Intelligent operation and management method and system for commodity auction based on artificial intelligence

Abstract

The present invention relates to the field of intelligent operation management technology, and in particular to an intelligent operation management method and system for commodity auctions based on artificial intelligence. The method comprises the following steps: collecting multi-source feature data of historical auction creation demonstration; involving a bipartite graph of user interest auctions based on the multi-source feature data of historical auction creation demonstration; analyzing dynamic prediction data of user interest auction propagation through the bipartite graph of user interest auctions; performing multi-window push processing of real-time interest auction content according to the dynamic prediction data of user interest auction propagation, and generating multi-window push data of real-time interest auction content; analyzing the probability distribution characteristics of user interest in auction creation demonstration behavior; performing intelligent layout analysis of real-time interest auction content multi-window push data through the probability distribution characteristics of user interest in auction creation demonstration behavior, and generating intelligent layout data of real-time auction content push. The present invention realizes intelligent commodity auction push operation management operations.

Description

Technical Field

[0001] The present invention relates to the field of intelligent operation management technology, and in particular to an intelligent operation management method and system for commodity auctions based on artificial intelligence. Background Art

[0002] Auction platforms still face many challenges, especially in terms of user interest matching, item recommendations, and auction efficiency improvement. Most online auction platforms still rely mainly on manual display and promotion strategies for items during operation. With the rise of artificial intelligence technology, it is particularly important to be able to better mine and analyze user behavior and provide intelligent operation and management support for auction platforms. However, the existing intelligent operation and management methods for commodity auctions lack intelligent encapsulation and dynamic analysis of item attributes, and cannot effectively combine item characteristics and user needs to provide an efficient matching mechanism; the accuracy of capturing and matching user interests is low, and it is impossible to dynamically track and analyze changes in user interests during the auction process; and the display form and promotion methods of items lack flexible and intelligent processing, and cannot make accurate push notifications based on users' real-time interests. Summary of the invention

[0003] Based on this, the present invention provides an artificial intelligence-based intelligent operation and management method and system for commodity auctions to solve at least one of the above technical problems.

[0004] To achieve the above purpose, an artificial intelligence-based commodity auction intelligent operation and management method includes the following steps:

[0005] Step S1: collecting multi-source features of historical auction item creation demonstration based on the commodity auction streaming media platform to generate multi-source feature data of historical auction item creation demonstration, wherein the multi-source feature data of historical auction item creation demonstration includes auction item creation demonstration video stream data and user interaction behavior data;

[0006] Step S2: Perform visual auction item attribute feature analysis and processing based on the auction item creation demonstration video stream data to generate auction item attribute feature data; perform auction item package instantiation object design based on the auction item attribute feature data to generate auction item package instantiation object data;

[0007] Step S3: Analyze the user interest interaction behavior characteristics according to the user interaction behavior data to generate user interest interaction behavior characteristic data;

[0008] Step S4: Design a matching bipartite graph between user interests and auction items based on the auction item encapsulation instantiation object data and the user interest interaction behavior feature data to generate a user interest auction item bipartite graph; perform dynamic prediction processing on the user interest auction item propagation based on the user interest auction item bipartite graph to generate dynamic prediction data on the user interest auction item propagation;

[0009] Step S5: Perform decision analysis on pushing user-interested items based on the dynamic prediction data of user-interested items, and generate decision data on pushing user-interested items;

[0010] Step S6: Based on the auction item creation demonstration video stream data and the user interest interaction behavior feature data, a probability distribution analysis of the user interest in the auction item creation demonstration behavior is performed to generate the user interest probability distribution data of the auction item creation demonstration behavior; and a personalized auction item creation demonstration interest prediction model is constructed according to the user interest probability distribution data of the auction item creation demonstration behavior;

[0011] Step S7: Perform real-time interest item content multi-window push processing based on the user's interest item push decision data to generate real-time interest item content multi-window push data; perform real-time interest item push content intelligent layout analysis on the real-time interest item content multi-window push data based on the personalized item creation demonstration interest prediction model to generate real-time item push content intelligent layout data; perform commodity auction push intelligent operation management operations through real-time item push content intelligent layout data.

[0012] The present invention collects multi-source features of the historical creation demonstration of the commodity auction streaming media platform, generates multi-source feature data including the video stream data of the creation demonstration of the auction item and the user interaction behavior data, and can provide a multi-dimensional information source for subsequent analysis and processing. Through the combination of video data and user interaction data, the attributes of the auction item and the user behavior can be fully understood, providing a basis for intelligent matching and recommendation. Visual analysis of the video stream data of the creation demonstration of the auction item can automatically extract the visual attribute features of the auction item, such as color, shape, material, etc., and generate the attribute feature data of the auction item. At the same time, the auction item encapsulation instantiation object design is performed based on these feature data to realize the standardized encapsulation of the auction item. This not only helps to improve the quality and consistency of the auction item display, but also provides structured input data for the subsequent recommendation and matching algorithms, and improves the processing efficiency and accuracy of the system. The analysis of the user interaction behavior data can accurately identify the user's interest points and behavior patterns, generate user interest interaction behavior feature data, help the system understand the user's preferences and habits, and thus lay the foundation for the subsequent interest and auction item matching. Accurate user interest recognition can significantly improve the relevance of recommendations and user experience. By combining the encapsulated instantiation object data of the auction items with the characteristic data of the user's interest interaction behavior, and designing a matching bipartite graph of user interests and auction items, it is possible to establish the association between users and auction items, efficiently reveal the user's potential interest in specific auction items, and improve the accuracy of recommendations. The propagation dynamic prediction processing based on this bipartite graph can also further predict the changing trend of user interests, provide a more forward-looking judgment basis for auction item push, and enhance the intelligence and responsiveness of the system. Push decision analysis is performed based on the dynamic prediction data of the user's interest auction item propagation, intelligently select the auction items that are most suitable for users, and determine the best push time. This step effectively improves the accuracy and timeliness of auction item push, avoids the problems of information redundancy or inappropriate timing in traditional auction push methods, and thus improves user participation and auction item transaction rate. The user interest probability distribution analysis of the auction item creation and demonstration behavior can identify the user's interest preferences during the auction item display process according to the user's behavior trajectory and interest changes, and generate interest probability distribution data. These data provide strong support for the construction of personalized interest prediction models, and can accurately predict the future changes in user interests, thereby providing a more accurate reference basis for the system's personalized recommendations and improving user experience. According to the decision data of the auction items that users are interested in, the auction items that users are interested in are pushed in real time in multiple windows, and the auction items are displayed in multiple dimensions to enhance the user's sense of participation. At the same time, based on the personalized interest prediction model, the system can perform intelligent layout analysis on the real-time pushed content to ensure that the display method of the pushed content is in line with the user's auction item creation, demonstration and browsing habits. This not only optimizes the display effect of the auction items, but also increases the acceptance rate of the pushed information, further improving the operational efficiency of the auction platform.

[0013] This specification provides an artificial intelligence-based commodity auction intelligent operation and management system, which is used to execute the artificial intelligence-based commodity auction intelligent operation and management method described above. The artificial intelligence-based commodity auction intelligent operation and management system includes:

[0014] A historical auction item creation demonstration multi-source feature collection module is used to collect multi-source features of historical auction item creation demonstration based on a commodity auction streaming media platform, and generate multi-source feature data of historical auction item creation demonstration, wherein the multi-source feature data of historical auction item creation demonstration includes auction item creation demonstration video stream data and user interaction behavior data;

[0015] The auction item visual attribute analysis module is used to perform visual auction item attribute feature analysis based on the auction item creation demonstration video stream data to generate auction item attribute feature data; perform auction item package instantiation object design based on the auction item attribute feature data to generate auction item package instantiation object data;

[0016] A user interest interaction behavior feature analysis module, used to analyze user interest interaction behavior features based on user interaction behavior data and generate user interest interaction behavior feature data;

[0017] A user interest item propagation dynamic prediction module is used to design a bipartite graph matching user interests and items based on item encapsulation instantiation object data and user interest interaction behavior feature data, and generate a user interest item bipartite graph; based on the user interest item bipartite graph, a user interest item propagation dynamic prediction process is performed to generate user interest item propagation dynamic prediction data;

[0018] A user interest item push decision analysis module, which is used to perform user interest item push decision analysis based on user interest item propagation dynamic prediction data, and generate user interest item push decision data;

[0019] The module for constructing a model for predicting the interest in auction item creation demonstration is used to analyze the probability distribution of user interest in auction item creation demonstration behavior based on the auction item creation demonstration video stream data and the user interest interaction behavior feature data, and generate the probability distribution data of user interest in auction item creation demonstration behavior; and construct a personalized auction item creation demonstration interest prediction model based on the probability distribution data of user interest in auction item creation demonstration behavior;

[0020] The intelligent operation job execution module is used to perform real-time interest auction item content multi-window push processing based on the user's interest auction item push decision data, and generate real-time interest auction item content multi-window push data; based on the personalized auction item creation demonstration interest prediction model, the real-time interest auction item content multi-window push data is subjected to real-time auction item push content intelligent layout analysis, and real-time auction item push content intelligent layout data is generated; and the commodity auction push intelligent operation management job is executed through the real-time auction item push content intelligent layout data.

[0021] The beneficial effect of the present application is that the present invention automatically analyzes and instantiates the visual attributes of the auction items into standardized object data based on the visual feature extraction and packaging of the auction item creation demonstration video stream data, and can intelligently match the auction items with users by dynamically analyzing the relationship between the auction item attributes and user interests, thereby effectively improving the efficiency of recommendation. By intelligently packaging the attribute characteristics of the auction items, an operable instantiated object is generated for each auction item, thereby achieving efficient matching of the auction items with user needs. This automated and intelligent processing method improves the accuracy of the system in processing the auction item characteristics and user interests, and optimizes the push effect of the auction items. By collecting and analyzing the user interaction behavior data, the user interest interaction behavior feature data is generated. At the same time, based on the deep learning algorithm such as the long short-term memory neural network (LSTM), the dynamic propagation prediction of user interests can be carried out, which can capture the changes in user interests in the online auction process in real time, effectively solving the problem that the existing technology cannot dynamically track and analyze the changes in user interests, and provides a comprehensive and accurate capture and prediction capability of user interests, which can perceive the fluctuation of user interests in real time and make corresponding adjustments, improve the accuracy of auction item recommendations, and enhance the user experience and the interactivity of the platform. By generating a bipartite graph of user-interested auction items and combining it with real-time user interest prediction, the push content can be dynamically adjusted according to the user's current needs, and the intelligent layout technology is designed to optimize the display form of the auction items by analyzing the auction item creation and display mode that the user likes, ensuring that users can browse auction items that meet their interests and prioritize display based on real-time auction item creation and display content. The display and push methods of auction items can flexibly adapt to the user's real-time interests, ensure that the push content is highly consistent with user needs, avoid interference from irrelevant content, and improve the accuracy and display effect of push, thereby increasing the exposure rate and transaction rate of auction items. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a schematic diagram of the steps of an intelligent operation and management method for commodity auctions based on artificial intelligence of the present invention;

[0023] Figure 2 for Figure 1 Detailed implementation steps of step S4 in FIG.

[0024] Figure 3 for Figure 1 Detailed implementation steps of step S6 in the flowchart;

[0025] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0026] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0027] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0028] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0029] To achieve this, please refer to Figures 1 to 3 The present invention provides an intelligent operation and management method for commodity auction based on artificial intelligence, comprising the following steps:

[0030] Step S1: collecting multi-source features of historical auction item creation demonstration based on the commodity auction streaming media platform to generate multi-source feature data of historical auction item creation demonstration, wherein the multi-source feature data of historical auction item creation demonstration includes auction item creation demonstration video stream data and user interaction behavior data;

[0031] Step S2: Perform visual auction item attribute feature analysis and processing based on the auction item creation demonstration video stream data to generate auction item attribute feature data; perform auction item package instantiation object design based on the auction item attribute feature data to generate auction item package instantiation object data;

[0032] Step S3: Analyze the user interest interaction behavior characteristics according to the user interaction behavior data to generate user interest interaction behavior characteristic data;

[0033] Step S4: Design a matching bipartite graph between user interests and auction items based on the auction item encapsulation instantiation object data and the user interest interaction behavior feature data to generate a user interest auction item bipartite graph; perform dynamic prediction processing on the user interest auction item propagation based on the user interest auction item bipartite graph to generate dynamic prediction data on the user interest auction item propagation;

[0034] Step S5: Perform decision analysis on pushing user-interested items based on the dynamic prediction data of user-interested items, and generate decision data on pushing user-interested items;

[0035] Step S6: Based on the auction item creation demonstration video stream data and the user interest interaction behavior feature data, a probability distribution analysis of the user interest in the auction item creation demonstration behavior is performed to generate the user interest probability distribution data of the auction item creation demonstration behavior; and a personalized auction item creation demonstration interest prediction model is constructed according to the user interest probability distribution data of the auction item creation demonstration behavior;

[0036] Step S7: Perform real-time interest item content multi-window push processing based on the user's interest item push decision data to generate real-time interest item content multi-window push data; perform real-time interest item push content intelligent layout analysis on the real-time interest item content multi-window push data based on the personalized item creation demonstration interest prediction model to generate real-time item push content intelligent layout data; perform commodity auction push intelligent operation management operations through real-time item push content intelligent layout data.

[0037] The present invention collects multi-source features of the historical creation demonstration of the commodity auction streaming media platform, generates multi-source feature data including the video stream data of the creation demonstration of the auction item and the user interaction behavior data, and can provide a multi-dimensional information source for subsequent analysis and processing. Through the combination of video data and user interaction data, the attributes of the auction item and the user behavior can be fully understood, providing a basis for intelligent matching and recommendation. Visual analysis of the video stream data of the creation demonstration of the auction item can automatically extract the visual attribute features of the auction item, such as color, shape, material, etc., and generate the attribute feature data of the auction item. At the same time, the auction item encapsulation instantiation object design is performed based on these feature data to realize the standardized encapsulation of the auction item. This not only helps to improve the quality and consistency of the auction item display, but also provides structured input data for the subsequent recommendation and matching algorithms, and improves the processing efficiency and accuracy of the system. The analysis of the user interaction behavior data can accurately identify the user's interest points and behavior patterns, generate user interest interaction behavior feature data, help the system understand the user's preferences and habits, and thus lay the foundation for the subsequent interest and auction item matching. Accurate user interest recognition can significantly improve the relevance of recommendations and user experience. By combining the encapsulated instantiation object data of the auction items with the characteristic data of the user's interest interaction behavior, and designing a matching bipartite graph of user interests and auction items, it is possible to establish the association between users and auction items, efficiently reveal the user's potential interest in specific auction items, and improve the accuracy of recommendations. The propagation dynamic prediction processing based on this bipartite graph can also further predict the changing trend of user interests, provide a more forward-looking judgment basis for auction item push, and enhance the intelligence and responsiveness of the system. Push decision analysis is performed based on the dynamic prediction data of the user's interest auction item propagation, intelligently select the auction items that are most suitable for users, and determine the best push time. This step effectively improves the accuracy and timeliness of auction item push, avoids the problems of information redundancy or inappropriate timing in traditional auction push methods, and thus improves user participation and auction item transaction rate. The user interest probability distribution analysis of the auction item creation and demonstration behavior can identify the user's interest preferences during the auction item display process according to the user's behavior trajectory and interest changes, and generate interest probability distribution data. These data provide strong support for the construction of personalized interest prediction models, and can accurately predict the future changes in user interests, thereby providing a more accurate reference basis for the system's personalized recommendations and improving user experience. According to the decision data of the auction items that users are interested in, the auction items that users are interested in are pushed in real time in multiple windows, and the auction items are displayed in multiple dimensions to enhance the user's sense of participation. At the same time, based on the personalized interest prediction model, the system can perform intelligent layout analysis on the real-time pushed content to ensure that the display method of the pushed content is in line with the user's auction item creation, demonstration and browsing habits. This not only optimizes the display effect of the auction items, but also increases the acceptance rate of the pushed information, further improving the operational efficiency of the auction platform.

[0038] As an embodiment of the present invention, refer to Figure 1 The above is a schematic diagram of the steps of an intelligent operation and management method for commodity auctions based on artificial intelligence of the present invention. In this embodiment, the intelligent operation and management method for commodity auctions based on artificial intelligence includes the following steps:

[0039] Step S1: collecting multi-source features of historical auction item creation demonstration based on the commodity auction streaming media platform to generate multi-source feature data of historical auction item creation demonstration, wherein the multi-source feature data of historical auction item creation demonstration includes auction item creation demonstration video stream data and user interaction behavior data;

[0040] In an embodiment of the present invention, multi-source feature data of historical auction creation demonstration is collected through a commodity auction streaming media platform, and the multi-source feature data includes video stream data of the auction creation demonstration and user interaction behavior data. In actual operation, the video stream of the auction creation demonstration is recorded through real-time multimedia transmission methods such as a live broadcast room, recording the multi-angle detail display, material display, and specific design details during the auction display process. The video stream data is saved at a specific frame rate and resolution, and the relevant timestamp is marked to ensure that subsequent visual analysis can be accurately aligned in time. At the same time, through the interactive function of the platform, the interactive behavior data of the user when watching the video is collected, and the user behavior data includes operations such as click volume, viewing time, page jump frequency, collection or sharing. These data are recorded in real time by the server side of the platform, and the user behavior is associated with a specific time point in the video stream to facilitate the analysis of the user's interest in certain auction details. The generated historical auction creation demonstration multi-source feature data includes two main parts: the auction creation demonstration video stream data and the user interaction behavior data, both of which are processed by data standardization and stored in the database for subsequent analysis.

[0041] Step S2: Perform visual auction item attribute feature analysis and processing based on the auction item creation demonstration video stream data to generate auction item attribute feature data; perform auction item package instantiation object design based on the auction item attribute feature data to generate auction item package instantiation object data;

[0042] In the embodiment of the present invention, a computer vision algorithm is used to analyze each frame in the video stream, and the visual features of the auction items, such as color, shape, material, texture and other attributes, are extracted through feature detection technology. For each type of auction item, feature matching is performed according to a pre-set attribute model to ensure the accuracy and consistency of feature extraction. Image processing technology can also automatically identify the special logos, artistic features or unique design elements of the auction items. The generated auction item attribute feature data then enters the encapsulation process to encapsulate these feature data into instantiated objects. The instantiated object includes the main visual attributes of the auction item and some additional metadata (such as the time and angle of display) for subsequent recommendation and matching. The auction item encapsulation instantiated object data is stored in the system's database as a standardized data source for future auction item display, analysis and push.

[0043] Step S3: Analyze the user interest interaction behavior characteristics according to the user interaction behavior data to generate user interest interaction behavior characteristic data;

[0044] In an embodiment of the present invention, the user's interactive behavior data is preprocessed by data mining technology to extract the interactive information between the user and the auction items. For example, the system analyzes the user's dwell time, interaction frequency (such as likes, comments, collections, etc.) when watching a video, and the user's historical behavior pattern. For each user, the system analyzes their comments and interactive texts through natural language processing technology to understand the user's preferences and interests in the auction items. Then, using machine learning models (such as collaborative filtering algorithms and user interest prediction models), these interactive behavior characteristics are analyzed to generate user interest feature data. This data mainly includes the user's interest categories, the types of auction items they are interested in, and the changing trends of user behavior patterns. The user interest interactive behavior feature data is then used to further optimize the push decision and recommendation algorithm of the auction items, ensuring that the system can provide more accurate auction item display and interactive services according to the personalized needs of users.

[0045] Step S4: Design a matching bipartite graph between user interests and auction items based on the auction item encapsulation instantiation object data and the user interest interaction behavior feature data to generate a user interest auction item bipartite graph; perform dynamic prediction processing on the user interest auction item propagation based on the user interest auction item bipartite graph to generate dynamic prediction data on the user interest auction item propagation;

[0046] In the embodiment of the present invention, the instantiated object data of the lot encapsulation and the user interest interaction behavior characteristic data are extracted from the database. For each lot, the instantiated object data provides the visual features, type and related metadata of the lot, while the user interest interaction behavior characteristic data records the interaction and preference information of each user for different types of lots. Using the graph model design method, the user and the lot are regarded as nodes of the graph, and the matching degree of the user's interest and the lot is connected as the edge between the nodes. The generated bipartite graph clearly shows the strength of the association between the user and the lot, and the edge between the user and the lot node with a higher matching degree will show a stronger connection. This matching relationship can reflect the user's potential interest in a specific lot, thereby providing a reference for subsequent recommendations and push. Based on the generated user interest lot bipartite graph, the system further performs dynamic prediction processing of the propagation of the user's interest lot. Specifically, the system predicts the changing trend of user interests in the future through a dynamic network analysis algorithm. For example, the system determines whether the user may increase or decrease his interest in a certain type of lot by analyzing the changes in the connection of the user in the bipartite graph. This process combines deep learning models such as the long short-term memory (LSTM) neural network to generate dynamic prediction data on the dissemination of user-interested items through learning from time series data. Finally, the system generates a detailed report based on the predicted data, explaining the types of items that users may be interested in in the future and the related interactive behavior trends. These data provide an accurate basis for the next decision on item push.

[0047] Step S5: Perform decision analysis on pushing user-interested items based on the dynamic prediction data of user-interested items, and generate decision data on pushing user-interested items;

[0048] In an embodiment of the present invention, dynamic prediction data on the dissemination of user-interested items is received and parsed. This data provides prediction results of the user's interest changes in different items over a period of time in the future. The system uses a push decision algorithm to analyze which items are more in line with the user's current needs, and based on this information, provides the user with the best item push solution. The push decision analysis process comprehensively considers multiple factors, including the user's current interest intensity, the characteristics and type of the items, the appropriateness of the auction time, and the user's historical behavior records for similar items. The push algorithm uses a weight distribution mechanism to comprehensively score these factors, give priority to relevant online auction product information that has a high degree of match with the user's interests and has a good predicted dissemination effect, generate user-interested items push decision data, and generate customized item push strategies for different users. Different push frequencies and content display methods will be used for different users to ensure user participation and item exposure.

[0049] Step S6: Based on the auction item creation demonstration video stream data and the user interest interaction behavior feature data, a probability distribution analysis of the user interest in the auction item creation demonstration behavior is performed to generate the user interest probability distribution data of the auction item creation demonstration behavior; and a personalized auction item creation demonstration interest prediction model is constructed according to the user interest probability distribution data of the auction item creation demonstration behavior;

[0050] In an embodiment of the present invention, video stream data of the auction creation demonstration and user interaction behavior data (such as viewing time, click frequency, collection or comment behavior) are extracted. A probability model is used to perform a probability distribution analysis of user interests, and quantify the changes in user interests at different times or in different video clips. For example, when a user stays for a long time and interacts frequently with a certain auction display part, the system determines that the user has a high interest in the display part and generates corresponding probability distribution data. The user interest probability distribution data not only reflects the user's interest intensity in a specific period of time, but also describes the user's behavior pattern during the auction display process. These data provide a basis for subsequent personalized recommendations and push notifications. On the basis of generating user interest probability distribution data, a personalized auction creation demonstration interest prediction model is further constructed. By using machine learning algorithms (such as variational autoencoders and deep learning models), the system can predict future changes in user interests and identify in advance which auction displays the user may be interested in. The model continuously optimizes the accuracy of prediction by continuously learning the user's historical behavior data and interest change trends, providing efficient support for subsequent push decisions.

[0051] Step S7: Perform real-time interest item content multi-window push processing based on the user's interest item push decision data to generate real-time interest item content multi-window push data; perform real-time interest item push content intelligent layout analysis on the real-time interest item content multi-window push data based on the personalized item creation demonstration interest prediction model to generate real-time item push content intelligent layout data; perform commodity auction push intelligent operation management operations through real-time item push content intelligent layout data.

[0052] In an embodiment of the present invention, after receiving the decision data of the user's interest item push, the user's current interest point is combined to select the appropriate item for multi-window display push. Multi-window push processing can simultaneously display the different characteristics of multiple items, allowing users to quickly browse, compare and filter items. The push content is prioritized according to the user's interest, and the system ensures real-time response to the user's interest changes during push. The real-time push data is transmitted to the personalized item creation demonstration interest prediction model for further interest prediction processing. The model adjusts the prediction results in real time by analyzing the user's feedback on different push content, ensures that the push content is highly matched with the user's current interest, and generates real-time push content interest prediction data. According to the interest prediction data of the real-time push content, the multi-window push content is intelligently typed and analyzed. Through the intelligent typesetting algorithm, the item creation display content that the user is interested in is limitedly placed in a prominent position, and the window layout and display effect are optimized at the same time, ensuring that the user can see the content of interest in the first time. The real-time item push content intelligent typesetting data generated by the analysis can further enhance the user's interactive experience with the item. The intelligent operation and management of commodity auction push is performed through the real-time item push content intelligent typesetting data. This process ensures that the pushed content is displayed to users in a timely and accurate manner, and dynamically adjusts the order and layout of the pushed content based on user feedback and behavior. Intelligent operation not only reduces manual intervention, but also improves push efficiency and user satisfaction, further promoting the transaction rate of auction items and the operational efficiency of the platform.

[0053] Preferably, step S1 comprises the following steps:

[0054] Step S11: monitoring and processing the multi-source data of the creation and demonstration of auction items according to the commodity auction streaming media platform to generate the multi-source data of the creation and demonstration of auction items;

[0055] Step S12: performing data synchronization processing on the auction item creation demonstration multi-source data to generate synchronized auction item creation demonstration multi-source data;

[0056] Step S13: extract historical multi-source features based on the synchronized auction item creation demonstration multi-source data to generate historical auction item creation demonstration multi-source feature data.

[0057] The present invention monitors and processes multi-source data for the creation and demonstration of auction items on a commodity auction streaming media platform, and collects auction item-related data from multiple dimensions in real time, such as video streams, images, user interaction behaviors, user comments, etc. The collection of multi-source data can provide rich contextual information, help the system form a more comprehensive understanding of auction items between different data dimensions, greatly improve the perception ability of the system, and provide more accurate basic data for subsequent data processing and analysis. By synchronously processing the collected multi-source data, it is ensured that various types of data (such as video frames, user interaction events, etc.) can maintain consistency in the time dimension during the creation and demonstration of auction items. The benefit of data synchronization is that data from different sources can be better used in conjunction to avoid timing problems caused by data asynchrony, so that the system can more accurately analyze the relationship between user behavior and auction item features, and provide a more reliable data basis for subsequent feature extraction and behavior analysis. Extracting historical multi-source features from the synchronous auction item creation and demonstration multi-source data can extract key feature data reflecting the attributes of the auction items and user interaction behaviors, effectively converting massive raw data into more structured and representative feature information, and helping the system to further understand the behavior patterns of historical auction items and users. At the same time, the extraction of historical features provides valuable training data for subsequent machine learning models, enhancing the system's ability to predict user interests and recommend auction items.

[0058] In an embodiment of the present invention, the creation process of the auction item is monitored throughout the process by a video recording device on a streaming platform to obtain multi-source data. The multi-source data includes video stream data, environmental information (such as light, angle, background, etc.) when the auction item is displayed, and information such as the auction item image, detailed description of the auction item, and introduction of the designer collected during the display process. User interaction behaviors on the platform are also recorded synchronously, such as viewing time, click volume, comments, and collections. In this way, the system can capture all aspects of the auction item display and the user's immediate response to form multi-source data for the auction item creation demonstration. Through the timing synchronization mechanism, the auction item creation demonstration data obtained from different sources are uniformly processed. For example, video stream data, image data, and user interaction behavior data all have their own independent timestamps. The time synchronization algorithm is used to align different types of data on the timeline to ensure that the specific segments of the video display are synchronized with the user's interactive behavior. After synchronization processing, the generated synchronization data can more accurately reflect the user's attention and response to the auction item at a specific time point. The key information of the historical auction item is extracted from the synchronized multi-source data through the feature extraction algorithm. For example, the visual features of the auction items (such as color, material, and design style) are extracted from the video stream data, and the user interaction behavior on the auction items (such as the emotional tendency of the comments and the frequency of collection) is extracted from the user interaction data. After these historical features are classified and processed, detailed historical feature data of the auction item creation demonstration is generated. The generated multi-source feature data of the historical auction item creation demonstration provides a reliable reference for subsequent user interest analysis, auction item recommendation, and personalized push notifications. Based on these historical data, the auction item creation display pattern recognition and user interest item prediction are carried out to better match the user's interest needs and improve the accuracy of auction item display and push notifications.

[0059] Preferably, step S2 comprises the following steps:

[0060] Step S21: performing video frame decoding processing according to the auction creation demonstration video stream data to generate auction creation demonstration video frame data;

[0061] Step S22: extract key video frames from the auction item creation demonstration video frame data to generate auction item creation demonstration key video frame data;

[0062] Step S23: Perform visual auction item attribute feature analysis and processing based on auction item creation demonstration key video frame data to generate auction item attribute feature data;

[0063] Step S24: performing data encapsulation processing on the auction item attribute feature data to generate encapsulated auction item attribute feature data;

[0064] Step S25: Designing an auction item packaging instantiation object according to the packaged auction item attribute feature data, and generating auction item packaging instantiation object data.

[0065] The present invention performs video frame decoding processing on the video stream data of the creation and demonstration of the auction item, and extracts independent video frame data from the continuous video stream. The decoding process can convert the video information into operable frame data, and help the system to perform a more detailed analysis of the display of the auction item in the video. This step not only improves the availability of the video data, but also lays a foundation for subsequent key frame extraction and attribute analysis. The video frame data is subjected to key frame extraction, and the key moments related to the display of the auction item are screened out from the massive video frames. This extraction method can effectively reduce redundant data and improve processing efficiency. The extracted key frames can focus on reflecting the main visual features of the auction item, such as appearance details, materials and design features, ensuring that the subsequent analysis is based on the most representative data, and improving the accuracy and efficiency of the analysis results. Based on the key video frame data, the visual attribute feature analysis of the auction item is performed, and the appearance features of the auction item, such as color, texture, shape, size, etc., are automatically identified. The visual information of the auction item is converted into quantifiable feature data through a computer vision algorithm, thereby providing a more accurate basis for the classification, matching and recommendation of the auction item. Through the automated visual feature analysis, the workload of manual participation is greatly reduced, and the intelligent level of data processing is improved. After the attribute feature data of the auction items are encapsulated, these attribute features are stored or transmitted in a standardized format. Encapsulation ensures the compatibility and operability of data between different system components, and improves the efficiency and stability of data exchange. The encapsulated data is directly used for subsequent instantiated object design and intelligent matching, further improving the modularization capability and processing efficiency of the system. According to the encapsulated attribute feature data of the auction items, instantiated object design is performed to generate structured digital objects for each auction item. These instantiated objects not only contain the key attributes of the auction items, but can also be efficiently matched with the interests and behaviors of users, converting abstract auction item data into operable objects, which is convenient for application in scenarios such as auction item recommendation and intelligent push, greatly enhancing the system's automated processing capabilities and intelligent push effects.

[0066] In the embodiment of the present invention, the video decoder is used to decode the video stream data of the auction creation demonstration, and the picture information of each frame is parsed. The decoding process adopts a timestamp-based mechanism to decompose each video clip into continuous video frame data. Each frame contains different angles, details, and environmental factors such as light for displaying the auction items. The frame data after decoding is stored in the system for subsequent key frame extraction and attribute analysis. The key frames are extracted from the decoded continuous video frame data by using the frame difference detection algorithm and the image change detection technology. The key video frame refers to the most representative and most variable frame for displaying the auction items, such as the frame showing the auction items from different angles, the frame with obvious light changes, or the frame showing important details. The extracted key frames can significantly reduce data redundancy while retaining enough visual information for subsequent processing. The key video frame data finally generated shows the core features of the auction items in a concentrated manner, providing high-quality input for subsequent analysis. The extracted key video frames are analyzed by computer vision technology. By using image processing algorithms such as edge detection, texture analysis, and color space transformation, various visual features of the auction items are identified, including color, shape, texture, gloss, etc. In addition, it can also identify the unique design elements or logos of the items and use this information as part of the item attribute characteristics. The generated item attribute feature data creates an independent visual feature file for each item, which is easy to classify, identify and recommend. The extracted item attribute feature data is structured through a standardized data encapsulation mechanism. The encapsulation process organizes all item attributes according to preset classification standards, including visual features such as color, shape, material, and metadata such as timestamp and display order. The main purpose of data encapsulation is to ensure that different system modules can efficiently share and access this data, improve processing speed and data transmission reliability. The encapsulated item attribute feature data generated in the end provides standardized input for the system's subsequent recommendation algorithm and push process. Using the encapsulated item attribute feature data, an instantiated object for each item is designed. The instantiated object contains important information such as the visual features of the item, the angle display order of the item, and the recommendation priority. The item encapsulated instantiated object is an important standardized object used by the system for recommendation, display and push. Through these objects, user interests and item attributes can be quickly matched, thereby achieving efficient personalized recommendations. The generated auction item encapsulation instantiation object data is stored in the database for subsequent push, display and auction item management.

[0067] Preferably, step S3 comprises the following steps:

[0068] Step S31: extracting text interaction behavior according to the user interaction behavior data to generate text interaction behavior data;

[0069] Step S32: performing text semantic analysis on the text interaction behavior data based on natural language technology to generate semantic analysis interaction behavior data;

[0070] Step S33: extract key interest text features from the semantic parsing interaction behavior data to generate key interest text feature data;

[0071] Step S34: extracting non-text interaction behaviors according to the user interaction behavior data to generate non-text interaction behavior data;

[0072] Step S35: performing non-text interaction behavior frequency feature analysis based on the non-text interaction behavior data to generate non-text interaction behavior frequency feature data;

[0073] Step S36: Perform user interest interaction behavior feature analysis based on the key interest text feature data and the non-text interaction behavior frequency feature data to generate user interest interaction behavior feature data.

[0074] The present invention extracts text interaction behavior from user interaction behavior data, and extracts text information generated by users during the auction process, such as comments, search keywords, chat records, etc. Text interaction behavior data can directly reflect user preferences and concerns, and provide a strong data basis for subsequent interest feature extraction. By extracting user text interaction behaviors, the changes in user interests on the auction platform are captured, and the sensitivity to user needs is improved. Based on natural language processing technology, semantic analysis is performed on text interaction behavior data, and the actual intention behind the user's language expression is deeply analyzed. Through semantic analysis, the deep meaning in the text, the user's emotional tendency, and the specific interest points related to the auction items are understood. Semantic analysis provides the system with a higher level of understanding ability, enhances the accurate interpretation of user behavior, and thus helps to improve the accuracy of subsequent recommendations. Key interest text features are extracted from the semantically analyzed interaction behavior data to identify the text features most relevant to the user's interests. These features include keywords repeatedly mentioned by users, specific description sentences, or emotional tendencies in comments. Key interest text feature data enables the system to effectively locate the core needs of users, and provides a key basis for the subsequent matching of interests and auction items. In addition to text interaction behaviors, non-text interaction behaviors such as clicks, browsing, collections, and bidding on the platform are equally important. By extracting these non-text interaction behavior data, we can capture the user's behavior patterns in terms of item selection and auction participation. Non-text interaction behavior data provides the system with more dimensional information, making the user portrait more complete and contributing to a more comprehensive analysis of user interest characteristics. Frequency feature analysis is performed on non-text interaction behavior data to identify the user's behavior habits at different time points or in specific scenarios. By analyzing the interaction frequency between users and items, we can find out the user's attention or preference for a certain type of item, which provides a quantitative basis for subsequent user interest feature analysis and further improves the accuracy of user interest prediction. Based on key interest text feature data and non-text interaction behavior frequency feature data, we conduct a comprehensive user interest interaction behavior feature analysis to help the system accurately capture the user's overall interest pattern and generate user interest interaction behavior feature data. By combining text and non-text data, we can have a more comprehensive understanding of the user's multi-dimensional interests, improve the accuracy of item recommendations and push notifications, and enhance user experience.

[0075] In an embodiment of the present invention, relevant information is extracted from the text interaction behaviors of users in the auction platform, which behaviors include comments, messages, questions and communication content of users on auction items. Each text interaction behavior of the user is recorded by the log monitoring and data collection module, and the user's input content is extracted. The generated text interaction behavior data contains text content, timestamp and basic information of the user, such as user ID and interaction object (auction item ID). These data provide a basis for subsequent text analysis. The extracted text data is semantically parsed using a natural language processing algorithm. First, the system preprocesses the text data, including word segmentation, removal of stop words and marking parts of speech. Using a semantic analysis model, such as word embedding or a pre-trained language model (such as BERT), the text content is semantically analyzed to extract the emotions, intentions and concerns expressed by the user in the text. For example, the user's praise, question or criticism of a certain auction item is identified. Through semantic parsing, the system generates semantic parsed interactive behavior data containing emotional tendencies and key topics. The semantically parsed data is further analyzed to extract key information related to user interests. Through keyword extraction technology and topic modeling (such as LDA topic model), we can identify the interests and preferences that users repeatedly mention, such as the material, design, function, etc. of a certain type of auction item. Key interest text feature data can accurately describe the user's interest intensity in a specific auction item or auction item feature. The generated feature data includes keywords, topic classification, and correlation score with user interests. These features provide the core basis for subsequent interest analysis and auction item recommendation. Extract important interactive data from users' non-text behaviors. Non-text interactive behaviors include users' clicks, browsing, collections, sharing, bidding, and other operations. Through system logs and interactive monitoring modules, every interactive behavior between users and the platform is recorded. The extracted non-text interactive behavior data includes behavior type, behavior object (such as a certain auction item), behavior time, and user information. These data can reflect users' interest tendencies and behavior patterns, and provide auxiliary data for further interest analysis. Perform frequency analysis on users' non-text interactive behaviors. For example, the system counts the number of times users browse a certain auction item, click frequency, bid frequency, etc., generates non-text behavior frequency features for each user, evaluates the user's interest in different auction items, and generates feature data including behavior type and frequency. Frequency feature data is used to determine whether users show sustained interest in a certain type of auction items, and to further optimize push strategies. Combining the user's text interaction features and non-text behavior features, the user's overall interest is evaluated through a multi-dimensional interest analysis model. By analyzing the user's comments and key topics mentioned in the text, combined with the user's non-text interaction frequency with the auction items, a user interest portrait is obtained, the user's preference for a certain type of auction items is identified, and each user's interest is quantitatively scored. User interest interaction behavior feature data is generated, including information such as the user's preferred auction item category, interaction intensity, and interest trend.These data provide the basis for personalized recommendations and precise push notifications.

[0076] Preferably, step S4 comprises the following steps:

[0077] Step S41: performing data association processing between the user's interest interaction behavior and the item packaging instantiation object according to the item packaging instantiation object data and the user's interest interaction behavior feature data, and generating user interest item association data;

[0078] Step S42: Design a bipartite graph matching user interests and auction items through the user-interest auction item association data to generate a user-interest auction item bipartite graph;

[0079] Step S43: Perform dynamic prediction processing on the propagation of user-interested items based on the bipartite graph of user-interested items to generate dynamic prediction data on the propagation of user-interested items.

[0080] The present invention associates the user interest interaction behavior feature data with the auction item encapsulation instantiation object data, establishes an effective association between the user behavior feature and the specific auction item attribute, and generates the user interest auction item association data, which accurately describes the user's interest in a specific auction item or a certain type of auction item. This association data provides an accurate basis for subsequent auction item recommendations, understands the user's needs and preferences, and further enhances the interaction effect between the user and the auction item. By designing a matching bipartite graph of user interest and auction items based on the user interest auction item association data, the user and the auction item are regarded as nodes in the graph, and the interest association between the user and the auction item is represented by the edge. The design of this bipartite graph can efficiently and intuitively display the different interest levels of users in multiple auction items, thereby providing a structured basis for subsequent auction item recommendations and matching optimization. The establishment of the matching bipartite graph helps to quickly find the auction items that users are potentially interested in, and improves the efficiency and accuracy of the recommendation. Based on the user interest auction item bipartite graph, the user interest auction item propagation dynamic prediction processing is performed to predict the changing trend and propagation path of the user's interest in the auction item. By understanding in advance which items attract more user attention and participation, we can optimize item push and quickly adjust item recommendations when user needs change, improving user experience and item transaction rates.

[0081] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed implementation step flow chart of step S4 in the embodiment, step S4 in this embodiment includes:

[0082] Step S41: performing data association processing between the user's interest interaction behavior and the item packaging instantiation object according to the item packaging instantiation object data and the user's interest interaction behavior feature data, and generating user interest item association data;

[0083] In the embodiment of the present invention, the data of the instantiation object of the lot package is extracted from the database, and these data include the attribute characteristics, packaging structure and display information of the lot. At the same time, the user interest interaction behavior feature data is extracted to record the user's interaction and interest intensity for various types of lots. Through the data association algorithm, the user interest interaction behavior characteristics are matched with the lot package instantiation object to generate the association data between the user and the lot. This process can match the features of the lot with the user's historical browsing records, collection behavior, click frequency and other factors to associate which lots are highly related to the user's interests. The generated user interest lot association data contains information such as the matching degree and association intensity of the user's interests and the lot, which serves as an important basis for subsequent recommendations and push.

[0084] Step S42: Design a bipartite graph matching user interests and auction items through the user-interest auction item association data to generate a user-interest auction item bipartite graph;

[0085] In the embodiment of the present invention, users and auction items are regarded as two different node sets in a bipartite graph. Based on the user interest auction item association data, an association edge is established between the user and the auction item node, and the weight of the edge represents the intensity of the user's interest in a certain auction item. The system finds the most relevant auction items for each user through a matching algorithm, and adjusts the edge weights in the bipartite graph based on the interaction frequency, browsing time, collection behavior, etc. between the user and the auction item. The generated user interest auction item bipartite graph structure intuitively displays the association relationship between users and auction items, efficiently identifies which users show strong interest in which auction items, and provides effective guidance for subsequent push and display.

[0086] Step S43: Perform dynamic prediction processing on the propagation of user-interested items based on the bipartite graph of user-interested items to generate dynamic prediction data on the propagation of user-interested items.

[0087] In the embodiment of the present invention, a long short-term memory (LSTM) neural network model is used to analyze the bipartite graph of user-interested items to predict the changes in user interests in the future. Through the bipartite graph relationship between users and items, and combined with historical interaction behavior data, it is predicted whether users will have a sustained interest in a certain type of items. By capturing the changing trend of user behavior, the propagation path of items in the user group and the potential interest transfer are analyzed. The generated user-interested item propagation dynamic prediction data includes items that users may be interested in in the future, interest intensity predictions, and the propagation trends of items in different user groups.

[0088] Preferably, step S43 includes the following steps:

[0089] Step S431: Performing item node similarity feature analysis on the user's interest item bipartite graph to generate item node similarity feature data; performing item node similarity measurement on the item node similarity feature data to generate item node similarity data;

[0090] Step S432: performing a user-interest-item adjacency matrix analysis on the user-interest-item bipartite graph based on the item node similarity data to generate a user-interest-item adjacency matrix;

[0091] Step S433: performing multi-dimensional interest propagation prediction processing on the user's interest item adjacency matrix based on the long short-term memory neural network algorithm to generate user's interest item propagation prediction data;

[0092] Step S434: dynamically update the user's interest item propagation prediction data to generate user's interest item propagation dynamic prediction data.

[0093] The present invention performs similar feature analysis processing on the item nodes in the user interest item bipartite graph, identifies the similarity between items in terms of attributes, user interests, etc., and generates item node similar feature data. These feature data reflect the similarity of items in multiple dimensions, such as appearance, category, historical user feedback, etc. Subsequently, by performing similarity measurement processing on the item node similar feature data, the similarity between items is quantified, and the item node similarity data is generated, which provides an accurate similarity basis for the subsequent recommendation algorithm, helps the system to make more accurate judgments based on similarity when pushing items, and improves the user's interest matching. The adjacency matrix of the user's interest items is constructed based on the item node similarity data, which can intuitively represent the similarity relationship between items and the matching degree with the user's interest. The adjacency matrix can not only clearly show the user's interest intensity in different items, but also help the system analyze the potential propagation path between items. Through the analysis of the user interest item adjacency matrix, it is possible to more efficiently find out those items that may cause a surge in user interest, provide a more accurate basis for subsequent recommendation and push, and improve the system's analysis ability in complex scenarios. By applying the long short-term memory (LSTM) neural network algorithm to perform multi-dimensional interest propagation prediction processing on the adjacency matrix of user-interested items, the dynamic trend of user interest changes is captured, and the possible future user interest trends are predicted. The LSTM algorithm can effectively process time series data and identify short-term fluctuations and long-term trends in user interest in items. The system's prediction ability in the time dimension is enhanced, so that item recommendations are not only based on current interests, but also on the user's future potential needs, thereby further optimizing the item push strategy and improving the foresight and accuracy of the recommendation. The user's interest item propagation prediction data is dynamically updated, and the prediction results are adjusted according to the user's real-time behavior changes to ensure that the recommendation and push strategies can timely reflect the latest user interest changes. This dynamic update process improves the system's sensitivity to changes in user needs and avoids the lag problem of traditional systems in pushing content. Through continuous dynamic updates, the accuracy and timeliness of item push are guaranteed, further improving user participation and item transaction rates.

[0094] In an embodiment of the present invention, by analyzing the item nodes in the user's interest item bipartite graph, the multidimensional features of each item are extracted, including visual features and other attributes related to the item display. Similarity feature analysis extracts similarity features between items through feature extraction algorithms, such as principal component analysis (PCA) or convolutional neural network (CNN). The generated similar feature data of the item nodes are used to further calculate the similarity between the items. In the similarity measurement process, the similarity between different item nodes is measured based on the similar feature data of the items using cosine similarity, Euclidean distance or a neural network-based embedding model. The measurement result is represented by assigning a similarity score, and the generated item node similarity data is used for subsequent adjacency matrix analysis and interest propagation prediction. Based on the item node similarity data, an adjacency matrix of the user's interest item bipartite graph is constructed. Each element of the adjacency matrix represents the connection relationship between the item nodes and their similarity strength. The size of the matrix depends on the number of items. Through adjacency matrix analysis, the distribution of users' similar interests in items is displayed, and the potential correlation between different items is revealed. Not only can the relationship between lots be displayed through similarity calculation, but also those lot combinations that may cause users to further spread their interest can be identified. The generated user interest lot adjacency matrix provides structured data support for the next step of interest propagation prediction. The LSTM neural network is used to perform time series analysis on the user interest lot adjacency matrix to predict the dynamic changes of user interests. LSTM is good at processing time-dependent sequence data, so it can effectively capture the propagation path and change trend of user interests in different periods of time. In the multi-dimensional interest propagation prediction processing, multiple dimensions of features are considered, such as the user's historical interest interaction behavior, lot similarity, and the frequency of interaction between users and lots. Through the LSTM model, the system generates the future propagation trend of user interest lots, predicts which lots will continue to maintain high user interest, and which lots may attract new attention. The generated user interest lot propagation prediction data includes the user's future interest score for different lots, potential interest characteristics, etc.

[0095] Preferably, step S6 comprises the following steps:

[0096] Step S61: analyzing the timing behavior characteristics of the auction item creation demonstration according to the auction item creation demonstration key video frame data to generate the timing behavior characteristic data of the auction item creation demonstration;

[0097] Step S62: Performing time-series attenuation weighting processing on the user interest interaction behavior feature data to generate weighted user interest interaction behavior feature data, and using the weighted user interest interaction behavior feature data as the auction item creation demonstration trajectory user interest score of the auction item creation demonstration time-series behavior feature data;

[0098] Step S63: performing a probability distribution analysis of user interest in the item creation demonstration behavior on the item creation demonstration temporal behavior feature data based on the user interest score of the item creation demonstration trajectory, and generating user interest probability distribution data of the item creation demonstration behavior;

[0099] Step S64: Based on the variational autoencoder algorithm, the user interest probability distribution data of the auction creation demonstration behavior is processed by user interest probability learning to build a personalized auction creation demonstration interest prediction model.

[0100] The present invention performs temporal behavior feature analysis on the key video frame data of the creation and demonstration of the auction item, captures the temporal features of the auction item during the display process, and generates the temporal behavior feature data of the creation and demonstration of the auction item. It reflects the user's attention and reaction to the auction item at different display stages, and identifies the key nodes in the auction item display process. The temporal behavior feature data provides a basis for understanding the gradual change of the user's interest in the auction item, thereby improving the system's ability to capture the evolution of user interests. The user interest interaction behavior feature data is subjected to temporal attenuation weighting processing, and its weight is adjusted according to the timeliness of the user behavior to generate weighted user interest interaction behavior feature data. It more reasonably reflects the decreasing trend of user interest over time, and avoids the excessive impact of old interaction behaviors on current recommendations. The user interest score generated based on the weighted data helps the system to more accurately evaluate the user's real-time interest in the auction item, making the push and recommendation more timely and effective. The user interest score based on the auction item creation demonstration trajectory can perform user interest probability distribution analysis on the auction item creation demonstration temporal behavior feature data. Through this probability distribution analysis, the user's interest in the auction item display at different times is quantified, and the user's preference for different auction item attributes is discovered. It provides a quantitative basis for the system to build personalized recommendations, improves the accuracy of recommendations and user satisfaction. The variational autoencoder (VAE) algorithm is used to learn potential behaviors of the probability distribution data of user interests in the creation and demonstration of auction items, which can automatically identify the potential patterns and behaviors of the creation and demonstration of auction items in user interests. This deep learning process can capture the user's imperceptible preferences, thereby building a personalized prediction model for the creation and demonstration of auction items. The model has efficient prediction capabilities, can continuously optimize the auction item recommendation effect, and further improve the intelligence level and user experience of the auction platform.

[0101] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 FIG. 5 is a flow chart of detailed implementation steps of step S6 in FIG. 5 . In this embodiment, step S6 includes:

[0102] Step S61: analyzing the timing behavior characteristics of the auction item creation demonstration according to the auction item creation demonstration key video frame data to generate the timing behavior characteristic data of the auction item creation demonstration;

[0103] In the embodiment of the present invention, the temporal behavior characteristics in the process of creating and demonstrating the auction items are analyzed by extracting key video frame data from the video stream, and a temporal analysis algorithm, such as a long short-term memory (LSTM) neural network or a hidden Markov model (HMM), is used to track the temporal changes in the display of the auction items, and the user's focus is recorded according to the display order, angle, and detail changes. The temporal behavior characteristics reflect which time periods or segments in the auction item display process attract more user attention, and the generated auction item creation and demonstration temporal behavior characteristic data can be used to analyze user interaction and interest changes.

[0104] Step S62: Performing time-series attenuation weighting processing on the user interest interaction behavior feature data to generate weighted user interest interaction behavior feature data, and using the weighted user interest interaction behavior feature data as the auction item creation demonstration trajectory user interest score of the auction item creation demonstration time-series behavior feature data;

[0105] In the embodiment of the present invention, when analyzing the characteristics of user interest interaction behavior, a time-series decay weighting process is adopted, which takes into account the gradual decay of the user's interaction behavior over time, that is, the influence of the user's earlier interaction behavior on the current interest gradually decreases. Different weights are assigned to different user interaction behaviors through a time-series weighting model (such as an exponential decay function), and weighted user interest interaction behavior characteristic data is generated. The weighted user interest interaction behavior characteristic data is associated with the auction creation demonstration time-series behavior characteristic data, and the interest intensity of each user during the auction demonstration process is calculated to generate the auction creation demonstration trajectory user interest score. The interest score reflects the user's interest changes in different time segments during the auction display process, and provides a quantitative basis for the subsequent interest distribution analysis.

[0106] Step S63: performing a probability distribution analysis of user interest in the item creation demonstration behavior on the item creation demonstration temporal behavior feature data based on the user interest score of the item creation demonstration trajectory, and generating user interest probability distribution data of the item creation demonstration behavior;

[0107] In the embodiment of the present invention, according to each user's interest score and temporal behavior characteristics, a probability distribution model is used to perform distribution analysis on user interests. The changes in user interests during the auction item display process are analyzed through a Gaussian mixture model (GMM) or a Bayesian probability model, and probability distribution data of user interests are generated. The probability distribution can characterize the strength of user interests at different display moments and predict the user's interest trend in subsequent display segments. The generated probability distribution data of user interests in auction item creation and demonstration behavior provides important input for the construction of a personalized interest prediction model.

[0108] Step S64: Based on the variational autoencoder algorithm, the user interest probability distribution data of the auction creation demonstration behavior is processed by user interest probability learning to build a personalized auction creation demonstration interest prediction model.

[0109] In the embodiment of the present invention, a variational autoencoder (VAE) is used to learn the generated user interest probability distribution data to identify the user's potential interest behavior in the auction display. The variational autoencoder compresses and reconstructs the user's interest distribution features, discovers the potential pattern of user interest, and extracts the user's implicit interest bias in the auction display through unsupervised learning. Based on these potential interest behaviors, a personalized auction creation demonstration interest prediction model is constructed. This model can effectively predict the changes in each user's interest in future auction display, and make personalized recommendations and push based on the behavior and preferences of individual users. This model is continuously optimized and dynamically updated with the input of new user interaction behavior data to ensure the accuracy and real-time nature of interest prediction.

[0110] Preferably, step S64 includes the following steps:

[0111] Step S641: performing auction item creation demonstration potential behavior analysis on the auction item creation demonstration behavior user interest probability distribution data based on the variational autoencoder to generate auction item creation demonstration potential behavior data;

[0112] Step S642: performing an interest probability distribution analysis of potential behavior of auction item creation demonstration according to the potential behavior data of auction item creation demonstration, and generating the interest probability distribution data of potential behavior of auction item creation demonstration;

[0113] Step S643: Based on the deep neural network algorithm, the user interest probability learning process of the potential behavior interest in the auction creation demonstration is performed on the potential behavior interest probability distribution of the auction creation demonstration to construct a personalized auction creation demonstration interest prediction model.

[0114] The present invention performs potential behavior analysis on the probability distribution data of user interest in the auction creation demonstration behavior based on variational autoencoder (VAE), identifies the potential behavior patterns behind user interest, and generates potential behavior data of auction creation demonstration. The user's interest preferences and behavioral tendencies that are not explicit during the auction display process are revealed. The potential behavior data of auction creation demonstration provides rich implicit information for subsequent user interest prediction, and improves the depth and accuracy of the recommendation system. The potential behavior interest probability distribution analysis is performed based on the potential behavior data of auction creation demonstration, and the potential interest of the user when watching the auction display is further quantified and modeled. By generating the probability distribution data of potential behavior interest in auction creation demonstration, the user's implicit interest is analyzed more carefully, so as to dig out the auction creation display features that are ignored but have strong appeal to users. The user interest probability learning processing is performed on the probability distribution data of potential behavior interest in auction creation demonstration by using a deep neural network algorithm, and a more accurate personalized auction creation demonstration interest prediction model is constructed. The deep neural network can capture complex nonlinear relationships, help the system better predict the future trend of user interest changes, continuously optimize the deeper recommendation effect, and improve the accuracy of auction recommendation and user participation.

[0115] In an embodiment of the present invention, the user interest probability distribution data of the auction creation demonstration behavior is analyzed by using a variational autoencoder (VAE) to identify the user's potential behavior pattern for the auction display. The variational autoencoder can extract implicit behavior features from complex user behavior data, and discover the user's hidden interest preferences when viewing the auction by compressing the input data. The user's potential interest changes at different time nodes and display details are analyzed by an unsupervised learning method, thereby generating potential behavior data of the auction creation demonstration, and this potential behavior data reveals the user's preferences that are not directly expressed. Based on the potential behavior data of the auction creation demonstration, the association between the auction creation demonstration potential behavior data and the auction creation demonstration behavior in the auction creation demonstration behavior user interest probability distribution data is analyzed, based on the association relationship, and the user interest probability distribution in the auction creation demonstration behavior user interest probability distribution data is used to preset the probability of potential behavior interest, such as by analyzing that the potential behavior is a potential mapping of the aggregation of several auction demonstration behaviors, analyzing the weight information of the auction demonstration behavior, and mapping the corresponding user interest probability distribution, to generate the probability distribution data of potential behavior interest. Through this process, the system can quantify the potential interest intensity of users in auction items at different display stages and establish a statistical relationship between user behavior and auction item characteristics. It reveals which auction item characteristics users show potential but not obvious interest in, thereby providing more accurate data basis for personalized recommendations. The generated auction item creation demonstration potential behavior interest probability distribution data provides input for the next deep learning model. The deep neural network (DNN) algorithm is used to process the user interest probability learning of the auction item creation display potential behavior on the potential behavior interest probability distribution data. The deep neural network can capture the complex nonlinear relationship between user behavior and interest preference, and model the user's interest trend through a multi-layer neural network structure. By training a large amount of user behavior data, the user's interest change pattern in different auction item creation displays is learned, and a personalized interest prediction model is generated. This model can predict which auction items users may show higher interest in in future auction item displays based on the user's historical behavior and potential interests, thereby providing a basis for personalized push. The construction of this model can not only predict the user's current interest preference through continuous optimization and learning, but also adapt to the dynamic changes of user interests through the learning of behavioral characteristics, ensuring the accuracy and real-time nature of recommendations.

[0116] Preferably, step S7 comprises the following steps:

[0117] Step S71: performing real-time multi-window push processing of the content of the auction items of interest to generate real-time multi-window push data of the content of the auction items of interest according to the decision data of pushing the auction items of interest to the user;

[0118] Step S72: transmitting the real-time interest auction item content multi-window push data to the personalized auction item creation demonstration interest prediction model for real-time push content interest prediction processing to generate real-time push content interest prediction data;

[0119] Step S73: performing real-time auction item push content intelligent layout analysis on the real-time interest auction item content multi-window push data according to the real-time push content interest prediction data, and generating real-time auction item push content intelligent layout data;

[0120] Step S74: Execute intelligent operation and management of commodity auction push through real-time auction item push content intelligent typesetting data.

[0121] The present invention performs multi-window push processing of real-time interest auction content according to user interest auction push decision data, and can push diversified auction display content in different windows according to the real-time interest dynamics of the user. Multi-window push allows users to pay attention to multiple auctions at the same time, effectively increases the user's interaction opportunities with more auctions, improves the coverage of push content, and enhances the user's participation and interactive experience, thereby improving the exposure rate and transaction rate of the auction. The real-time interest auction content multi-window push data is transmitted to the personalized auction creation demonstration interest prediction model, and the real-time push content can be processed for interest prediction, and real-time push content interest prediction data can be generated. By predicting the user's potential interest in different push content, the auction that best meets the user's current interest is screened out, thereby further optimizing the push content, improving the accuracy of the recommendation, avoiding pushing irrelevant or repeated auctions to the user, and improving the user experience. According to the interest prediction data of the real-time push content, the multi-window push data is intelligently typed and analyzed to generate real-time auction push content intelligent typesetting data. Through intelligent typesetting analysis, the layout of the push content is optimized according to the strength of the user's interest and the display effect, ensuring that the user sees the auction that best meets the interest at the first time. Intelligent layout not only improves the visual appeal of pushed content, but also increases the probability of user interaction with auction items, thereby further improving the recommendation effect. Through the intelligent layout data of real-time auction item push content, the intelligent operation of commodity auction push is performed to achieve automated and efficient auction item push operation, effectively reduce manual intervention, and improve the overall efficiency of the auction process. Intelligent operation ensures the timeliness and accuracy of push content, thereby optimizing the user experience and the operation effect of the auction platform, promoting auction item transactions and user retention rate.

[0122] In the embodiment of the present invention, decision data is pushed according to user interests, and auction items that are highly relevant to the user's interests are selected. Then, the system uses a multi-window push mechanism to simultaneously present multiple auction items in the user's interface for user selection and interaction. Multi-window push can simultaneously display different details, characteristics and information of multiple auction items. This multi-dimensional display method enhances the user's comprehensive understanding of the auction items and the flexibility of selection. During the push process, the system adjusts the push content according to the user's current interests and behaviors to ensure that the pushed auction items are highly consistent with the user's interests. The generated real-time interest auction item content multi-window push data includes the display order, display content and real-time feedback of the auction items in each push window. The multi-window push data is transmitted to the personalized auction item creation demonstration interest prediction model. The model predicts the user's interest change trend in the current push content based on the user's real-time interaction data on the multi-window content, such as clicks, browsing time, scrolling behavior, etc. By analyzing the user's behavior data in real time, the system can identify which auction items the user shows a higher interest in, thereby generating more accurate real-time push content interest prediction data. These data will help the system continuously optimize the push content so that subsequent pushes are more in line with the user's personalized needs. Based on the real-time push content interest prediction data, the auction display content in the multi-window push data is intelligently layout analyzed. The intelligent layout algorithm intelligently sorts and layouts the auction items in different windows according to factors such as the user's interest intensity, the priority of the push window, and the visual appeal of the auction items. The auction items that users are most interested in will be displayed in a prominent position first, ensuring that users can find content that meets their needs more quickly. The real-time auction push content intelligent layout data generated in the end will be used to increase the attention of the auction items and user experience, and improve the push effect. According to the real-time auction push content intelligent layout data, the intelligent operation of commodity auction push is automatically executed. The core of the intelligent operation is to dynamically adjust the auction push strategy based on the user's real-time interest and personalized needs. Continuously monitor user behavior and feedback, optimize the push timing, display method and frequency of the auction items, so as to ensure that users can receive the auction item recommendations that best suit their interests in a timely manner. Through this automated and intelligent push operation, not only the user experience is improved, but also the exposure rate and transaction rate of the auction items are improved, further optimizing the overall operation efficiency of the platform.

[0123] This specification provides an artificial intelligence-based commodity auction intelligent operation and management system, which is used to execute the artificial intelligence-based commodity auction intelligent operation and management method described above. The artificial intelligence-based commodity auction intelligent operation and management system includes:

[0124] A historical auction item creation demonstration multi-source feature collection module is used to collect multi-source features of historical auction item creation demonstration based on a commodity auction streaming media platform, and generate multi-source feature data of historical auction item creation demonstration, wherein the multi-source feature data of historical auction item creation demonstration includes auction item creation demonstration video stream data and user interaction behavior data;

[0125] The auction item visual attribute analysis module is used to perform visual auction item attribute feature analysis based on the auction item creation demonstration video stream data to generate auction item attribute feature data; perform auction item package instantiation object design based on the auction item attribute feature data to generate auction item package instantiation object data;

[0126] A user interest interaction behavior feature analysis module, used to analyze user interest interaction behavior features based on user interaction behavior data and generate user interest interaction behavior feature data;

[0127] A user interest item propagation dynamic prediction module is used to design a bipartite graph matching user interests and items based on item encapsulation instantiation object data and user interest interaction behavior feature data, and generate a user interest item bipartite graph; based on the user interest item bipartite graph, a user interest item propagation dynamic prediction process is performed to generate user interest item propagation dynamic prediction data;

[0128] A user interest item push decision analysis module, which is used to perform user interest item push decision analysis based on user interest item propagation dynamic prediction data, and generate user interest item push decision data;

[0129] The module for constructing a model for predicting the interest in auction item creation demonstration is used to analyze the probability distribution of user interest in auction item creation demonstration behavior based on the auction item creation demonstration video stream data and the user interest interaction behavior feature data, and generate the probability distribution data of user interest in auction item creation demonstration behavior; and construct a personalized auction item creation demonstration interest prediction model based on the probability distribution data of user interest in auction item creation demonstration behavior;

[0130] The intelligent operation job execution module is used to perform real-time interest auction item content multi-window push processing based on the user's interest auction item push decision data, and generate real-time interest auction item content multi-window push data; based on the personalized auction item creation demonstration interest prediction model, the real-time interest auction item content multi-window push data is subjected to real-time auction item push content intelligent layout analysis, and real-time auction item push content intelligent layout data is generated; and the commodity auction push intelligent operation management job is executed through the real-time auction item push content intelligent layout data.

[0131] The beneficial effect of the present application is that the present invention automatically analyzes and instantiates the visual attributes of the auction items into standardized object data based on the visual feature extraction and packaging of the auction item creation demonstration video stream data, and can intelligently match the auction items with users by dynamically analyzing the relationship between the auction item attributes and user interests, thereby effectively improving the efficiency of recommendation. By intelligently packaging the attribute characteristics of the auction items, an operable instantiated object is generated for each auction item, thereby achieving efficient matching of the auction items with user needs. This automated and intelligent processing method improves the accuracy of the system in processing the auction item characteristics and user interests, and optimizes the push effect of the auction items. By collecting and analyzing the user interaction behavior data, the user interest interaction behavior feature data is generated. At the same time, based on the deep learning algorithm such as the long short-term memory neural network (LSTM), the dynamic propagation prediction of user interests can be carried out, which can capture the changes in user interests in the online auction process in real time, effectively solving the problem that the existing technology cannot dynamically track and analyze the changes in user interests, and provides a comprehensive and accurate capture and prediction capability of user interests, which can perceive the fluctuation of user interests in real time and make corresponding adjustments, improve the accuracy of auction item recommendations, and enhance the user experience and the interactivity of the platform. By generating a bipartite graph of user-interested auction items and combining it with real-time user interest prediction, the push content can be dynamically adjusted according to the user's current needs, and the intelligent layout technology is designed to optimize the display form of the auction items by analyzing the auction item creation and display mode that the user likes, ensuring that users can browse auction items that meet their interests and prioritize display based on real-time auction item creation and display content. The display and push methods of auction items can flexibly adapt to the user's real-time interests, ensure that the push content is highly consistent with user needs, avoid interference from irrelevant content, and improve the accuracy and display effect of push, thereby increasing the exposure rate and transaction rate of auction items.

[0132] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0133] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. An intelligent operation and management method for commodity auction based on artificial intelligence, characterized in that: The following steps are involved: Step S1: collecting multi-source features of historical auction item creation demonstration based on the commodity auction streaming media platform to generate multi-source feature data of historical auction item creation demonstration, wherein the multi-source feature data of historical auction item creation demonstration includes auction item creation demonstration video stream data and user interaction behavior data; Step S2: Perform visual auction item attribute feature analysis based on the auction item creation demonstration video stream data to generate auction item attribute feature data; Design the instantiation object of the lot packaging according to the property feature data of the lot, and generate the instantiation object data of the lot packaging; Step S3: Analyze the user interest interaction behavior characteristics according to the user interaction behavior data to generate user interest interaction behavior characteristic data; Step S4: Design a matching bipartite graph between user interests and auction items based on the auction item encapsulation instantiation object data and the user interest interaction behavior feature data to generate a user interest auction item bipartite graph; perform dynamic prediction processing on the user interest auction item propagation based on the user interest auction item bipartite graph to generate dynamic prediction data on the user interest auction item propagation; Step S5: Perform decision analysis on pushing user-interested items based on the dynamic prediction data of user-interested items, and generate decision data on pushing user-interested items; Step S6: Based on the auction item creation demonstration video stream data and the user interest interaction behavior feature data, a probability distribution analysis of the user interest in the auction item creation demonstration behavior is performed to generate the user interest probability distribution data of the auction item creation demonstration behavior; and a personalized auction item creation demonstration interest prediction model is constructed according to the user interest probability distribution data of the auction item creation demonstration behavior; Wherein, step S6 comprises the following steps: Step S61: analyzing the timing behavior characteristics of the auction item creation demonstration according to the auction item creation demonstration key video frame data to generate the timing behavior characteristic data of the auction item creation demonstration; Step S62: Performing time-series attenuation weighting processing on the user interest interaction behavior feature data to generate weighted user interest interaction behavior feature data, and using the weighted user interest interaction behavior feature data as the auction item creation demonstration trajectory user interest score of the auction item creation demonstration time-series behavior feature data; Step S63: performing a probability distribution analysis of user interest in the item creation demonstration behavior on the item creation demonstration temporal behavior feature data based on the user interest score of the item creation demonstration trajectory, and generating user interest probability distribution data of the item creation demonstration behavior; Step S64: performing user interest probability learning processing of the potential behavior of auction item creation demonstration on the user interest probability distribution data of the auction item creation demonstration behavior based on the variational autoencoder algorithm to construct a personalized auction item creation demonstration interest prediction model; Wherein, step S64 includes the following steps: Step S641: performing auction item creation demonstration potential behavior analysis on the auction item creation demonstration behavior user interest probability distribution data based on the variational autoencoder to generate auction item creation demonstration potential behavior data; Step S642: performing an interest probability distribution analysis of potential behavior of auction item creation demonstration according to the potential behavior data of auction item creation demonstration, and generating the interest probability distribution data of potential behavior of auction item creation demonstration; Step S643: performing user interest probability learning processing on the potential behavior interest probability distribution of the auction item creation demonstration based on the deep neural network algorithm to construct a personalized auction item creation demonstration interest prediction model; Step S7: Perform real-time interest item content multi-window push processing based on the user's interest item push decision data to generate real-time interest item content multi-window push data; perform real-time interest item push content intelligent layout analysis on the real-time interest item content multi-window push data based on the personalized item creation demonstration interest prediction model to generate real-time item push content intelligent layout data; perform commodity auction push intelligent operation management operations through real-time item push content intelligent layout data.

2. The intelligent operation and management method for commodity auction based on artificial intelligence according to claim 1 is characterized in that: Step S1 includes the following steps: Step S11: monitoring and processing the multi-source data of the creation and demonstration of auction items according to the commodity auction streaming media platform to generate the multi-source data of the creation and demonstration of auction items; Step S12: performing data synchronization processing on the auction item creation demonstration multi-source data to generate synchronized auction item creation demonstration multi-source data; Step S13: extract historical multi-source features based on the synchronized auction item creation demonstration multi-source data to generate historical auction item creation demonstration multi-source feature data.

3. The intelligent operation and management method for commodity auction based on artificial intelligence according to claim 1 is characterized in that: Step S2 includes the following steps: Step S21: performing video frame decoding processing according to the auction creation demonstration video stream data to generate auction creation demonstration video frame data; Step S22: extract key video frames from the auction item creation demonstration video frame data to generate auction item creation demonstration key video frame data; Step S23: Perform visual auction item attribute feature analysis and processing based on auction item creation demonstration key video frame data to generate auction item attribute feature data; Step S24: performing data encapsulation processing on the auction item attribute feature data to generate encapsulated auction item attribute feature data; Step S25: Designing an auction item packaging instantiation object according to the packaged auction item attribute feature data, and generating auction item packaging instantiation object data.

4. The intelligent operation and management method for commodity auctions based on artificial intelligence according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: extracting text interaction behavior according to the user interaction behavior data to generate text interaction behavior data; Step S32: performing text semantic analysis on the text interaction behavior data based on natural language technology to generate semantic analysis interaction behavior data; Step S33: extract key interest text features from the semantic parsing interaction behavior data to generate key interest text feature data; Step S34: extracting non-text interaction behaviors according to the user interaction behavior data to generate non-text interaction behavior data; Step S35: performing non-text interaction behavior frequency feature analysis based on the non-text interaction behavior data to generate non-text interaction behavior frequency feature data; Step S36: Perform user interest interaction behavior feature analysis based on the key interest text feature data and the non-text interaction behavior frequency feature data to generate user interest interaction behavior feature data.

5. The intelligent operation and management method for commodity auction based on artificial intelligence according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: performing data association processing between the user's interest interaction behavior and the item packaging instantiation object according to the item packaging instantiation object data and the user's interest interaction behavior feature data, and generating user interest item association data; Step S42: Design a bipartite graph matching user interests and auction items through the user-interest auction item association data to generate a user-interest auction item bipartite graph; Step S43: Perform dynamic prediction processing on the propagation of user-interested items based on the bipartite graph of user-interested items to generate dynamic prediction data on the propagation of user-interested items.

6. The intelligent operation and management method for commodity auction based on artificial intelligence according to claim 5 is characterized in that: Step S43 includes the following steps: Step S431: Performing item node similarity feature analysis on the user's interest item bipartite graph to generate item node similarity feature data; performing item node similarity measurement on the item node similarity feature data to generate item node similarity data; Step S432: performing a user-interest-item adjacency matrix analysis on the user-interest-item bipartite graph based on the item node similarity data to generate a user-interest-item adjacency matrix; Step S433: performing multi-dimensional interest propagation prediction processing on the user's interest item adjacency matrix based on the long short-term memory neural network algorithm to generate user's interest item propagation prediction data; Step S434: dynamically update the user's interest item propagation prediction data to generate user's interest item propagation dynamic prediction data.

7. The intelligent operation and management method for commodity auction based on artificial intelligence according to claim 1 is characterized in that: Step S7 includes the following steps: Step S71: performing real-time multi-window push processing of the content of the auction items of interest to generate real-time multi-window push data of the content of the auction items of interest according to the decision data of pushing the auction items of interest to the user; Step S72: transmitting the real-time interest auction item content multi-window push data to the personalized auction item creation demonstration interest prediction model for real-time push content interest prediction processing to generate real-time push content interest prediction data; Step S73: performing real-time auction item push content intelligent layout analysis on the real-time interest auction item content multi-window push data according to the real-time push content interest prediction data, and generating real-time auction item push content intelligent layout data; Step S74: Execute intelligent operation and management of commodity auction push through real-time auction item push content intelligent typesetting data.

8. An intelligent operation and management system for commodity auctions based on artificial intelligence, characterized in that: Used to execute the commodity auction intelligent operation management method based on artificial intelligence as claimed in claim 1, the commodity auction intelligent operation management system based on artificial intelligence includes: A historical auction item creation demonstration multi-source feature collection module is used to collect multi-source features of historical auction item creation demonstration based on a commodity auction streaming media platform, and generate multi-source feature data of historical auction item creation demonstration, wherein the multi-source feature data of historical auction item creation demonstration includes auction item creation demonstration video stream data and user interaction behavior data; The auction item visual attribute analysis module is used to perform visual auction item attribute feature analysis based on the auction item creation demonstration video stream data to generate auction item attribute feature data; perform auction item package instantiation object design based on the auction item attribute feature data to generate auction item package instantiation object data; A user interest interaction behavior feature analysis module, used to analyze user interest interaction behavior features based on user interaction behavior data and generate user interest interaction behavior feature data; A user interest item propagation dynamic prediction module is used to design a bipartite graph matching user interests and items based on item encapsulation instantiation object data and user interest interaction behavior feature data, and generate a user interest item bipartite graph; based on the user interest item bipartite graph, a user interest item propagation dynamic prediction process is performed to generate user interest item propagation dynamic prediction data; A user interest item push decision analysis module, which is used to perform user interest item push decision analysis based on user interest item propagation dynamic prediction data, and generate user interest item push decision data; The module for constructing a model for predicting the interest in auction item creation demonstration is used to analyze the probability distribution of user interest in auction item creation demonstration behavior based on the auction item creation demonstration video stream data and the user interest interaction behavior feature data, and generate the probability distribution data of user interest in auction item creation demonstration behavior; and construct a personalized auction item creation demonstration interest prediction model based on the probability distribution data of user interest in auction item creation demonstration behavior; The intelligent operation job execution module is used to perform real-time interest auction item content multi-window push processing based on the user's interest auction item push decision data, and generate real-time interest auction item content multi-window push data; based on the personalized auction item creation demonstration interest prediction model, the real-time interest auction item content multi-window push data is subjected to real-time auction item push content intelligent layout analysis, and real-time auction item push content intelligent layout data is generated; and the commodity auction push intelligent operation management job is executed through the real-time auction item push content intelligent layout data.

Images