data, and context information of the target interactive scene is determined based on the key data, including: using a topic model to extract content topics from text feature data; using a keyword extraction model to extract keywords from text feature data; using a sentiment analysis model to perform sentiment prediction on text feature data to obtain sentiment information; performing a fusion analysis on the extracted content topics, keywords and sentiment information to determine the context information of the target interactive scene; and the method also includes: updating the context information of the target interactive scene in real time based on the collected real-time content data of the target interactive scene.

[0010] In some embodiments, the target user's portrait data is determined based on the target user's attribute information and the behavior data in the target interactive scenario, including: extracting key behavior features of the target user's behavior data in the target interactive scenario, wherein the key behavior features include at least one of the following: interaction frequency, interaction method, comment content, comment sentiment information, and dwell time distribution; using a natural language processing model to analyze the comment content in the key behavior features and extract key data; determining the target user's portrait data based on the target user's attribute information and the extracted key behavior features and key data, and storing the portrait data, wherein the attribute information includes at least one of the following: basic attribute information, historical behavior data on the interactive platform, and behavior data in a set application.

[0011] In some embodiments, the method further includes: updating the target user's portrait data in real time based on the collected real-time behavior data of the target user in the target interaction scenario, wherein the portrait data includes the user's preference data.

[0012] In some embodiments, an item recommendation list is generated based on context information, portrait data, and a pre-trained recommendation model, including: inputting the context information and portrait data into the pre-trained recommendation model to determine the target user's predicted score for the item; selecting a specified number of items from each item in order of predicted score from high to low to generate an item recommendation list.

[0013] In some embodiments, the method also includes: adjusting the recommendation strategy according to dynamically changing data to adjust the items or the order in the item recommendation list, wherein the dynamically changing data includes at least one of the following: real-time content data of the target interactive scene, real-time behavior data of the target user in the target interactive scene, feedback data of the target user on the items in the item recommendation list, item inventory data, and item activity information.

[0014] In some embodiments, the method also includes: collecting scene data in the target interactive scene, and user data of the target user in the target interactive scene; cleaning the scene data and user data, and using the cleaned scene data as content data of the target interactive scene, and using the cleaned user data as behavior data of the target user in the target interactive scene.

[0015] In a second aspect, some embodiments of the present disclosure provide an item recommendation device for an interactive platform, comprising: a context determination unit, configured to analyze and process content data of a target interactive scene on the interactive platform, and determine context information of the target interactive scene; a user data determination unit, configured to determine portrait data of the target user based on attribute information of the target user and behavior data in the target interactive scene, wherein the target user is a user in the target interactive scene; a recommended item generation unit, configured to generate an item recommendation list based on the context information, portrait data and a pre-trained recommendation model; and an item display unit, configured to display items indicated in the item recommendation list in the target interactive scene.

[0016] In some embodiments, the context determination unit is further configured to use a natural language processing model to analyze and process the content data of the target interactive scene, and extract text feature data, wherein the content data includes at least one of video data, audio data, and text data; extract key data from the text feature data, and determine the context information of the target interactive scene based on the key data, wherein the key data includes at least one of the following: content theme, keywords, and emotional information.

[0017] In some embodiments, the context determination unit is further configured to use a topic model to extract content topics from text feature data; use a keyword extraction model to extract keywords from text feature data; use a sentiment analysis model to perform sentiment prediction on text feature data to obtain sentiment information; perform a fusion analysis on the extracted content topics, keywords and sentiment information to determine the context information of the target interaction scene; and the item recommendation device also includes a context update unit, which is configured to update the context information of the target interaction scene in real time based on the collected real-time content data of the target interaction scene.

[0018] In some embodiments, the user data determination unit is further configured to extract key behavior features of the target user's behavior data in the target interaction scenario, wherein the key behavior features include at least one of the following: interaction frequency, interaction method, comment content, comment sentiment information, and dwell time distribution; use a natural language processing model to analyze the comment content in the key behavior features and extract key data; determine the target user's portrait data based on the target user's attribute information, and the extracted key behavior features and key data, and store the portrait data, wherein the attribute information includes at least one of the following: basic attribute information, historical behavior data on the interactive platform, and behavior data in the set application.

[0019] In some embodiments, the item recommendation device also includes a user data updating unit, which is configured to update the target user's portrait data in real time based on the collected real-time behavior data of the target user in the target interaction scenario, wherein the portrait data includes the user's preference data.

[0020] In some embodiments, the recommended item generation unit is further configured to input context information and portrait data into a pre-trained recommendation model to determine the target user's predicted score for the item; select a specified number of items from each item in order of predicted score from high to low to generate an item recommendation list.

[0021] In some embodiments, the item recommendation device also includes a recommendation strategy adjustment unit, which is configured to adjust the recommendation strategy according to dynamically changing data to adjust the items or the order in the item recommendation list, wherein the dynamically changing data includes at least one of the following: real-time content data of the target interactive scene, real-time behavior data of the target user in the target interactive scene, feedback data of the target user on the items in the item recommendation list, item inventory data, and item activity information.

[0022] In some embodiments, the item recommendation device also includes a data collection unit, which is configured to collect scene data in the target interactive scene and user data of the target user in the target interactive scene; clean the scene data and user data, and use the cleaned scene data as content data of the target interactive scene, and use the cleaned user data as behavior data of the target user in the target interactive scene.

[0023] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the item recommendation method for an interactive platform described in any implementation method in the first aspect above.

[0024] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for recommending items for an interactive platform as described in any one of the implementations in the first aspect above is implemented.

[0025] In a fifth aspect, some embodiments of the present disclosure provide a computer program product, including a computer program, which, when executed by a processor, implements the item recommendation method for an interactive platform described in any implementation manner in the first aspect above.

[0026] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: the item recommendation method for an interactive platform of some embodiments of the present disclosure can realize the real-time recommendation of items in the interactive scene, improve the accuracy of item recommendation, and thus help to increase the outbound volume of items. Specifically, unlike the commodity recommendation scene of the online e-commerce platform, the live broadcast scene of the short video platform is a real-time interactive scene of the anchor, the commodity and the audience. The audience may come to the live broadcast room with the demand for purchase, because they have a certain mental and trust basis for the anchors they often see. There are also viewers who come to the live broadcast room without the need for purchase. The purpose of visiting the live broadcast room is to kill time or to show support for the anchor. However, the demand for purchase may often arise when interacting with the anchor during the stay in the live broadcast room. The anchor needs to adjust the next recommended commodity in real time based on the current live broadcast atmosphere, the commodities to be displayed, and the behavioral feedback of the audience to maximize the sales of the commodities.

[0027] However, existing technologies often fail to deeply understand the personalized needs and preferences of each user, have poor real-time performance, and are unable to respond to changes in user behavior in real time. In addition, existing technologies may overly rely on historical user data and ignore real-time contextual information. Existing technologies rarely analyze the contextual information of live broadcast content, such as the host's introduction and the audience's interaction. These will affect the accuracy of item recommendations and thus affect the sales volume of items.

[0028] Based on this, the item recommendation method for an interactive platform of some embodiments of the present disclosure can not only analyze the content data of the interactive scene, but also analyze the behavior data of the user in the interactive scene. In this way, the context information of the interactive scene and the user's portrait data can be obtained in real time. Thereby, it is possible to respond to the live broadcast content and the audience interaction in real time, and deeply understand the personalized needs of users. In this way, the item recommendation list is determined based on the context information and portrait data, which can improve the accuracy of item recommendations, especially for real-time interactive live broadcast sales. It not only helps to increase the outbound volume of items, but also reduces the dependence on manual operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0030] Figure 1 is a flow chart of some embodiments of the item recommendation method for an interactive platform disclosed in the present invention;

[0031] Figure 2 Schematic diagrams of some application scenarios of the item recommendation method disclosed in the present invention;

[0032] Figure 3A yes Figure 2 Flowcharts of some embodiments of the context-aware module;

[0033] Figure 3B yes Figure 2 Flowcharts of some embodiments of the audience behavior analysis module;

[0034] Figure 3C yes Figure 2 Flowcharts of some embodiments of a product recommendation engine;

[0035] Figure 4 It is a schematic diagram of the structure of some embodiments of the item recommendation device for an interactive platform disclosed in the present invention;

[0036] Figure 5 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0037] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0038] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0039] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0040] In addition, the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0041] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0042] With regard to the collection, storage, and use of user personal information (such as user basic attribute information, user portraits, user historical behavior data, etc.) involved in this disclosure, before performing the corresponding operations, the relevant organizations or individuals shall fulfill their obligations, including conducting personal information security impact assessments, fulfilling the obligation to inform the personal information subject, and obtaining the authorization and consent of the personal information subject in advance.

[0043] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0044] Figure 1 The process 100 of some embodiments of the item recommendation method for an interactive platform according to the present disclosure is shown. The method may include the following steps:

[0045] Step 101 : analyzing and processing content data of a target interactive scene on an interactive platform to determine context information of the target interactive scene.

[0046] In some embodiments, the execution subject (such as a platform server) of the item recommendation method for an interactive platform in some embodiments of the present disclosure can be connected to other electronic devices (such as user terminals, databases, etc.) through wired or wireless connections. Here, users can access the interactive platform through user terminals. The interactive platform here can be a platform with interactive functions, especially a platform that can interact with users in real time, such as a short video platform, a live video platform, a live broadcast module in a shopping application or a social application, etc. Interactive scenes are usually provided and displayed on the interactive platform. In this way, users can use user terminals to create and / or view interactive scenes on the interactive platform, and can enter the interactive scene for interactive operations. The interactive scene here is usually a specific interactive scene on the interactive platform, such as a live broadcast room of an account or a store.

[0047] Here, when the execution subject receives an instruction representing the execution of the item recommendation process, it can execute the item recommendation method of the present disclosure to generate an item recommendation list. It is understandable that the instruction representing the execution of the item recommendation process can include multiple situations. For example, an instruction to start an interactive scene sent by a user terminal is received; another example is an instruction to start an item recommendation service is received from a user terminal; another example is detecting (for the first time) that a user has entered or interacted in an already opened interactive scene, etc.

[0048] In some embodiments, after determining to execute the item recommendation process, the execution subject may first analyze and process the content data of the target interactive scene on the interactive platform, thereby determining the context information of the target interactive scene. The target interactive scene here is usually the interactive scene indicated by the above instructions. It can be any interactive scene on the interactive platform; it can also be an interactive scene that starts the item recommendation method service of the present disclosure; it can also be a specified specific interactive scene, such as a test interactive scene, etc.

[0049] It is understandable that the content data of the target interactive scene can be any data indicating the content of the interactive scene, such as video data, audio data, etc. of the scene. Here, the execution subject can extract key information, such as keywords, from these content data to obtain the context information of the target interactive scene. For example, the content data of different modalities can be input into the corresponding models respectively, or can be input into a multimodal large model to extract the respective key information. The extracted key information can then be integrated. The integrated key information can then be used as the context information of the target interactive scene.

[0050] Optionally, for Figure 2 In the application scenario of the live broadcast room product recommendation shown, in order to better perceive the context information of the live broadcast room, the execution subject can also use a natural language processing model (NLP, Natural Language Processing) to analyze and process the content data of the target interactive scene and extract text feature data. Among them, the content data may include at least one of video data, audio data, and text data. Extract key data from the text feature data, and determine the context information of the target interactive scene based on the key data. Among them, the key data may include at least one of the following: content theme, keywords, and emotional information.

[0051] Specifically, topic models, such as LDA (Latent Dirichlet Allocation, a document topic generation model), can be used to extract content topics from text feature data. Keyword extraction models, such as TF-IDF (term frequency-inverse document frequency), can be used to extract keywords from text feature data. Sentiment analysis models can be used to predict sentiment on text feature data to obtain sentiment information. Finally, the extracted content topics, keywords, and sentiment information can be fused and analyzed to determine the context information of the target interactive scene. For more information, see Figure 3A The processing flow shown.

[0052] Step 102, determining the target user's portrait data based on the target user's attribute information and behavior data in the target interactive scenario.

[0053] In some embodiments, the execution subject can determine the portrait data of the target user based on the attribute information of the target user and the behavior data of the target user in the target interactive scene. Among them, the target user is usually a user in the target interactive scene, such as an audience entering the live broadcast room. The attribute information here can be the basic attribute information of the user, such as age, gender, address location, occupation, etc. The behavior data usually characterizes various behavioral operation data of the user in the interactive scene, such as likes, collections, shares, comments, viewing time, etc. As an example, the execution subject can extract preset data from the attribute information and behavior data of the target user. Thereby, the portrait data of the target user is constructed using these preset data. Among them, the preset data can be set according to the actual situation. For example, for the interactive scene of live sales, the preset data may include comments, viewing time, adding to shopping cart, purchase, etc. For the interactive scene of only live video broadcast, the preset data may include likes, comments, sharing, collection, etc.

[0054] In some embodiments, for Figure 2 In the audience behavior analysis module in the application scenario of product recommendation in the live broadcast room shown, the execution entity can extract key behavioral features from the behavioral data of the target user in the target interactive scenario. Among them, the key behavioral features may include at least one of the following: interaction frequency, interaction method, comment content, comment sentiment information, residence time distribution, etc. As an example, machine learning algorithms such as matrix factorization, GBDT (Gradient Boosted Decision Trees), MAB (Multi-Armed Bandit), autoencoders, etc. can be used to classify, identify and predict audience behavior.

[0055] Furthermore, a natural language processing model (NLP) can be used to analyze the comment content in the key behavior features and extract key data. The key data here can also include at least one of themes, keywords and sentiment information. Finally, the target user's portrait data can be determined based on the target user's attribute information, and the extracted key behavior features and key data. And the portrait data can be stored. It is understandable that in order to improve the accuracy of user portrait data analysis, the attribute information here may include at least one of the following: basic attribute information, historical behavior data on the interactive platform (such as information about other live broadcast rooms that have been watched recently, historical purchase records, etc.), and behavior data in set applications (such as social software or other e-commerce platforms, etc.). For details, please refer to Figure 3B The processing flow shown.

[0056] It is understandable that this method can also integrate social network data. By analyzing user behaviors on social platforms (such as sharing, liking, commenting, etc.), it is possible to more comprehensively understand user needs and preferences, and further improve the accuracy and relevance of recommendations.

[0057] Step 103, generating an item recommendation list based on the context information, the portrait data and the pre-trained recommendation model.

[0058] In some embodiments, the execution subject can generate a list of recommended items based on the context information obtained in step 101 and the portrait data obtained in step 102, using a pre-trained recommendation model. The recommendation model here can also be set according to actual needs, such as a recommendation algorithm based on a graph neural network (GNN) or a recommendation model based on a Transformer (a deep learning model based on a self-attention mechanism). Here, the context information, portrait data, and item information in the target interactive scene can be input into the recommendation model, and the model can sort the input item information to generate a list of recommended items.

[0059] As an example, after inputting the context information and portrait data into the recommendation model, the model can determine the predicted score of each item by the target user. Then, a specified number of items can be selected from each item in the order of predicted score from high to low to generate an item recommendation list. It is understandable that the same target interaction scene often contains multiple target users. In this case, the execution entity can first determine the personalized item recommendation list for each target user according to the above method. Then, the items in the item recommendation list of each target user can be statistically analyzed to generate the final item recommendation list for the target interaction scene.

[0060] For example, you can select common items in each item recommendation list and calculate the total score or average score of each common item. Then, sort the common items according to the calculation results to generate a final item recommendation list. For another example, in order to increase the number and richness of recommended items, you can calculate the total score or average score of each item in each item recommendation list and sort the items accordingly. Then, you can select a specified number of items to generate a final item recommendation list.

[0061] It should be noted that the above recommendation model can be obtained by training with training data, and the training method can be set by oneself. For example, the training data can include sample data such as sample user portrait data, sample item information, sample context information, and sample labels of the actual scores of sample users on sample items. In this case, the recommendation model can be initialized, and the model can be used to perform embedded representation learning on the sample data to determine the predicted score of the sample user on the sample item. Then, the loss function value between the predicted score and the corresponding actual score is calculated. Then, the parameters of the model are updated using the gradient descent algorithm to minimize the loss function value.

[0062] Step 104: display the items indicated in the item recommendation list in the target interactive scene.

[0063] In some embodiments, when an item recommendation list is generated, the items indicated in the item recommendation list can be displayed in the target interactive scene. The display method here is also not limited. For example, it can include but is not limited to pop-up window recommendation display in the live broadcast room, sidebar recommendation display, anchor oral recommendation, etc.

[0064] It can be seen from the above description that the item recommendation method for an interactive platform of some embodiments of the present disclosure can not only analyze the content data of the interactive scene, but also analyze the behavior data of the user in the interactive scene. In this way, the context information of the interactive scene and the user's portrait data can be obtained in real time. Thereby, it is possible to respond to the live broadcast content and the audience interaction in real time, and deeply understand the personalized needs of users. In this way, the item recommendation list is determined based on the context information and portrait data, which can improve the accuracy of item recommendations, especially for real-time interactive live broadcast sales. It not only helps to increase the outbound volume of items, but also reduces the dependence on manual operations.

[0065] It is understandable that before executing the item recommendation method disclosed in the present invention, it is usually necessary to collect corresponding data, such as Figures 2 to 3BThe data acquisition module shown in . That is, first, the scene data in the target interactive scene and the user data of the target user in the target interactive scene can be collected. The scene data here can be data that characterizes the content of the scene, and can include not only the above-mentioned content data, but also the scene identifier (such as the live broadcast room ID), information about items displayed in the scene, etc. Next, the scene data and user data can be cleaned, so that the cleaned scene data can be used as the content data of the target interactive scene, and the cleaned user data can be used as the behavior data of the target user in the target interactive scene.

[0066] In addition, in order to improve the accuracy of item recommendations and meet the real-time requirements of interactive scenarios, the item recommendation method disclosed in the present invention can also dynamically adjust the corresponding data in real time based on dynamically changing data. Among them, the dynamically changing data may include at least one of the following: real-time content data of the target interactive scenario, real-time behavior data of the target user in the target interactive scenario, feedback data of the target user on the items in the item recommendation list, item inventory data, item activity information, etc. For example, the context information of the target interactive scenario can be updated in real time based on the collected real-time content data of the target interactive scenario. For another example, the portrait data of the target user, such as the user's preference data, can be updated in real time based on the collected real-time behavior data of the target user in the target interactive scenario. In addition, the recommendation strategy can also be adjusted based on the dynamically changing data to adjust the items or sorting in the item recommendation list.

[0067] In other words, the method disclosed in the present invention aims to solve the deficiencies of existing live streaming product recommendation systems in terms of personalization, real-time, context awareness, dynamic adjustment, scalability, transparency, data utilization, long-tail product recommendation, reliance on manual operations, and social data integration. By developing a live streaming product recommendation method based on context-aware algorithms and audience behavior analysis technology. The present invention is committed to providing an intelligent recommendation system that can respond to live streaming content and audience interaction in real time, deeply understand user personalized needs, dynamically adjust recommendation strategies, adapt to large-scale users and products, provide explainable recommendation logic, comprehensively utilize real-time and historical data, optimize long-tail product exposure, reduce reliance on manual operations, and effectively integrate social network data. This will greatly improve the user experience, enhance the relevance and accuracy of product recommendations, and thus improve the conversion rate and commercial value of live streaming products.

[0068] Figure 2The process shown briefly summarizes the core steps of the present invention. It starts with data collection, goes through preprocessing, context awareness, audience behavior analysis, and generates a personalized recommendation list. Subsequently, the system collects real-time feedback and real-time inventory management, performs dynamic policy adjustments, ensures user privacy is protected, and finally displays the recommendation results. The whole process is cyclical, and the system will continuously update the recommendation results as new live content and audience behavior are generated.

[0069] Here, first, the data collection module is mainly responsible for real-time collection of multi-dimensional data during the live broadcast process, including but not limited to live broadcast content, audience comments, likes, purchase records, viewing time, audience geographic location, device information, etc. In addition, this module can also integrate historical data to provide a rich data source for subsequent context perception and audience behavior analysis. This is used for subsequent data processing and recommendation algorithms. The following is a detailed description of the data collection module, including sample collection, relevant parameter variable definitions, and the model description involved:

[0070] 1) Sample collection

[0071] Live content data: Collect multimedia content such as video streams, audio streams, and the host’s real-time speech in the live broadcast room. These data will be used for subsequent context perception and keyword extraction.

[0072] Audience behavior data: records all behaviors of viewers in the live broadcast room, including but not limited to likes, comments, purchases, shares, viewing time, etc. These data are the key to building a user interest model.

[0073] User portrait data: collect basic information about users, such as age, gender, geographic location, occupation, etc., as well as historical behavior data of users on the platform.

[0074] 2) Definition of relevant parameter variables

[0075] User ID: A number that uniquely identifies a user and is used to track and analyze user behavior.

[0076] Live Room ID: A number that uniquely identifies a live room, making it easier to analyze data from a specific live room.

[0077] Action Type: defines the specific behavior of the audience in the live broadcast room, such as likes, comments, purchases, etc.

[0078] Action Timestamp: records the exact time when the viewer's behavior occurs, and is used to analyze the timing characteristics of user behavior.

[0079] Action Content: If it is a comment or share action, record the specific content information to facilitate subsequent text analysis and keyword extraction.

[0080] Product ID: The unique identifier of the product displayed or purchased in the live broadcast room.

[0081] 3) Data Model

[0082] In the data acquisition module, although the main focus is on data collection and organization, some simple models can also be combined to perform preliminary data screening and preprocessing.

[0083] Data cleaning model: used to remove duplicate, invalid or abnormal data records to ensure the accuracy and validity of the data.

[0084] User behavior recognition model: Based on user portraits and historical behavior data, the user's behavior in the current live broadcast room is preliminarily identified and classified, providing a basis for subsequent user interest modeling.

[0085] It should be noted that the design and implementation of the data acquisition module must comply with relevant laws and regulations on user privacy protection and data security to ensure the legality and compliance of user data. At the same time, in order to improve the real-time and response speed of the system, the data acquisition module needs to adopt efficient data transmission and processing technology to ensure the timeliness and accuracy of the data.

[0086] Second, the context-aware module mainly extracts context information by analyzing the real-time content of the live broadcast room, providing an important basis for subsequent personalized recommendations. Figure 3A As shown, it mainly includes the following core algorithm models:

[0087] Natural Language Processing (NLP) model: used to analyze and understand the text content of the live broadcast room. The NLP model can include sub-modules such as lexical analysis, syntactic analysis, and semantic understanding, so as to extract key information and topics from the text.

[0088] Topic model: such as LDA (Latent Dirichlet Allocation, a document topic generation model), or more advanced topic models based on deep learning, are used to discover potential themes or topics from the text content of the live broadcast room. These themes or topics constitute the context information of the live broadcast room.

[0089] Keyword extraction algorithms: For example, the TF-IDF (term frequency-inverse document frequency) algorithm is used to extract keywords from text. These keywords reflect the focus of the live broadcast content and are also an important part of contextual information.

[0090] Sentiment analysis model: used to analyze the sentiment in the live broadcast room, including positive, negative or neutral. Sentiment information can be used as part of the context to influence the recommendation system's judgment of user interests.

[0091] The processing flow of the context-aware module is as follows:

[0092] (1) Data input: The context perception module receives real-time data from the live broadcast room from the data acquisition module, including multimedia content such as text, audio, and video.

[0093] (2) Preprocessing: Clean and standardize the input data, such as removing noise, stop words and special symbols, and performing operations such as stemming or word form restoration.

[0094] (3) Feature extraction:

[0095] The NLP model is used to perform operations such as word segmentation, part-of-speech tagging and syntactic analysis on the text content to extract the basic features of the text.

[0096] Apply topic models (such as LDA) to discover potential themes or topics from text.

[0097] Use keyword extraction algorithms (such as TF-IDF) to extract keywords from the text.

[0098] The sentiment tendency of the text is determined through the sentiment analysis model.

[0099] (4) Context information construction: The extracted topics, keywords, and sentiment tendencies are integrated to construct a context information model for the live broadcast room. This model can fully reflect the real-time content and atmosphere of the live broadcast room.

[0100] (5) Context information output: The constructed context information model is output to the subsequent recommendation algorithm module as an important reference for generating personalized recommendations.

[0101] (6) Dynamic update: As the live broadcast content continues to change, the context awareness module needs to update the context information model in real time to ensure that it is always consistent with the current state of the live broadcast room.

[0102] Understandably, Figure 3A The processing flow shown usually requires the collaboration of various models. For example, data preprocessing and feature extraction: First, the natural language processing (NLP) model is used to preprocess the text data of the live broadcast room, including word segmentation, stop word removal, part-of-speech tagging, etc., for subsequent feature extraction and analysis. Then, keywords are extracted from the preprocessed text through a keyword extraction model (such as TF-IDF). These keywords can reflect the focus of the content of the live broadcast room.

[0103] Another example is topic modeling and sentiment analysis: topic models (such as LDA) are applied to the extracted text features to discover potential themes or topics in the live broadcast room. These themes constitute one of the core contextual information of the live broadcast room. At the same time, the sentiment analysis model judges the sentiment tendency of the live broadcast room text, identifies positive, negative or neutral sentiment, and provides the context of user emotional response for the recommendation system.

[0104] Contextual information fusion: The extracted keywords, topics, and sentiment tendencies are integrated to form a comprehensive live broadcast context information model. This model not only includes the keywords and topics of the live broadcast content, but also reflects the emotional response of the audience.

[0105] Dynamic updates and real-time feedback: As the live broadcast progresses, the context-aware module needs to continuously receive new data input and update the context information model in real time. This ensures that the model is always synchronized with the current state of the live broadcast room. In addition, the module can also fine-tune the model in the context-aware module based on real-time feedback data (such as user interaction, viewing time, etc.) to improve the accuracy and relevance of recommendations.

[0106] Collaboration with the recommendation engine: Finally, the constructed context information is passed to the recommendation algorithm module. When generating personalized recommendations, the recommendation algorithm will comprehensively consider the user's personal interests, historical behaviors, and current live broadcast room context information. In this way, the recommendation system can more accurately grasp the user's needs and preferences, thereby providing recommended content that better meets the user's expectations.

[0107] Third, the audience behavior analysis module mainly uses big data analysis technology to deeply explore the audience's behavior patterns in the live broadcast room. Figure 3B As shown in the figure, this module will monitor the audience's interaction frequency, comment content, dwell time and other multivariate data to build an accurate user portrait, and update the user's interests and preferences in real time through machine learning and deep learning algorithms. These user portrait data will serve as the key input of the recommendation engine, significantly improving the personalization of recommendations.

[0108] Data collection and processing: Real-time collection of audience behavior data in the live broadcast room, including viewing time, likes, comments, shares, purchases, etc. Clean and pre-process the collected data to remove invalid and abnormal data to ensure data quality.

[0109] Feature Engineering: Extract key features of audience behavior, such as interaction frequency, comment sentiment, dwell time distribution, etc. Use NLP technology to perform text analysis on comment content and extract topics, keywords, and sentiment tags.

[0110] User portrait construction: Based on the extracted features, construct the audience's user portrait, including basic information, interest preferences, consumption habits, etc. Use big data storage technology (such as HBase, Cassandra) to store user portrait data for subsequent analysis.

[0111] Application of machine learning and deep learning models: Use machine learning algorithms (such as matrix decomposition, GBDT, MAB, autoencoders, etc.) to classify and predict audience behavior. Introduce deep learning models (such as RNN, LSTM, GNN, etc.) to process time series data and capture the dynamic changes of audience interests.

[0112] Real-time update and optimization: Dynamically update user profiles and interest preferences based on the real-time behavior of viewers in the live broadcast room. Utilize online learning technology to continuously optimize the parameters of machine learning and deep learning models.

[0113] For example, after generating the user's portrait data, the user's interest vector can be initialized such as U = [0, 0, ..., 0]. The dimension is set according to the specific business. For each new behavior data of the user, the machine learning model can be used to parse the behavior data and extract key features (such as behavior type, related product category, etc.). After that, the deep learning model can be used to predict the user's interest changes based on the extracted key features, and update the corresponding dimension value of the user's interest vector. For example: If the audience likes a certain type of product, the value of the corresponding dimension of this type of product in the user's interest vector is increased. In addition, the user interest vector can be normalized to ensure that the sum of each dimension is 1. Finally, the updated user interest vector can be returned.

[0114] Data output and interface: Output the updated user portrait data to the recommendation engine as the basis for generating personalized recommendations. Provide API interface to support other systems or services to call the audience behavior analysis module.

[0115] Fourth, the product recommendation engine module can integrate the output of the context perception module and the audience behavior analysis module, and use advanced recommendation algorithms to generate a personalized product recommendation list. This module also has a dynamic adjustment function, which can optimize the recommendation strategy according to the real-time changes in the live broadcast content and audience behavior, thereby providing more accurate and timely recommendation services. The specific process is as follows Figure 3C shown.

[0116] 1) Data Integration

[0117] Receive live broadcast room context information (such as theme, atmosphere, time, etc.) provided by the context perception module. Receive user portrait data (such as interest preferences, consumption habits, etc.) provided by the audience behavior analysis module. Integrate the above data to form a comprehensive user-context feature vector.

[0118] 2) Recommendation algorithm selection

[0119] Adopt the current cutting-edge deep learning recommendation algorithms, such as the recommendation system based on graph neural network (GNN), the recommendation model based on Transformer, etc. These algorithms can deeply explore the complex relationship between users and items, as well as the impact of contextual information on user preferences.

[0120] 3) Model training and optimization

[0121] The recommendation model is trained with a large amount of historical data, enabling it to learn the potential connections between user behavior, item characteristics, and contextual information. Online learning technology is used to update model parameters in real time to adapt to changes in live content and audience behavior.

[0122] It should be noted that online learning is a machine learning paradigm that allows the model to update its parameters in real time as it receives new data. Online learning is very effective for processing data streams or real-time data because it can quickly adapt to changes in data.

[0123] 4) Recommendation list generation

[0124] Based on the trained model and the user-context feature vector input in real time, predict the items that the user may be interested in. Sort the prediction results and generate a personalized product recommendation list.

[0125] 5) Dynamic adjustment and feedback

[0126] Monitor the changes in live content and audience behavior in real time. Dynamically adjust the recommendation strategy based on these changes, such as adjusting the parameters of the recommendation model, re-ordering the recommendation list, etc. And collect user feedback on the recommendation results to further optimize the recommendation model.

[0127] As an example, during the live broadcast, the system can continuously monitor and obtain the comments, likes, and gifts of online viewers in the comments and public screens about the products they are interested in. The recommendation model will adjust the products that the anchor needs to display next in real time based on the concentration of interest shown by the audience, combined with the relevance and inventory of the currently displayed products. For example, real-time feedback learning: if the click-through rate and purchase rate of a recommended product by online viewers suddenly increase, the system can update the recommendation weight of the product in real time to increase its position in the recommendation list. Another example is audience behavior analysis: if the system detects that historical viewers have shown a higher interest in a specific category of products in a certain period of time (such as through clicks, comments, or sharing behaviors), the recommendation system can adjust parameters to increase the frequency of recommendations for such products. Another example is inventory and promotion information update: when inventory is tight or there are new promotions, the recommendation system can adjust the recommendation list to give priority to products with sufficient inventory or discounts. Another example is related to advertising: during the live broadcast, if multiple brands want to promote their own products, the recommendation system can adjust the order of product recommendations based on real-time bidding results.

[0128] also, Figure 2 The real-time commodity inventory management module in is also a key component. It can continuously monitor the inventory status of each commodity through the real-time inventory management system to ensure the accuracy and update of inventory data. The system keeps in real-time synchronization with the commodity inventory database, and any changes in inventory can be reflected in the recommendation system immediately. Before the recommendation engine generates a list of recommended commodities, the real-time inventory management system verifies the inventory status of the recommended commodities to avoid recommending sold-out or low-stock commodities to users. At the same time, the system can dynamically adjust inventory according to sales speed and forecasting models, replenish stocks in time or adjust recommendation strategies to avoid overselling. When the inventory of commodities is lower than the preset threshold, the system will automatically issue an early warning to prompt the operation team to take corresponding measures, such as increasing inventory or adjusting recommendation priorities. Through these mechanisms, the purchase availability of recommended commodities is ensured, and user satisfaction and purchase conversion rate are improved.

[0129] Further references Figure 4 , as a response to the above Figures 1 to 3C The present disclosure provides some embodiments of an item recommendation device for an interactive platform. Figures 1 to 3C The device can be specifically applied to various electronic devices.

[0130] like Figure 4As shown, some embodiments of an item recommendation device 400 for an interactive platform may include: a context determination unit 401, configured to analyze and process content data of a target interactive scene on the interactive platform to determine context information of the target interactive scene; a user data determination unit 402, configured to determine portrait data of a target user based on attribute information of the target user and behavior data in the target interactive scene, wherein the target user is a user in the target interactive scene; a recommended item generation unit 403, configured to generate an item recommendation list based on context information, portrait data and a pre-trained recommendation model; and an item display unit 404, configured to display items indicated in the item recommendation list in the target interactive scene.

[0131] In some embodiments, the context determination unit 401 can be further configured to use a natural language processing model to analyze and process the content data of the target interactive scene, and extract text feature data, wherein the content data includes at least one of video data, audio data, and text data; extract key data from the text feature data, and determine the context information of the target interactive scene based on the key data, wherein the key data includes at least one of the following: content theme, keywords, and emotional information.

[0132] In some embodiments, the context determination unit 401 can be further configured to use a topic model to extract content topics from text feature data; use a keyword extraction model to extract keywords from text feature data; use a sentiment analysis model to perform sentiment prediction on text feature data to obtain sentiment information; perform a fusion analysis on the extracted content topics, keywords and sentiment information to determine the context information of the target interaction scene; and the item recommendation device 400 also includes a context update unit (not shown in the figure), which is configured to update the context information of the target interaction scene in real time based on the collected real-time content data of the target interaction scene.

[0133] In some embodiments, the user data determination unit 402 can be further configured to extract key behavior features of the target user's behavior data in the target interaction scenario, wherein the key behavior features include at least one of the following: interaction frequency, interaction method, comment content, comment sentiment information, and dwell time distribution; use a natural language processing model to analyze the comment content in the key behavior features and extract key data; determine the target user's portrait data based on the target user's attribute information, and the extracted key behavior features and key data, and store the portrait data, wherein the attribute information includes at least one of the following: basic attribute information, historical behavior data on the interactive platform, and behavior data in the set application.

[0134] In some embodiments, the item recommendation device 400 may also include a user data updating unit (not shown in the figure), which is configured to update the target user's portrait data in real time based on the collected real-time behavior data of the target user in the target interaction scenario, wherein the portrait data includes the user's preference data.

[0135] In some embodiments, the recommended item generation unit 404 can be further configured to input context information and portrait data into a pre-trained recommendation model to determine the target user's predicted score for the item; select a specified number of items from each item in order of predicted score from high to low to generate an item recommendation list.

[0136] In some embodiments, the item recommendation device 400 may further include a recommendation strategy adjustment unit (not shown in the figure), which is configured to adjust the recommendation strategy according to dynamically changing data to adjust the items or the order in the item recommendation list, wherein the dynamically changing data includes at least one of the following: real-time content data of the target interactive scene, real-time behavior data of the target user in the target interactive scene, feedback data of the target user on the items in the item recommendation list, item inventory data, and item activity information.

[0137] In some embodiments, the item recommendation device 400 may also include a data acquisition unit (not shown in the figure), which is configured to collect scene data in the target interactive scene and user data of the target user in the target interactive scene; clean the scene data and user data, and use the cleaned scene data as content data of the target interactive scene, and use the cleaned user data as behavior data of the target user in the target interactive scene.

[0138] It is understandable that the units recorded in the item recommendation device 400 are similar to those in the reference Figures 1 to 3C Therefore, the operations, features and beneficial effects described above for the method are also applicable to the item recommendation device 400 and the units included therein, and will not be described in detail here.

[0139] Reference below Figure 5 , which shows a structural schematic diagram of an electronic device 500 suitable for implementing some embodiments of the present disclosure. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0140] like Figure 5As shown, the electronic device 500 may include a processing device 501 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0141] Typically, the following devices may be connected to the I / O interface 505: input devices 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 507 including, for example, a speaker, a vibrator, etc.; storage devices 508 including, for example, a disk, a hard disk, a memory card, etc.; and communication devices 509. The communication devices 509 may allow the electronic device 500 to communicate with other devices wirelessly or by wire to exchange data. Figure 5 The electronic device 500 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 5 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0142] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0143] It should be noted that the computer-readable medium recorded in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0144] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0145] The computer-readable medium may be included in the electronic device; or it may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: analyzes and processes the content data of the target interactive scene on the interactive platform to determine the context information of the target interactive scene; determines the portrait data of the target user based on the attribute information of the target user and the behavior data in the target interactive scene, wherein the target user is the user in the target interactive scene; generates an item recommendation list based on the context information, the portrait data and the pre-trained recommendation model; and displays the items indicated in the item recommendation list in the target interactive scene.

[0146] In addition, computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0147] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0148] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The described units may also be provided in a processor, for example, may be described as: a processor including a context determination unit, a user data determination unit, a recommended item generation unit, and an item display unit. Among them, the names of these units do not constitute a limitation on the units themselves in certain circumstances, for example, the context determination unit may also be described as "a unit that analyzes and processes the content data of the target interactive scene on the interactive platform to determine the context information of the target interactive scene".

[0149] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0150] Some embodiments of the present disclosure further provide a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned methods for recommending items on an interactive platform.

[0151] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.

Claims

1. A method for recommending items on an interactive platform, comprising: Analyze and process content data of a target interactive scene on an interactive platform to determine context information of the target interactive scene; Determine the portrait data of the target user according to the attribute information of the target user and the behavior data in the target interaction scene, wherein the target user is a user in the target interaction scene; Generate an item recommendation list according to the context information, the portrait data and a pre-trained recommendation model; In the target interactive scene, the items indicated by the item recommendation list are displayed.

2. The method for recommending items on an interactive platform according to claim 1, wherein: The analyzing and processing the content data of the target interactive scene on the interactive platform to determine the context information of the target interactive scene includes: Using a natural language processing model, analyzing and processing the content data of the target interactive scene to extract text feature data, wherein the content data includes at least one of video data, audio data, and text data; Extract key data from the text feature data, and determine context information of the target interactive scene based on the key data, wherein the key data includes at least one of the following: content theme, keywords, and emotional information.

3. The method for recommending items on an interactive platform according to claim 2, wherein: The extracting key data from the text feature data, and determining the context information of the target interactive scene according to the key data, includes: Using a topic model to extract content topics from the text feature data; Using a keyword extraction model to extract keywords from the text feature data; Using a sentiment analysis model to perform sentiment prediction on the text feature data to obtain sentiment information; Performing a fusion analysis on the extracted content topics, keywords and sentiment information to determine the context information of the target interactive scene; and The method further comprises: According to the collected real-time content data of the target interactive scene, the context information of the target interactive scene is updated in real time.

4. The method for recommending items on an interactive platform according to claim 1, wherein: Determining the target user's portrait data according to the target user's attribute information and the behavior data in the target interaction scenario includes: Extracting key behavior features from the behavior data of the target user in the target interaction scenario, wherein the key behavior features include at least one of the following: interaction frequency, interaction mode, comment content, comment sentiment information, and stay time distribution; A natural language processing model is used to analyze the comments in the key behavior features and extract key data; Based on the attribute information of the target user, and the extracted key behavior features and key data, the portrait data of the target user is determined, and the portrait data is stored, wherein the attribute information includes at least one of the following: basic attribute information, historical behavior data on the interactive platform, and behavior data in a set application.

5. The method for recommending items on an interactive platform according to claim 4, wherein: The method further comprises: Based on the collected real-time behavior data of the target user in the target interactive scenario, the portrait data of the target user is updated in real time, wherein the portrait data includes the user's preference data.

6. The method for recommending items on an interactive platform according to claim 1, wherein: The generating of an item recommendation list according to the context information, the portrait data and a pre-trained recommendation model includes: Inputting the context information and the portrait data into a pre-trained recommendation model to determine the predicted rating of the target user for the item; In descending order of predicted scores, a specified number of items are selected from each item to generate an item recommendation list.

7. The method for recommending items on an interactive platform according to claim 6, wherein: The method further comprises: The recommendation strategy is adjusted according to dynamically changing data to adjust the items or the order in the item recommendation list, wherein the dynamically changing data includes at least one of the following: real-time content data of the target interactive scene, real-time behavior data of the target user in the target interactive scene, feedback data of the target user on the items in the item recommendation list, item inventory data, and item activity information.

8. The item recommendation method for an interactive platform according to any one of claims 1 to 7, wherein: The method further comprises: Collecting scene data in the target interactive scene and user data of the target user in the target interactive scene; The scene data and the user data are cleaned, and the cleaned scene data is used as the content data of the target interactive scene, and the cleaned user data is used as the behavior data of the target user in the target interactive scene.

9. An item recommendation device for an interactive platform, comprising: A context determination unit configured to analyze and process content data of a target interactive scene on an interactive platform to determine context information of the target interactive scene; A user data determination unit is configured to determine the portrait data of the target user according to the attribute information of the target user and the behavior data in the target interaction scene, wherein the target user is a user in the target interaction scene; A recommended item generation unit, configured to generate an item recommendation list according to the context information, the portrait data and a pre-trained recommendation model; The item display unit is configured to display the items indicated by the item recommendation list in the target interactive scene.

10. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the item recommendation method for an interactive platform as described in any one of claims 1-8.

11. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the item recommendation method for an interactive platform as described in any one of claims 1 to 8 is implemented.

12. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for recommending items for an interactive platform as claimed in any one of claims 1 to 8 is implemented.

Images