An intelligent demand forecasting and analysis method and system for commodity auctions

Abstract

The present invention relates to the field of commercial data analysis technology, and in particular to an intelligent demand prediction and analysis method and system for commodity auctions. The method comprises the following steps: acquiring auction item data; analyzing auction item feature data according to the auction item data; performing dynamic association network analysis of the auction item based on the auction item feature data to generate a dynamic auction item association network; detecting user interaction data; performing user interaction behavior feature analysis and processing according to the user interaction data to generate user interaction behavior feature data; performing instant user demand auction item probability prediction based on the user interaction behavior feature data to generate instant user demand auction item probability prediction data; transmitting the instant user demand auction item probability prediction data to the dynamic auction item association network to perform instant user associated demand auction item intelligent analysis to generate instant user associated demand auction item prediction data. The present invention achieves the improvement of the accuracy and efficiency of demand prediction in the commodity auction platform.

Description

Technical Field

[0001] The present invention relates to the technical field of commercial data analysis, and in particular to an intelligent demand forecasting and analysis method and system for commodity auctions. Background Art

[0002] With the development of the Internet and the popularization of e-commerce, online commodity auction platforms have gradually become an important way of commodity trading. As a bidding purchase method, auctions have attracted a large number of users to participate. With the increase in the number of users and the types of auctioned goods, how to accurately predict user demand has become a key issue. Demand forecasting can not only help auction platforms optimize auction strategies and improve user experience, but also provide data support for platforms and sellers to promote accurate recommendation and sales of goods. However, the existing commodity demand analysis methods cannot accurately express the commodity information, and they also need to manually label the commodities or classify simple type information. In addition, the analysis of user demand is based on historical data or user browsing behavior analysis, which makes it impossible to accurately capture the real-time changes in user demand for auction items. Summary of the invention

[0003] Based on this, the present invention provides an intelligent demand forecasting and analysis method and system for commodity auctions to solve at least one of the above technical problems.

[0004] To achieve the above purpose, an intelligent demand forecasting and analysis method for commodity auctions includes the following steps:

[0005] Step S1: Acquire the auction item ontology data of the online commodity auction platform; perform auction item ontology feature analysis based on the auction item ontology data to generate auction item ontology feature data;

[0006] Step S2: performing a screening process on the item body gain feature according to the item body feature data to generate the item body gain feature data;

[0007] Step S3: establishing an item ontology attribute matrix set through the item ontology gain feature data; performing a dynamic association network analysis of the item ontology based on the item ontology attribute matrix set to generate a dynamic item ontology association network;

[0008] Step S4: Detecting user interaction data in the online commodity auction platform; performing user interaction behavior feature analysis and processing based on the user interaction data to generate user interaction behavior feature data; performing interaction behavior feature analysis and processing of clustering user interaction behavior habit types based on the user interaction behavior feature data to generate clustered user interaction behavior feature data;

[0009] Step S5: construct an optimized probability tree model for classifying the demand items of clustered users based on the clustered user interaction behavior feature data to generate an optimized probability tree model for the demand items of clustered users; transmit the user interaction behavior feature data to the optimized probability tree model for the demand items of clustered users to predict the probability of the real-time user demand items and generate the real-time user demand item probability prediction data; transmit the real-time user demand item probability prediction data to the dynamic item ontology association network to perform intelligent analysis of the real-time user associated demand items and generate the real-time user associated demand item prediction data; and transmit the real-time user associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand items.

[0010] Further, step S1 includes the following steps:

[0011] Step S11: Acquire the auction item data of the online commodity auction platform;

[0012] Step S12: extracting the auction item body structured data and the auction item body unstructured data of the auction item body data;

[0013] Step S13: performing unstructured mapping feature analysis on the auction item based on the auction item body unstructured data to generate unstructured mapping feature data of the auction item;

[0014] Step S14: performing auction item ontology feature analysis based on the auction item ontology structured data and the auction item unstructured mapping feature data to generate auction item ontology feature data.

[0015] Further, step S13 includes the following steps:

[0016] Step S131: performing unstructured type classification processing on the auction item body unstructured data to generate classified auction item body unstructured data;

[0017] Step S132: performing convolutional layer analysis and processing of multimodal unstructured mapping features of auction items based on the classified auction item body unstructured data to generate a convolutional layer of multimodal mapping features of auction items;

[0018] Step S133: designing a multi-modal mapping feature convolutional neural network model based on the multi-modal mapping feature convolutional layer of the auction items;

[0019] Step S134: performing preliminary unstructured item mapping feature processing on the unstructured item data according to the multimodal item mapping feature convolutional neural network model to generate preliminary unstructured item mapping feature data; and performing feature full-connection mapping optimization processing on the preliminary unstructured item mapping feature data to generate unstructured item mapping feature data.

[0020] Further, step S2 includes the following steps:

[0021] Step S21: performing feature discretization processing on the auction item feature data to generate auction item feature discretization data;

[0022] Step S22: performing conditional entropy analysis of the item's main features based on the discretized data of the item's main features to generate conditional entropy data of the item's main features;

[0023] Step S23: performing auction item body gain feature screening processing on the auction item body feature data according to the auction item body feature condition entropy data to generate auction item body gain feature data.

[0024] Further, step S22 includes the following steps:

[0025] Step S221: performing frequency domain data analysis on the discretized data of the auction item's main features to generate frequency domain data of the auction item's main features;

[0026] Step S222: performing an entropy analysis on the item body feature label according to the frequency domain data of the item body feature to generate the item body feature label entropy data;

[0027] Step S223: performing auction item feature conditional entropy analysis on the auction item feature frequency domain data and the auction item feature label entropy data to generate auction item feature conditional entropy data.

[0028] Further, step S3 includes the following steps:

[0029] Step S31: designing and classifying item ontology attribute matrix nodes according to each gain feature type in the item ontology gain feature data, and transmitting the item ontology gain feature data to the item ontology attribute matrix nodes to establish an item ontology attribute matrix set;

[0030] Step S32: performing association aggregation analysis on the auction item ontology attribute matrix based on the auction item ontology attribute matrix set to generate an auction item matrix association aggregation coefficient;

[0031] Step S33: Designing an item ontology association network through the item ontology attribute matrix set, and dynamically updating the item ontology association degree of the item ontology association network through the matrix association aggregation coefficient as the item association attention weight parameter to generate a dynamic item ontology association network.

[0032] Further, step S32 includes the following steps:

[0033] Step S321: using matrix nodes of various types in the auction item ontology attribute matrix set as first auction item ontology attribute data, and randomly selecting matrix nodes of the remaining auction item ontology attribute matrix set in the type corresponding to the first auction item ontology attribute node data as second auction item ontology attribute node data;

[0034] Step S322: performing node similarity calculation on the first lot body attribute node data and the second lot body attribute node data to generate node similarity data;

[0035] Step S323: Calculate the node access overlap of the first lot body attribute node data and the second lot body attribute node data to generate node access overlap data;

[0036] Step S324: performing association aggregation analysis on the auction item ontology attribute matrix set according to the node similarity data and the node access overlap data, and generating an auction item matrix association aggregation coefficient.

[0037] Further, step S4 includes the following steps:

[0038] Step S41: Detecting user interaction data in the online commodity auction platform;

[0039] Step S42: performing user interaction behavior feature analysis and processing according to the user interaction data to generate user interaction behavior feature data;

[0040] Step S43: performing user interaction behavior habit cluster analysis based on the user interaction behavior feature data to generate user interaction behavior habit cluster data;

[0041] Step S44: performing data division processing on the user interaction behavior characteristic data into clustered user interaction behavior habit types according to the user interaction behavior habit clustering data to generate clustered user interaction behavior characteristic data.

[0042] Further, step S5 includes the following steps:

[0043] Step S51: performing clustered user demand auction item conversion feature analysis based on clustered user interaction behavior feature data to generate clustered user demand auction item conversion feature data;

[0044] Step S52: using clustered user interaction behavior feature data as clustered user demand item probability factors; establishing clustered user demand item classification probability mapping relationships for clustered user demand item probability factors based on a preset decision tree classification algorithm to obtain a clustered user demand item probability tree model; optimizing and adjusting the tree node probability coefficients of the clustered user demand item probability tree model through clustered user demand item conversion feature data to generate an optimized clustered user demand item probability tree model;

[0045] Step S53: transmitting the user interaction behavior feature data to the optimized clustering user demand item probability tree model to perform instant user demand item probability prediction, and generate instant user demand item probability prediction data;

[0046] Step S54: Transmit the probability prediction data of the real-time user demand item to the dynamic item ontology association network for intelligent analysis of the real-time user-associated demand item, generate the real-time user-associated demand item prediction data, and transmit the real-time user-associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand item.

[0047] This specification provides an intelligent demand forecasting and analysis system for commodity auctions, which is used to execute the intelligent demand forecasting and analysis method for commodity auctions as described above. The intelligent demand forecasting and analysis system for commodity auctions includes:

[0048] The item body feature analysis module is used to obtain the item body data of the online commodity auction platform; perform item body feature analysis based on the item body data to generate item body feature data;

[0049] A gain feature screening module is used to perform a screening process on the gain features of the auction item itself according to the auction item itself feature data to generate the auction item itself gain feature data;

[0050] A dynamic auction item ontology association module is used to establish an auction item ontology attribute matrix set through the auction item ontology gain feature data; perform a dynamic association network analysis of the auction item ontology based on the auction item ontology attribute matrix set to generate a dynamic auction item ontology association network;

[0051] The clustering user interaction behavior analysis module is used to detect user interaction data in the online commodity auction platform; perform user interaction behavior feature analysis and processing based on the user interaction data to generate user interaction behavior feature data; perform clustering user interaction behavior habit type interaction behavior feature analysis and processing based on the user interaction behavior feature data to generate clustered user interaction behavior feature data;

[0052] The instant user-associated demand item prediction module is used to construct an optimized demand item classification probability tree model for clustered users based on the clustered user interaction behavior feature data, and generate an optimized clustered user demand item probability tree model; transmit the user interaction behavior feature data to the optimized clustered user demand item probability tree model to perform instant user demand item probability prediction, and generate instant user demand item probability prediction data; transmit the instant user demand item probability prediction data to the dynamic item ontology association network to perform instant user-associated demand item intelligent analysis, generate instant user-associated demand item prediction data, and transmit the instant user-associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand item.

[0053] The beneficial effect of the present application is that, by acquiring and analyzing the auction item data in the commodity auction platform, combining the structured and unstructured features of the commodity, and using the multimodal unstructured mapping feature convolutional neural network model, the complex characteristics of the auction item are deeply analyzed. Traditional methods usually rely only on simple analysis of structured data, ignoring the rich information of the commodity ontology unstructured data. Through the convolutional layer neural network, the unstructured data is deeply processed, and the potential characteristics of the auction item can be captured more accurately, thereby improving the accurate prediction of commodity demand. At the same time, through multi-dimensional feature processing such as feature gain screening and conditional entropy analysis, redundant features are effectively eliminated, and the efficiency and accuracy of the prediction model are improved. By performing association analysis on the auction item ontology gain feature data and constructing an association network, the network is continuously updated dynamically based on the similarity between the commodity ontology features and the associated user access behavior, reflecting the degree of association between commodities and changes in user demand in real time, and can be continuously updated with changes in user demand and commodity features, to achieve dynamic management and analysis of the degree of association between commodities. By analyzing the interactive behavior characteristics of users in the auction platform in detail, and classifying the interactive behavior habits of different types of users through cluster analysis. On this basis, a clustering user demand item classification probability tree model is constructed to perform personalized demand forecasting based on the behavior patterns and demand characteristics of different user groups. This deep clustering analysis based on user behavior characteristics not only helps the platform better understand the behavior habits of different users and optimize the item recommendation strategy, but also improves the personalization and pertinence of the prediction results, thereby meeting the diverse needs of different users. The clustering user demand item classification probability tree model can quickly update the prediction results based on the user's instant interactive behavior, so that the demand forecast can be adjusted in real time as the user behavior changes, greatly improving the response speed and dynamic adaptability of the prediction analysis. The generated user demand forecast results are transmitted to the terminal through an intelligent prediction feedback mechanism to form a visual feedback operation. This instant feedback mechanism not only provides decision support for platform managers, but also provides users with a more intuitive display of demand forecast results, which helps to improve the user experience of auction participants and increase users' stickiness and satisfaction with the platform. In addition, the platform can optimize the recommendation and auction strategy of the items based on the demand forecast results, thereby improving transaction efficiency and success rate, and ultimately promoting the improvement of the overall benefits of the platform. Therefore, the intelligent demand prediction and analysis method for commodity auctions of the present invention can extract the key representation features of the auction item body, and screen the high-gain feature types in the auction item body to improve data processing efficiency while accurately expressing the auction item information, and establish a dynamic auction item body association network through the association characteristics in the high-gain features in the auction item body, thereby dynamically reflecting the association relationship between each commodity. When a user is interested in a certain auction item, intelligent recommendations are made to the user based on the dynamic association relationship of the auction items in the dynamic auction item body association network.In addition, the analysis of user needs is carried out by clustering user analysis on user interaction behaviors to reflect the interaction habits of users in different clusters, and then the probability of favorite items of users in different clusters is predicted through the decision tree algorithm to reflect the distribution information of different users for different favorite items to reflect the predicted probability of user item demand. Combined with the dynamic item ontology association network and the user item demand prediction probability, the intelligent analysis of the instant user-associated demand items is carried out to reflect the user's instant demand items and the corresponding quantitative demand information, so as to improve the accuracy and efficiency of demand prediction in the commodity auction platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A schematic diagram of the steps of an intelligent demand forecasting and analysis method for commodity auctions according to the present invention;

[0055] Figure 2 for Figure 1 Detailed implementation steps of step S3 in FIG.

[0056] Figure 3 for Figure 1 Detailed implementation steps of step S5;

[0057] 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

[0058] 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.

[0059] 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.

[0060] 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.

[0061] To achieve this, please refer to Figures 1 to 3 The present invention provides an intelligent demand forecasting and analysis method for commodity auctions. In the embodiments of the present invention, please refer to Figure 1 FIG. 1 is a schematic diagram of a process flow of an intelligent demand forecasting and analysis method for commodity auctions according to the present invention. The intelligent demand forecasting and analysis method for commodity auctions comprises the following steps:

[0062] Step S1: Acquire the auction item ontology data of the online commodity auction platform; perform auction item ontology feature analysis based on the auction item ontology data to generate auction item ontology feature data;

[0063] In an embodiment of the present invention, by connecting with the database system or data interface of the online commodity auction platform, the relevant API (application programming interface) in the online commodity auction platform is called or the ontology data of the auction item is directly read from the database. The ontology data usually contains basic information of the auction item, such as the auction item name, description, category, auction item number, upload time, auction item status, etc. The auction item ontology data usually consists of two parts, namely structured data and unstructured data. In this step, firstly, the structured part and the unstructured part in the auction item ontology data are identified and distinguished according to the pre-set classification standard. Structured data includes commodity name, auction item number, category, specification, etc., all of which have fixed fields and formats. Unstructured data includes multimedia data such as commodity description, image, video and text data. The formats of these data are complex and diverse, and cannot be extracted by simple field matching. The relevant unstructured data is saved to obtain the auction item ontology unstructured data, and the separation and extraction of structured and unstructured data are accelerated by multi-threaded concurrent processing. The unstructured data is analyzed and processed by mapping features. This process is mainly carried out through convolutional neural networks (CNN). Unstructured data, such as text description data, is processed by word embedding and other natural language processing technologies to convert text content into a numerical matrix representation that can be calculated. Then, for unstructured data such as images or videos, image features are extracted through pre-trained convolutional neural network models such as ResNet or VGG, mapped to a high-dimensional feature space, and a multimodal feature matrix is ​​generated. During the entire processing process, the multi-layer convolutional layer and pooling layer of the convolutional neural network are used to extract the visual features of the auction items layer by layer, and these features are integrated through the fully connected layer to finally form the unstructured mapping feature data of the auction items for subsequent analysis. As an important part of the analysis of the auction item's ontology features, these data provide support for subsequent feature screening and predictive analysis. The extracted auction item structured data and auction item unstructured mapping feature data are jointly analyzed. First, for the structured data of the auction items, feature engineering methods are used to preprocess and normalize the data. For example, for category information, One-Hot Encoding is used to convert category features into a computable numerical form; for continuous numerical features, the Z-score standardization method is used for normalization so that each feature data can be analyzed within the same scale range. For unstructured mapping feature data, the high-dimensional feature vector extracted by the previous convolutional neural network (such as ResNet or VGG) is used in combination with natural language processing technology (such as the BERT model) to conduct in-depth analysis of the auction item description text and image features.For text descriptions, the word embedding model is used to convert text into vector form to capture the semantic information in the description. For image data, the visual feature vector extracted from the convolution layer is used as input data, and the feature is further enhanced in combination with the structured attributes of the auction item. When integrating the structured and unstructured features of the auction item, feature fusion technology is used. The features of different modalities are merged into a unified feature space through feature concatenation to generate the feature data of the auction item itself.

[0064] Step S2: performing a screening process on the item body gain feature according to the item body feature data to generate the item body gain feature data;

[0065] In the embodiment of the present invention, the feature data of the auction item body is discretized. The purpose of the discretization process is to convert continuous feature data into discrete features so as to perform more complex feature analysis operations later. First, for the numerical features in the auction item body data, such as the continuous features of the size and weight of the auction item, the equal frequency discretization or equal width discretization method is used to divide the continuous numerical features into multiple discrete intervals. Text features and category features do not need to be discretized, and they are discrete in themselves. For example, the category information of the auction item can be directly used as a discrete feature. Through the discretization process, a set of auction item body feature data containing discrete values ​​is generated. Conditional entropy analysis is performed based on the discretized feature data of the auction item body. Conditional entropy is used to measure the ability of a certain feature to reduce the uncertainty of the target variable (such as user demand or the popularity of the auction item) when other features are known. The probability distribution of each discrete feature is calculated, and then the conditional entropy value of the feature is calculated in combination with the target variable to obtain the conditional entropy value of each feature, so as to judge the ability of the feature to reduce the uncertainty of the target variable. The lower the conditional entropy value, the more important the feature is in reducing uncertainty and improving prediction accuracy. By calculating the conditional entropy of each feature, the conditional entropy data of the auction item feature is generated. These data reflect the importance of each auction item feature to the prediction of user demand. The auction item feature data is screened for auction item gain features based on the conditional entropy data of the auction item feature. Feature gain is to screen out useful features by comparing the degree of improvement in the prediction ability of the target variable before and after the feature. A feature gain threshold is set, and the gain of each feature on the prediction of user demand is calculated based on the conditional entropy value. The gain formula is: . Among them, H(Y) is the entropy of the target variable, which represents the system uncertainty when there is no feature, and H(Y|X) is the conditional entropy of feature X. By calculating the gain value, the features whose gain is greater than the set threshold are screened out. Those invalid or noise features that contribute less to demand forecasting are filtered out, and the gain features are retained to generate the gain feature data of the auction item itself. These gain feature data will be used in the subsequent demand forecasting and association analysis to improve the accuracy and efficiency of the forecast.

[0066] Step S3: establishing an item ontology attribute matrix set through the item ontology gain feature data; performing a dynamic association network analysis of the item ontology based on the item ontology attribute matrix set to generate a dynamic item ontology association network;

[0067] In an embodiment of the present invention, based on the selected auction item gain feature data, auction item attribute matrix nodes are designed and classified according to different types. The gain feature data includes multiple key features of the auction item, such as material, visual features, category, user interest tags, etc. For each type of feature, a corresponding matrix node is created to store the numerical value or classification information of the feature. For example, the material feature is mapped to a matrix node, the size feature is mapped to another matrix node, and each node represents an independent dimension of the category. These nodes correspond to different dimensions in the matrix, the rows of the matrix represent the auction items, and the columns represent the values ​​or categories of each feature dimension. The auction item gain feature data is input into these attribute matrix nodes to establish a complete auction item attribute matrix set. The matrix set contains the gain features of all auction items, and each feature is weighted by its gain coefficient. For example, the material, texture complexity, texture distribution features, etc. of a certain auction item, as well as the weight of the corresponding feature type, these values ​​are filled in the corresponding nodes of the matrix to generate the auction item attribute matrix set. Based on the set of lot ontology attribute matrices, the association aggregation analysis of the lot ontology attribute matrices is performed. The purpose of the association aggregation analysis is to calculate the similarity of each feature in the lot ontology attribute matrix set by analyzing the feature similarity between different lots. For example, for two lots in the matrix, if their values ​​in the gain features such as material, size, and category are similar, their degree of association will also be higher. By comparing all lots in pairs, the similarity between each pair of lots is calculated to generate an initial similarity matrix. Then, using an aggregation algorithm (such as weighted clustering or fuzzy C-means clustering algorithm), the lots are clustered according to the values ​​in the similarity matrix, and the association coefficient of each cluster is calculated, which is called the lot matrix association aggregation coefficient, to generate an aggregation coefficient that describes the association of the lot ontology. Design the lot ontology association network based on the lot ontology attribute matrix set. Each node of the lot ontology association network represents a lot, and the edges between the nodes represent the association between the lots. The degree of association is determined by the lot matrix association aggregation coefficient calculated in the previous step. Each lot in the lot ontology attribute matrix is ​​used as a node in the association network, and the similarity between the lots (association aggregation coefficient) is used as the weight of the edge. The edge between the nodes reflects the similarity of the lots in multiple gain feature dimensions. The larger the edge weight value, the stronger the association between the lots. Based on the lot matrix association aggregation coefficient, it is used as the attention weight parameter of the association network to dynamically update the lot ontology association network. The dynamic update process is completed through the network propagation algorithm. Each node dynamically adjusts its association degree through interaction with adjacent nodes. After each update, the association weight of the node in the lot association network will be adjusted according to the new lot interaction data or feature changes. For example, when the lot feature changes, the network will update the edge weight between adjacent nodes in real time to ensure that the association network can always reflect the latest lot feature association.Through this process, the generated dynamic item ontology association network can flexibly reflect the association relationship between items and adapt to the dynamic changes of item characteristics.

[0068] Step S4: Detecting user interaction data in the online commodity auction platform; performing user interaction behavior feature analysis and processing based on the user interaction data to generate user interaction behavior feature data; performing interaction behavior feature analysis and processing of clustering user interaction behavior habit types based on the user interaction behavior feature data to generate clustered user interaction behavior feature data;

[0069] In an embodiment of the present invention, the user behavior of the online commodity auction platform is monitored in real time through a pre-designed API communication interface to collect user interaction data. The interaction data includes operations such as user click behavior, search history, commodity browsing time, commodity addition to favorites, and auction participation. Obtain each interactive action of the user on the platform and store these data in a database regularly. The system can record each user's unique identifier (such as user ID), timestamp, visited commodity ID, and related interaction types. According to different interaction types (such as click, search, auction, etc.), a feature extraction algorithm is used to construct a feature vector for each user. Each feature vector includes important information such as the user's behavior frequency, interaction time, and click sequence. For example, if a user browses the same category of commodities many times and stays for a long time, the user's interest weight in the relevant commodity category will increase. After these features are quantified, they are converted into user interaction behavior feature data for further analysis. The system generates a feature vector matrix for each user, in which each element represents the weight or frequency of a certain behavior of the user. A clustering algorithm (such as K-means or DBSCAN) is used to cluster users with similar behavior patterns into the same category. The behavioral feature data of all users are constructed into a multidimensional feature space, where each dimension represents a behavioral feature (such as click frequency, browsing time, etc.). Then, a clustering algorithm is used to group users, and users with similar behavioral features are grouped into the same cluster. The K value (number of clusters) is determined by cross-validation or elbow method. For example, the initial K value is set to 5, and as the analysis process progresses, the number of clusters is finally determined to be 6. In this way, clustering data of user interaction behavior habits are generated, and each cluster represents a group of users with similar behavior habits. Based on the user behavior habit clustering data, the interactive behavior feature data of all users are clustered and divided into user interaction behavior habit types, and each user is classified into the corresponding cluster group according to his or her behavioral characteristics. The distance between each user's feature vector and each cluster center is checked to determine the cluster category to which the user belongs. The behavioral data of each user is labeled with corresponding cluster labels, which represent the user's behavioral habit type. The generated clustered user interaction behavior feature data records the cluster category and corresponding behavioral characteristics of each user.

[0070] Step S5: construct an optimized probability tree model for classifying the demand items of clustered users based on the clustered user interaction behavior feature data to generate an optimized probability tree model for the demand items of clustered users; transmit the user interaction behavior feature data to the optimized probability tree model for the demand items of clustered users to predict the probability of the real-time user demand items and generate the real-time user demand item probability prediction data; transmit the real-time user demand item probability prediction data to the dynamic item ontology association network to perform intelligent analysis of the real-time user associated demand items and generate the real-time user associated demand item prediction data; and transmit the real-time user associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand items.

[0071] In the embodiment of the present invention, clustered user interaction behavior feature data is used to analyze the conversion features of the auction items required by clustered users. The conversion relationship between the behavior pattern of each type of user and the auction items that are finally sold or auctioned is analyzed. The conversion features reflect the user's behavior path from paying attention to the auction items to the final bidding or purchase. By analyzing these data, the system can extract the conversion features of the auction items required by each user group, and the generated clustered user demand auction item conversion feature data contains the relationship between the behavior pattern of each user group and the auction item conversion features. The clustered user interaction behavior feature data is converted into the probability factor of the auction items required by clustered users, and the behavior features of each user group (such as click frequency, browsing time, number of auctions, etc.) are quantified to form a set of factors representing the probability of user demand. For example, the click frequency can be mapped to a probability value between 0 and 1, and the browsing time is also standardized to generate a corresponding probability factor. Subsequently, these factors are used as input variables, and the preset decision tree classification algorithm is used for model training to construct a clustered user demand auction item classification probability tree model. In the decision tree classification algorithm, each node of the decision tree represents a user behavior feature, and the node splitting is based on the probability factor of the user's demand item. Each splitting process will calculate the information gain or Gini coefficient, and select the optimal feature for node division. For example, the splitting of click frequency may divide users into two categories: high click and low click, and the generated child nodes represent the demand item classification of high click users and low click users respectively. As the depth of the tree increases, a complete demand item classification probability tree model is gradually formed. Combined with the conversion feature data of the clustered user demand items, each node of the decision tree model is optimized and adjusted. By analyzing the actual conversion behavior of users (such as the conversion rate from browsing to bidding), the probability coefficient of the tree node is adjusted so that each node better reflects the relationship between user demand and item conversion. For example, if the user group represented by a node has a significantly low conversion rate, it is necessary to adjust the splitting standard of the node and reduce the weight of the feature. Conversely, nodes with higher conversion rates enhance their influence in the tree model. After this optimization and adjustment, an optimized clustered user demand item probability tree model is generated, providing a more accurate basis for subsequent instant user demand item prediction. The optimized clustering user demand item probability tree model generated by the real-time collected user interaction behavior feature data is used to predict the probability of real-time user demand items. The user interaction behavior feature data includes the user's current click, browse, search, bid and other behavior information. In order to predict real-time demand, the user's behavior data is first automatically collected by the system, and these data are transmitted to the demand prediction model in real time. The optimized clustering user demand item probability tree model dynamically predicts the user's demand probability for a certain type of item based on the user's behavior characteristics. According to the trained node information in the decision tree model, the user's demand probability for browsing and accessing items is predicted.The output value of the probability prediction is a floating probability value, which indicates the user's demand intensity for a certain item. Assuming that the user's demand probability for items in related auction activities is 0.35, and the demand probability for items visited by browsing is 0.15, etc. (the sum of the demand probabilities of items is 1), the instant user demand item probability prediction data is generated, and the output result contains the user's demand probability values ​​for different items. The instant user demand item probability prediction data is transmitted to the dynamic item body association network for intelligent analysis. The instant demand item prediction data will be combined with the relevant item information in the item body association network for analysis. During the analysis process, the association relationship in the network, the user behavior similarity and other data are used to further predict the user's demand for other related items. Each node in the network represents an item, and the weight of the edge represents the strength of the association between items. For example, if the user has a high probability of demanding a certain item, and the item has a high degree of correlation with other items in the association network (such as similar materials and styles), the association network will also increase the probability of demand for these related items, and set priorities based on the quantitative probability information of demand in the instant user demand item probability prediction data, such as the associated items corresponding to the highest probability prediction value in the instant user demand item probability prediction data have a higher priority. Through this intelligent analysis, the system can predict not only the user's demand for the items of current interest, but also the user's potential demand for related items. The generated instant user-related demand item prediction data includes the user's comprehensive demand prediction results for multiple types of items, which are transmitted to the terminal for performing the visual feedback operation of the intelligent prediction of the demand items. On the terminal, the user demand prediction results are displayed in a visual way to feedback the specific situation of the user's demand to the operator of the auction platform. The terminal user can clearly obtain the information related to the pushed instant user-related demand item prediction data through these visualization tools.

[0072] Further, step S1 includes the following steps:

[0073] Step S11: Acquire the auction item data of the online commodity auction platform;

[0074] In an embodiment of the present invention, by connecting to the database system or data interface of the online commodity auction platform, the relevant API (application programming interface) in the online commodity auction platform is called or the ontological data of the auction item is directly read from the database. The ontological data usually contains the basic information of the auction item, such as the auction item name, description, category, auction item number, upload time, auction item status, etc. These data are automatically extracted by the system from the commodity information library of the auction platform to ensure that the latest and most complete auction item data can be obtained. The auction item ontological data is the basic data source for subsequent analysis. The auction items of all auction platforms are stored in the platform database in a certain standard format. Therefore, during the data extraction process, SQL statements or RESTful API interfaces can be used for batch query and reading operations to ensure the integrity and accuracy of the data.

[0075] Step S12: extracting the auction item body structured data and the auction item body unstructured data of the auction item body data;

[0076] In an embodiment of the present invention, the auction item data generally consists of two parts, namely structured data and unstructured data. In this step, firstly, the structured part and unstructured part in the auction item data are identified and distinguished according to the pre-set classification standard. Structured data includes product name, lot number, category, specification, etc. All these data have fixed fields and formats. When extracted, they can be directly queried through SQL database or data can be captured based on XML, JSON and other format parsing. Unstructured data includes multimedia data and text data such as product descriptions, images, videos, etc. The formats of these data are complex and diverse, and they cannot be extracted by simple field matching. The relevant unstructured data is saved to obtain the auction item unstructured data, and the separation and extraction of structured and unstructured data are accelerated by multi-threaded concurrent processing.

[0077] Step S13: performing unstructured mapping feature analysis on the auction item based on the auction item body unstructured data to generate unstructured mapping feature data of the auction item;

[0078] In the embodiment of the present invention, after the extraction of the unstructured data of the auction item is completed, it is necessary to analyze and process the mapping features of these unstructured data. This process is mainly carried out through convolutional neural networks (CNN). First, unstructured data, such as text description data, is vectorized by natural language processing technologies such as word embedding, and the text content is converted into a numerical matrix representation that can be calculated. Then, for unstructured data such as images or videos, image features are extracted through pre-trained convolutional neural network models such as ResNet or VGG, and mapped to a high-dimensional feature space to generate a multimodal feature matrix. During the entire processing process, the multi-layer convolution layer and pooling layer of the convolutional neural network are used to extract the visual features of the auction item layer by layer, and these features are integrated through the fully connected layer, and finally the unstructured mapping feature data of the auction item for subsequent analysis is formed. These data, as an important part of the analysis of the characteristics of the auction item itself, provide support for subsequent feature screening and predictive analysis.

[0079] Step S14: performing auction item ontology feature analysis based on the auction item ontology structured data and the auction item unstructured mapping feature data to generate auction item ontology feature data.

[0080] In an embodiment of the present invention, the extracted structured data of the auction items and the unstructured mapping feature data of the auction items are jointly analyzed. First, for the structured data of the auction items, the feature engineering method is used to preprocess and normalize the data. For example, for category information, the category features can be converted into a computable numerical form using One-Hot Encoding; for continuous numerical features, the Z-score normalization method is used to normalize the features so that each feature data is analyzed within the same scale range. For the unstructured mapping feature data, the high-dimensional feature vector extracted by the previous convolutional neural network (such as ResNet or VGG) is first used in combination with natural language processing technology (such as the BERT model) to conduct an in-depth analysis of the auction item description text and image features. For text descriptions, the Word Embedding model is used to convert the text into a vector form to capture the semantic information in the description. For image data, the visual feature vector extracted from the convolution layer is used as input data, and the feature enhancement is further performed in combination with the structured attributes of the auction items. When integrating the structured and unstructured features of the auction items, feature fusion technology is used. The features of different modes are merged into a unified feature space through feature concatenation. For example, for the structured data of each lot, such as the lot category and size, its structured feature vector can be concatenated with the unstructured image feature vector and text feature vector to form a high-dimensional feature vector. This concatenation method ensures that different types of data can be comprehensively analyzed in the same feature space, retaining the information of each type of data. Structured data and unstructured data are fused through feature concatenation to form a multi-dimensional feature space and generate the lot body feature data.

[0081] Further, step S13 includes the following steps:

[0082] Step S131: performing unstructured type classification processing on the auction item body unstructured data to generate classified auction item body unstructured data;

[0083] In the embodiment of the present invention, the unstructured data of the auction item body is processed by type classification. Unstructured data includes text data, image data, video data, audio data, etc. In the processing process, the data recognition algorithm is used to classify the data according to the file format and content characteristics, for example, the text description file format, image file format, and video file format of the auction item are identified. For text data, natural language processing (NLP) technology is used for further language analysis, while image and video data are pre-processed using image recognition technology and video frame segmentation technology, and the original unstructured data is divided into different category data sets by type to generate classified auction item body unstructured data.

[0084] Step S132: performing convolutional layer analysis and processing of multi-modal unstructured mapping features of auction items based on the classified auction item body unstructured data to generate a convolutional layer of multi-modal mapping features of auction items;

[0085] In the embodiment of the present invention, a convolutional neural network (CNN) model is used to extract features from unstructured data of different categories. For image data, a deep convolutional neural network (such as ResNet, VGG, etc.) is used to perform convolution operations on the features of each level of the image. Each convolution layer extracts visual features such as edges, colors, and textures of the auction item image through a convolution kernel (filter). For text data, a convolution-based text convolutional neural network (Text-CNN) model is used to convert the text into a vector form through word embedding to extract important vocabulary and semantic information in the text. In the processing of the convolution layer, the convolution operation of the feature includes multi-layer convolution, pooling (Max Pooling) and activation function (ReLU) calculation, and the deep-level features of the image and text data are extracted layer by layer. The generated convolution layer is marked as the multi-modal mapping feature convolution layer of the auction item processed by the corresponding modality. For example, after the image is processed by three convolutions, the generated convolution feature vector can represent the visual features such as the object contour and color distribution in the image, and the text can extract key semantic fragments such as brand, material and other information through convolution.

[0086] Step S133: designing a multi-modal mapping feature convolutional neural network model based on the multi-modal mapping feature convolutional layer of the auction items;

[0087] In an embodiment of the present invention, a multimodal convolutional neural network model is designed based on a multimodal mapping feature convolution layer. The model includes multiple input layers for receiving feature data of different modalities, for example, one layer receives a text feature vector and another layer receives an image feature vector. Each input layer is connected to a set of convolutional layers and pooling layers to extract the depth information of each modal feature respectively. In order to fuse the features of different modalities, a feature splicing layer is designed to splice the outputs of the text feature convolution layer and the image feature convolution layer. Next, the spliced ​​features are reduced in dimension through a fully connected layer to generate a unified feature vector. When designing the network, specific parameters are set, such as a convolution kernel size of 3x3, a number of convolutional layers of 5, an activation function using a ReLU function, an optimization algorithm using an Adam optimizer, and a learning rate set to 0.001. Through this multimodal convolutional neural network model, the features of different unstructured data can be effectively fused together to generate a joint feature vector containing text and image information.

[0088] Step S134: performing preliminary unstructured item mapping feature processing on the unstructured item data according to the multimodal item mapping feature convolutional neural network model to generate preliminary unstructured item mapping feature data; and performing feature full-connection mapping optimization processing on the preliminary unstructured item mapping feature data to generate unstructured item mapping feature data.

[0089] In the embodiment of the present invention, the unstructured data of the auction item body is preliminarily processed according to the designed multimodal convolutional neural network model. In this stage, each unstructured data is input into different input layers of the model, and after multi-layer convolution operation, the system generates preliminary unstructured mapping feature data of the auction item. The preliminary feature data is a set of high-dimensional feature vectors, which reflects the complex features of multimodal data such as text information and image information of the auction item. For example, for image data, the feature vector after convolution may contain information such as the shape, material texture, etc. of the object, while for text data, the feature vector after convolution may include important keywords and sentence structure. The preliminarily generated unstructured mapping feature data contains redundancy and noise of data full connection fusion, so full connection mapping optimization processing is performed on this basis. The fully connected layer inputs all preliminary features, and reduces the high-dimensional features to a smaller feature space through linear transformation and activation function. Through this fully connected mapping processing, the expression ability of useful features can be enhanced and the influence of noise data can be reduced. For example, the convolution features of the image may contain background information, and the fully connected layer can reduce the weight of the background features through weight adjustment to enhance the feature performance of the foreground object. The unstructured mapping feature data of the auction items finally generated is a set of feature vectors after dimensionality reduction, which contains the multimodal data features of the auction items and has been optimized to have stronger prediction and classification capabilities.

[0090] Further, step S2 includes the following steps:

[0091] Step S21: performing feature discretization processing on the auction item feature data to generate auction item feature discretization data;

[0092] In an embodiment of the present invention, the acquired auction item feature data is discretized. The purpose of the discretization process is to convert continuous feature data into discrete features so as to facilitate subsequent more complex feature analysis operations. First, for the numerical features in the auction item data, such as the size, weight and other continuous features of the auction item, an equal-frequency discretization or equal-width discretization method is used to divide the continuous numerical features into multiple discrete intervals. Text features and category features do not need to be discretized, as they are discrete in themselves. For example, the category information of the auction item can be directly used as a discrete feature. Through the discretization process, a set of auction item feature data containing discrete values ​​is generated.

[0093] Step S22: performing conditional entropy analysis of the item's main features based on the discretized data of the item's main features to generate conditional entropy data of the item's main features;

[0094] In the embodiment of the present invention, conditional entropy analysis is performed based on the discretized feature data of the auction item itself. Conditional entropy is used to measure the ability of a certain feature to reduce the uncertainty of the target variable (such as user demand or the popularity of the auction item) when other features are known. The probability distribution of each discrete feature is calculated, and then the conditional entropy value of the feature is calculated in combination with the target variable. The specific calculation method of conditional entropy is that for each feature, the probability of the occurrence of the feature is first calculated, and then the probability distribution of the target variable under the condition of the feature is calculated when the feature is known. Then, based on these probability distributions, the uncertainty, that is, the entropy value, is calculated. In this way, the conditional entropy value of each feature is obtained, so as to judge the ability of the feature to reduce the uncertainty of the target variable. The lower the conditional entropy value, the more important the feature is to reduce uncertainty and improve prediction accuracy. By calculating the conditional entropy of each feature, the conditional entropy data of the auction item feature is generated, which reflects the importance of each auction item feature to user demand prediction.

[0095] Step S23: performing auction item body gain feature screening processing on the auction item body feature data according to the auction item body feature condition entropy data to generate auction item body gain feature data.

[0096] In the embodiment of the present invention, feature gain is to screen out useful features by comparing the degree of improvement in the prediction ability of the target variable before and after the feature. In the specific implementation process, a feature gain threshold is first set, and the gain of the feature for user demand prediction is calculated based on the conditional entropy value of each feature. The gain formula is: . Among them, H(Y) is the entropy of the target variable, which represents the system uncertainty when there is no feature, and H(Y|X) is the conditional entropy of feature X. By calculating the gain value, the features whose gain is greater than the set threshold are screened out. For example, if the gain threshold is set to 0.05, if the material feature gain of a certain item is 0.07, it means that it helps to significantly reduce the uncertainty of the target variable, and it is retained; the feature with a gain value of 0.03 is discarded. In this way, invalid or noise features that contribute less to demand forecasting can be filtered out, the gain features can be retained, and the gain feature data of the item itself can be generated. These gain feature data will be used in subsequent demand forecasting and association analysis to improve the accuracy and efficiency of the forecast.

[0097] Further, step S22 includes the following steps:

[0098] Step S221: performing frequency domain data analysis on the discretized data of the auction item's main features to generate frequency domain data of the auction item's main features;

[0099] In an embodiment of the present invention, frequency domain data analysis is performed on the discretized data of the auction item's main features. The main purpose of frequency domain data analysis is to analyze the changing frequency of the features to discover potential periodic features or high-frequency features, and use Fast Fourier Transform (FFT) to transform the discretized features and convert them to the frequency domain. Through FFT, the discrete feature data in the time domain is mapped to the frequency domain to generate spectrum information containing different frequency components. In order to ensure the accuracy of the analysis, the frequency sampling rate is set to 100Hz, and the sampling results are filtered to filter out low-frequency noise and retain the main high-frequency features. The frequency components in the auction item's main features can be effectively extracted, and the frequency domain data of the auction item's main features can be generated. This frequency domain data contains the distribution of features in different frequency ranges, which provides a basis for subsequent entropy analysis and feature screening.

[0100] Step S222: performing an entropy analysis on the item body feature label according to the frequency domain data of the item body feature to generate the item body feature label entropy data;

[0101] In the embodiment of the present invention, label entropy analysis is performed based on the frequency domain data of the auction item's main features. Label entropy is to evaluate the importance of the feature by calculating the information uncertainty between the feature and the corresponding category label (such as the auction item's classification or user behavior label). Determine the label of the auction item, such as the auction item's category, material, or user behavior type, and then calculate the entropy value between each feature frequency domain data and the corresponding label. The calculation of label entropy is based on the definition of entropy in statistics. By calculating the joint distribution between the statistical feature value and the label, the distribution entropy of the label on the feature is calculated to measure the ability of the feature to reduce uncertainty. For example, if the material feature of a certain auction item has a strong correlation with the category to which it belongs, the label entropy of the feature will be lower, and the label entropy data of the auction item's main features will be generated to evaluate the strength of the relationship between each frequency domain feature and the label.

[0102] Step S223: performing auction item feature conditional entropy analysis on the auction item feature frequency domain data and the auction item feature label entropy data to generate auction item feature conditional entropy data.

[0103] In an embodiment of the present invention, conditional entropy is used to evaluate the degree to which a certain feature eliminates the uncertainty of a target variable when other features are known. For each auction item feature, the frequency domain data and label entropy data are used to calculate the joint probability distribution of the feature and the target variable (such as the user behavior label). The joint distribution is used to represent the correlation between the feature and the target variable. According to the joint distribution, the conditional entropy formula is used for analysis. The calculation of conditional entropy represents the degree to which the uncertainty of the target variable is reduced given the label entropy of the auction item feature. For each auction item feature, the smaller its conditional entropy value is, the more effectively the feature can reduce the uncertainty of the prediction when other features are known, thereby being used to screen out the features that have the greatest benefit for the prediction. Finally, the conditional entropy data of the auction item feature is obtained by calculation, and these data provide an important basis for subsequent intelligent demand forecasting analysis.

[0104] Further, as an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S3 in the embodiment, step S3 includes the following steps:

[0105] Step S31: designing and classifying item ontology attribute matrix nodes according to each gain feature type in the item ontology gain feature data, and transmitting the item ontology gain feature data to the item ontology attribute matrix nodes to establish an item ontology attribute matrix set;

[0106] In an embodiment of the present invention, based on the selected auction item gain feature data, auction item attribute matrix nodes are designed and classified according to different types. The gain feature data includes multiple key features of the auction item, such as material, visual features, category, user interest tags, etc. For each type of feature, a corresponding matrix node is created to store the numerical value or classification information of the feature. For example, the material feature is mapped to a matrix node, the size feature is mapped to another matrix node, and each node represents an independent dimension of the category. These nodes correspond to different dimensions in the matrix, the rows of the matrix represent the auction items, and the columns represent the values ​​or categories of each feature dimension. The auction item gain feature data is input into these attribute matrix nodes to establish a complete auction item attribute matrix set. The matrix set contains the gain features of all auction items, and each feature is weighted by its gain coefficient. For example, the material, texture complexity, texture distribution features, etc. of a certain auction item, as well as the weight of the corresponding feature type, these values ​​are filled in the corresponding nodes of the matrix to generate the auction item attribute matrix set. The matrix set provides a basic data structure for subsequent association analysis and network construction.

[0107] Step S32: performing association aggregation analysis on the auction item ontology attribute matrix based on the auction item ontology attribute matrix set to generate an auction item matrix association aggregation coefficient;

[0108] In the embodiment of the present invention, the purpose of the association aggregation analysis is to calculate the similarity of each feature in the auction item attribute matrix set by analyzing the feature similarity between different auction items. Commonly used similarity measurement methods include cosine similarity, Euclidean distance or Jaccard coefficient. For example, for two auction items in the matrix, if their value similarity in gain features such as material, size, and category is high, their degree of association will also be higher. By comparing all auction items in pairs, the similarity between each pair of auction items is calculated to generate an initial similarity matrix. Then, using an aggregation algorithm (such as weighted clustering or fuzzy C-means clustering algorithm), the auction items are clustered according to the values ​​in the similarity matrix, and the association coefficient of each cluster is calculated, which is called the auction item matrix association aggregation coefficient. The aggregation coefficient describing the association of the auction item body is generated, which provides an important reference for the subsequent association network.

[0109] Step S33: Designing an item ontology association network through the item ontology attribute matrix set, and dynamically updating the item ontology association degree of the item ontology association network through the matrix association aggregation coefficient as the item association attention weight parameter to generate a dynamic item ontology association network.

[0110] In the embodiment of the present invention, based on the auction item ontology attribute matrix set, an auction item ontology association network is designed. Each node of the auction item ontology association network represents an auction item, and the edge between the nodes represents the association between the auction items. The degree of association is determined by the auction item matrix association aggregation coefficient calculated in the previous step. Each auction item in the auction item ontology attribute matrix set is used as a node in the association network, and the similarity between the auction items (association aggregation coefficient) is used as the weight of the edge. The edge between the nodes reflects the similarity of the auction items in multiple gain feature dimensions. The larger the edge weight value, the stronger the association between the auction items. Based on the auction item matrix association aggregation coefficient, it is used as the attention weight parameter of the association network to dynamically update the auction item ontology association network. The dynamic update process is completed through the network propagation algorithm. Each node dynamically adjusts its degree of association by interacting with adjacent nodes. After each update, the association weight of the node in the auction item association network will be adjusted according to the new auction item interaction data or feature changes. For example, when the auction item feature changes, the network will update the edge weight between adjacent nodes in real time to ensure that the association network can always reflect the latest auction item feature association. Through this process, the generated dynamic item ontology association network can flexibly reflect the association relationship between items and adapt to the dynamic changes of item characteristics.

[0111] Further, step S32 includes the following steps:

[0112] Step S321: using matrix nodes of various types in the auction item ontology attribute matrix set as first auction item ontology attribute data, and randomly selecting matrix nodes of the remaining auction item ontology attribute matrix set in the type corresponding to the first auction item ontology attribute node data as second auction item ontology attribute node data;

[0113] In an embodiment of the present invention, attribute matrix node data of each type is extracted from the auction item ontology attribute matrix set as the first auction item ontology attribute data. The auction item ontology attribute matrix set includes gain feature data of multiple dimensions, such as attributes such as material, color, size, and category. Each attribute matrix node represents feature information of a specific dimension, and the attribute matrix nodes of all auction items are organized into a matrix structure according to their feature categories. For each type of auction item ontology feature, a node is randomly selected from the category as the first auction item ontology attribute node data. Then, another node is randomly selected from the remaining auction item ontology attribute matrix set of the same type as the node as the second auction item ontology attribute node data. For example, if the material category is selected, the system first selects the material information of a auction item from the "material" category as the first node data, and then randomly extracts the material information of another auction item from other auction items as the second node data. This random selection process ensures fairness and comprehensiveness of all auction item features, laying a foundation for subsequent node similarity and association analysis.

[0114] Step S322: performing node similarity calculation on the first lot body attribute node data and the second lot body attribute node data to generate node similarity data;

[0115] In an embodiment of the present invention, node similarity calculation is performed for the first and second lot ontology attribute node data selected in the previous step. Node similarity measures the degree of similarity between two nodes in a specific attribute. Different types of attribute data use different similarity calculation methods. For example, for continuous data (such as size), the Euclidean distance formula is used to calculate the similarity; for categorical data (such as material, color, etc.), the cosine similarity or Jaccard coefficient is used to calculate its similarity. For example, the continuous attribute data is standardized to ensure that the features of different dimensions are comparable. Then, the similarity formula is applied separately according to the type of different features. For example, for the "material" feature, the Jaccard coefficients of the two nodes are calculated. If the two lots belong to the same material category, their similarity is close to 1, otherwise it is close to 0. Finally, a set of similarity data is generated to represent the similarity of each pair of nodes in a specific attribute. These similarity data are used to evaluate the correlation between different lots in the gain feature.

[0116] Step S323: Calculate the node access overlap of the first lot body attribute node data and the second lot body attribute node data to generate node access overlap data;

[0117] In an embodiment of the present invention, the node access overlap is calculated to further measure the degree of overlap between the two nodes in terms of access frequency or user behavior. The node access overlap is measured by analyzing user interaction data, that is, evaluating whether there is overlap in the user's access behavior between different lots. In the specific calculation, first obtain the user's access records to the first and second lots on the auction platform, and count the user groups visited by each lot. Then, calculate the ratio of the number of users who jointly visit the two lots to the total number of users who visit. For example, if there are 100 visiting users for the first lot and the second lot, and 50 users visit these two lots, the access overlap is 50 / 100=0.5. In this way, node access overlap data is generated, indicating the degree of association between the two lots at the user interaction level. This data provides an important basis for the user behavior dimension for the subsequent auction attribute matrix association aggregation analysis.

[0118] Step S324: performing association aggregation analysis on the auction item ontology attribute matrix set according to the node similarity data and the node access overlap data, and generating an auction item matrix association aggregation coefficient.

[0119] In an embodiment of the present invention, node similarity data and node access overlap data are used to perform association aggregation analysis of the auction item ontology attribute matrix. The purpose of association aggregation analysis is to comprehensively evaluate the correlation between the auction items in terms of feature similarity and user access behavior. For example, the similarity data and access overlap data are weighted and combined, and the weight is determined by the needs of the specific application scenario. For example, if the weight of the user access behavior is high, the weight of the access overlap is increased, and the association score of each pair of auction items is calculated. Next, for each auction item, the aggregation coefficient between it and all other auction items is calculated, which is called the auction item matrix association aggregation coefficient. The auction item association aggregation coefficient reflects the association strength between the auction item and other auction items in multiple dimensions (feature similarity and user behavior overlap). For example, if the material and size similarity of two auction items are high, and they have a large overlap in user access behavior, their association aggregation coefficient will be high. Finally, the generated association aggregation coefficient provides an important basis for the subsequent design and dynamic update of the auction item association network.

[0120] Further, step S4 includes the following steps:

[0121] Step S41: Detecting user interaction data in the online commodity auction platform;

[0122] In an embodiment of the present invention, the user behavior of the online commodity auction platform is monitored in real time through a pre-designed API communication interface to collect user interaction data. The interaction data includes operations such as user click behavior, search history, commodity browsing time, adding commodities to favorites, and auction participation. Each interaction action of the user on the platform is obtained, and these data are stored in the database regularly. The system can record each user's unique identifier (such as user ID), timestamp, visited commodity ID, and related interaction types. These interaction data provide a basis for subsequent behavioral analysis. For example, if a user clicks on a specific auction item multiple times in a day and participates in the auction of the auction item, these behavioral data will be detected and recorded by the system.

[0123] Step S42: performing user interaction behavior feature analysis and processing according to the user interaction data to generate user interaction behavior feature data;

[0124] In an embodiment of the present invention, the user's interaction data is cleaned and preprocessed to remove redundant information or noise data, such as abnormal click frequencies or invalid access records. Next, a feature extraction algorithm is used to construct a feature vector for each user based on different interaction types (such as clicks, searches, auctions, etc.). Each feature vector includes important information such as the user's behavior frequency, interaction duration, and click sequence. For example, if a user browses the same category of goods many times and stays for a long time, the user's interest weight in the relevant product category will increase. After these features are quantified, they are converted into user interaction behavior feature data for further analysis. The system generates a feature vector matrix for each user, in which each element represents the weight or frequency of a certain behavior of the user.

[0125] Step S43: performing user interaction behavior habit cluster analysis based on the user interaction behavior feature data to generate user interaction behavior habit cluster data;

[0126] In an embodiment of the present invention, a clustering algorithm (such as K-means or DBSCAN) is used to cluster users with similar behavior patterns into the same class. The behavioral feature data of all users are constructed into a multidimensional feature space, and each dimension represents a behavioral feature (such as click frequency, browsing time, etc.). Then, a clustering algorithm is used to group the users, and users with similar behavioral characteristics are classified into the same cluster. For example, if a group of users exhibit high-frequency bidding behavior and have a strong interest in high-value goods, these users will be clustered into one group. The K value (number of clusters) is determined by cross-validation or the elbow method. For example, the initial K value is set to 5. As the analysis process progresses, the number of clusters is finally determined to be 6. In this way, clustering data of user interaction behavior habits are generated, and each cluster represents a group of users with similar behavior habits.

[0127] Step S44: performing data division processing on the user interaction behavior characteristic data into clustered user interaction behavior habit types according to the user interaction behavior habit clustering data, and generating clustered user interaction behavior characteristic data.

[0128] In an embodiment of the present invention, the interactive behavior feature data of all users are clustered into user interactive behavior habit types based on user behavioral habit clustering data, and each user is classified into a corresponding cluster group according to his or her behavioral characteristics. The system determines the cluster category to which the user belongs by checking the distance between each user's feature vector and each cluster center. Each user's behavioral data is labeled with corresponding cluster labels, which represent the user's behavioral habit type. For example, a user is classified as a "high-frequency bidding group" because of his or her bidding behavior and browsing behavior for high-value goods. The generated clustered user interactive behavior feature data records the cluster category and corresponding behavioral characteristics of each user. Such data can be further used in demand forecasting and personalized recommendation systems to better serve user needs.

[0129] Further, as an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 FIG. 5 is a schematic diagram of a detailed step flow chart of step S5 in FIG. 5 . In this embodiment, step S5 includes the following steps:

[0130] Step S51: performing clustered user demand auction item conversion feature analysis based on clustered user interaction behavior feature data to generate clustered user demand auction item conversion feature data;

[0131] In an embodiment of the present invention, clustered user interaction behavior feature data is used to analyze the conversion features of the auction items demanded by clustered users. The conversion relationship between the behavior pattern of each type of user and the auction items that are finally sold or auctioned is analyzed. User behavior features include the number of clicks, browsing time, frequency of participating in auctions, etc., while the conversion features reflect the user's behavior path from paying attention to the auction items to the final bidding or purchase. By analyzing these data, the system can extract the conversion features of the auction items demanded by each user group. For example, for high-frequency bidding groups, it may be found that their conversion rate is highly correlated with the brand or price range of the auction items. The final generated clustered user demand auction item conversion feature data contains the relationship between the behavior pattern of each user group and the auction item conversion features.

[0132] Step S52: using clustered user interaction behavior feature data as clustered user demand item probability factors; establishing clustered user demand item classification probability mapping relationships for clustered user demand item probability factors based on a preset decision tree classification algorithm to obtain a clustered user demand item probability tree model; optimizing and adjusting the tree node probability coefficients of the clustered user demand item probability tree model through clustered user demand item conversion feature data to generate an optimized clustered user demand item probability tree model;

[0133] In an embodiment of the present invention, clustered user interaction behavior feature data is converted into probability factors of clustered user demand items, and the behavior features of each user group (such as click frequency, browsing time, number of times of participating in the auction, etc.) are quantified to form a set of factors representing the probability of user demand. For example, the click frequency can be mapped to a probability value between 0 and 1, and the browsing time is also standardized to generate a corresponding probability factor. Subsequently, these factors are used as input variables, and a preset decision tree classification algorithm is used for model training to construct a clustered user demand item classification probability tree model. In the decision tree classification algorithm, each node of the decision tree represents a user behavior feature, and the node splits according to the probability factor of the user demand item. Each splitting process calculates the information gain or the Gini coefficient, and selects the optimal feature for node division. For example, the splitting of the click frequency may divide users into two categories: high click and low click, and the generated child nodes represent the demand item classification of high click users and low click users, respectively. As the depth of the tree increases, a complete demand item classification probability tree model is gradually formed. For example, the information gain threshold is set to 0.01, which means that each time a node splits, only features with information gain greater than 0.01 will be used for node division. This threshold ensures that the model can effectively select the most discriminative features. The maximum tree depth is set to 10, limiting the depth of the decision tree to no more than 10 layers to avoid overfitting. This controls the complexity of the tree and ensures that the model has good generalization ability when making predictions. The minimum number of samples for each leaf node is set to 20, that is, the latest interaction data, to ensure that each terminal node (leaf node) contains at least 20 user samples, avoid inaccurate predictions caused by over-refinement of nodes, and be able to update the number of samples in a timely manner. Then, combined with the conversion feature data of the clustered user demand auction items, the various nodes of the decision tree model are optimized and adjusted. By analyzing the actual conversion behavior of users (such as the conversion rate from browsing to bidding), the probability coefficient of the tree node is adjusted so that each node better reflects the relationship between user demand and auction item conversion. For example, if the user group represented by a node has a significantly low conversion rate, it is necessary to adjust the splitting criteria of the node and reduce the weight of the feature. On the contrary, nodes with higher conversion rates increase their influence in the tree model. After this optimization adjustment, an optimized clustering user demand item probability tree model is generated, providing a more accurate basis for subsequent instant user demand item prediction.

[0134] Step S53: transmitting the user interaction behavior feature data to the optimized clustering user demand item probability tree model to perform instant user demand item probability prediction, and generate instant user demand item probability prediction data;

[0135] In an embodiment of the present invention, an optimized clustering user demand item probability tree model is generated from the user interaction behavior feature data collected in real time to predict the probability of the user's immediate demand item. The user interaction behavior feature data includes the user's current click, browse, search, bid and other behavior information. In order to predict the immediate demand, the user's behavior data is first automatically collected by the system, and these data are transmitted to the demand prediction model in real time. The optimized clustering user demand item probability tree model dynamically predicts the user's demand probability for a certain type of item based on the user's behavior characteristics. For example, if a user frequently browses items of the same category in a short period of time, and participates in related bidding activities, browsing visits and other behaviors, the model will input such behavior characteristics, and predict the user's demand probability for browsing and visiting items based on the trained node information in the decision tree model. The output value of the probability prediction is a floating probability value, which indicates the user's demand intensity for a certain item. Assuming that the user's demand probability for the items in the relevant auction activities is 0.35, and the demand probability for the items visited is 0.15, etc. (the total demand probability of the items is 1), it is said that the instant user demand probability prediction data for the items is generated, and the output result includes the user's demand probability values ​​for different items.

[0136] Step S54: Transmit the probability prediction data of the real-time user demand item to the dynamic item ontology association network for intelligent analysis of the real-time user-associated demand item, generate the real-time user-associated demand item prediction data, and transmit the real-time user-associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand item.

[0137] In an embodiment of the present invention, the instant user demand item probability prediction data is transmitted to the dynamic item body association network for intelligent analysis. The dynamic item body association network is a dynamically updated network constructed based on the gain characteristics, user behavior and item association between different items. The instant demand item prediction data will be combined with the relevant item information in the item body association network for analysis. During the analysis process, the association relationship in the network, the user behavior similarity and other data are used to further predict the user's demand for other related items. Each node in the network represents an item, and the weight of the edge represents the strength of the association between the items. For example, if the user has a high probability of demanding a certain item, and the item has a high degree of association with other items in the association network (such as similar materials and styles), the association network will also increase the demand probability of these related items, and set the priority based on the demand quantification probability information in the instant user demand item probability prediction data, such as the associated product corresponding to the highest probability prediction value in the instant user demand item probability prediction data has a higher priority. Through this intelligent analysis, the system can predict not only the user's demand for the item currently of interest, but also the user's potential demand for the related items. The generated instant user-related demand item forecast data includes the comprehensive demand forecast results of users for multiple types of items. These data are transmitted to the terminal for visual feedback of intelligent forecast of demand items. On the terminal, the user demand forecast results are displayed in a visual way to provide feedback to the operators of the auction platform on the specific situation of user demand. Terminal users can clearly obtain the relevant information of the instant user-related demand item forecast data through these visualization tools.

[0138] This specification provides an intelligent demand forecasting and analysis system for commodity auctions, which is used to execute the intelligent demand forecasting and analysis method for commodity auctions as described above. The intelligent demand forecasting and analysis system for commodity auctions includes:

[0139] The item body feature analysis module is used to obtain the item body data of the online commodity auction platform; perform item body feature analysis based on the item body data to generate item body feature data;

[0140] A gain feature screening module is used to perform a screening process on the gain features of the auction item itself according to the auction item itself feature data to generate the auction item itself gain feature data;

[0141] A dynamic auction item ontology association module is used to establish an auction item ontology attribute matrix set through the auction item ontology gain feature data; perform a dynamic association network analysis of the auction item ontology based on the auction item ontology attribute matrix set to generate a dynamic auction item ontology association network;

[0142] The clustering user interaction behavior analysis module is used to detect user interaction data in the online commodity auction platform; perform user interaction behavior feature analysis and processing based on the user interaction data to generate user interaction behavior feature data; perform clustering user interaction behavior habit type interaction behavior feature analysis and processing based on the user interaction behavior feature data to generate clustered user interaction behavior feature data;

[0143] The instant user-associated demand item prediction module is used to construct an optimized demand item classification probability tree model for clustered users based on the clustered user interaction behavior feature data, and generate an optimized clustered user demand item probability tree model; transmit the user interaction behavior feature data to the optimized clustered user demand item probability tree model to perform instant user demand item probability prediction, and generate instant user demand item probability prediction data; transmit the instant user demand item probability prediction data to the dynamic item ontology association network to perform instant user-associated demand item intelligent analysis, generate instant user-associated demand item prediction data, and transmit the instant user-associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand item.

[0144] The beneficial effect of the present application is that, by acquiring and analyzing the auction item data in the commodity auction platform, combining the structured and unstructured features of the commodity, and using the multimodal unstructured mapping feature convolutional neural network model, the complex characteristics of the auction item are deeply analyzed. Traditional methods usually rely only on simple analysis of structured data, ignoring the rich information of the commodity ontology unstructured data. Through the convolutional layer neural network, the unstructured data is deeply processed, and the potential characteristics of the auction item can be captured more accurately, thereby improving the accurate prediction of commodity demand. At the same time, through multi-dimensional feature processing such as feature gain screening and conditional entropy analysis, redundant features are effectively eliminated, and the efficiency and accuracy of the prediction model are improved. By performing association analysis on the auction item ontology gain feature data and constructing an association network, the network is continuously updated dynamically based on the similarity between the commodity ontology features and the associated user access behavior, reflecting the degree of association between commodities and changes in user demand in real time, and can be continuously updated with changes in user demand and commodity features, to achieve dynamic management and analysis of the degree of association between commodities. By analyzing the interactive behavior characteristics of users in the auction platform in detail, and classifying the interactive behavior habits of different types of users through cluster analysis. On this basis, a clustering user demand item classification probability tree model is constructed to perform personalized demand forecasting based on the behavior patterns and demand characteristics of different user groups. This deep clustering analysis based on user behavior characteristics not only helps the platform better understand the behavior habits of different users and optimize the item recommendation strategy, but also improves the personalization and pertinence of the prediction results, thereby meeting the diverse needs of different users. The clustering user demand item classification probability tree model can quickly update the prediction results based on the user's instant interactive behavior, so that the demand forecast can be adjusted in real time as the user behavior changes, greatly improving the response speed and dynamic adaptability of the prediction analysis. The generated user demand forecast results are transmitted to the terminal through an intelligent prediction feedback mechanism to form a visual feedback operation. This instant feedback mechanism not only provides decision support for platform managers, but also provides users with a more intuitive display of demand forecast results, which helps to improve the user experience of auction participants and increase users' stickiness and satisfaction with the platform. In addition, the platform can optimize the recommendation and auction strategy of the items based on the demand forecast results, thereby improving transaction efficiency and success rate, and ultimately promoting the improvement of the overall benefits of the platform. Therefore, the intelligent demand prediction and analysis method for commodity auctions of the present invention can extract the key representation features of the auction item body, and screen the high-gain feature types in the auction item body to improve data processing efficiency while accurately expressing the auction item information, and establish a dynamic auction item body association network through the association characteristics in the high-gain features in the auction item body, thereby dynamically reflecting the association relationship between each commodity. When a user is interested in a certain auction item, intelligent recommendations are made to the user based on the dynamic association relationship of the auction items in the dynamic auction item body association network.In addition, the analysis of user needs is carried out by clustering user analysis on user interaction behaviors to reflect the interaction habits of users in different clusters, and then the probability of favorite items of users in different clusters is predicted through the decision tree algorithm to reflect the distribution information of different users for different favorite items to reflect the predicted probability of user item demand. Combined with the dynamic item ontology association network and the user item demand prediction probability, the intelligent analysis of the instant user-associated demand items is carried out to reflect the user's instant demand items and the corresponding quantitative demand information, so as to improve the accuracy and efficiency of demand prediction in the commodity auction platform.

[0145] 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.

[0146] 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 demand forecasting and analysis method for commodity auctions, characterized in that: The following steps are involved: Step S1: Acquire the auction item ontology data of the online commodity auction platform; perform auction item ontology feature analysis based on the auction item ontology data to generate auction item ontology feature data; Wherein, step S1 comprises the following steps: Step S11: Acquire the auction item data of the online commodity auction platform; Step S12: extracting the auction item body structured data and the auction item body unstructured data of the auction item body data; Step S13: performing unstructured mapping feature analysis on the auction item based on the auction item body unstructured data to generate unstructured mapping feature data of the auction item; Wherein, step S13 comprises the following steps: Step S131: performing unstructured type classification processing on the auction item body unstructured data to generate classified auction item body unstructured data; Step S132: performing convolutional layer analysis and processing of multi-modal unstructured mapping features of auction items based on the classified auction item body unstructured data to generate a convolutional layer of multi-modal mapping features of auction items; Step S133: designing a multi-modal mapping feature convolutional neural network model based on the multi-modal mapping feature convolutional layer of the auction items; Step S134: performing preliminary unstructured item mapping feature processing on the unstructured item data according to the multimodal item mapping feature convolutional neural network model to generate preliminary unstructured item mapping feature data; and performing feature full-connection mapping optimization processing on the preliminary unstructured item mapping feature data to generate unstructured item mapping feature data; Step S14: performing auction item ontology feature analysis based on the auction item ontology structured data and the auction item unstructured mapping feature data to generate auction item ontology feature data; Step S2: performing a screening process on the item body gain feature according to the item body feature data to generate the item body gain feature data; Wherein, step S2 comprises the following steps: Step S21: performing feature discretization processing on the auction item feature data to generate auction item feature discretization data; Step S22: performing conditional entropy analysis of the item's main features based on the discretized data of the item's main features to generate conditional entropy data of the item's main features; Wherein, step S22 includes the following steps: Step S221: performing frequency domain data analysis on the discretized data of the auction item's main features to generate frequency domain data of the auction item's main features; Step S222: performing an entropy analysis on the item body feature label according to the frequency domain data of the item body feature to generate the item body feature label entropy data; Step S223: performing an item body feature conditional entropy analysis on the item body feature frequency domain data and the item body feature label entropy data to generate item body feature conditional entropy data; Step S23: performing auction item body gain feature screening processing on the auction item body feature data according to the auction item body feature condition entropy data to generate auction item body gain feature data; Step S3: establishing an item ontology attribute matrix set through the item ontology gain feature data; performing a dynamic association network analysis of the item ontology based on the item ontology attribute matrix set to generate a dynamic item ontology association network; Step S4: Detecting user interaction data in the online commodity auction platform; performing user interaction behavior feature analysis and processing based on the user interaction data to generate user interaction behavior feature data; performing interaction behavior feature analysis and processing of clustering user interaction behavior habit types based on the user interaction behavior feature data to generate clustered user interaction behavior feature data; Step S5: construct an optimized probability tree model for classifying the demand items of clustered users based on the clustered user interaction behavior feature data to generate an optimized probability tree model for the demand items of clustered users; transmit the user interaction behavior feature data to the optimized probability tree model for the demand items of clustered users to predict the probability of the real-time user demand items and generate the real-time user demand item probability prediction data; transmit the real-time user demand item probability prediction data to the dynamic item ontology association network to perform intelligent analysis of the real-time user associated demand items and generate the real-time user associated demand item prediction data; and transmit the real-time user associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand items.

2. The intelligent demand forecasting and analysis method for commodity auction according to claim 1 is characterized in that: Step S3 includes the following steps: Step S31: designing and classifying item ontology attribute matrix nodes according to each gain feature type in the item ontology gain feature data, and transmitting the item ontology gain feature data to the item ontology attribute matrix nodes to establish an item ontology attribute matrix set; Step S32: performing association aggregation analysis on the auction item ontology attribute matrix based on the auction item ontology attribute matrix set to generate an auction item matrix association aggregation coefficient; Step S33: Designing an item ontology association network through the item ontology attribute matrix set, and dynamically updating the item ontology association degree of the item ontology association network through the matrix association aggregation coefficient as the item association attention weight parameter to generate a dynamic item ontology association network.

3. The intelligent demand forecasting and analysis method for commodity auction according to claim 2 is characterized in that: Step S32 includes the following steps: Step S321: using matrix nodes of various types in the auction item ontology attribute matrix set as first auction item ontology attribute data, and randomly selecting matrix nodes of the remaining auction item ontology attribute matrix set in the type corresponding to the first auction item ontology attribute node data as second auction item ontology attribute node data; Step S322: performing node similarity calculation on the first lot body attribute node data and the second lot body attribute node data to generate node similarity data; Step S323: Calculate the node access overlap of the first lot body attribute node data and the second lot body attribute node data to generate node access overlap data; Step S324: performing association aggregation analysis on the auction item ontology attribute matrix set according to the node similarity data and the node access overlap data, and generating an auction item matrix association aggregation coefficient.

4. The intelligent demand forecasting and analysis method for commodity auction according to claim 1 is characterized in that: Step S4 includes the following steps: Step S41: Detecting user interaction data in the online commodity auction platform; Step S42: performing user interaction behavior feature analysis and processing according to the user interaction data to generate user interaction behavior feature data; Step S43: performing user interaction behavior habit cluster analysis based on the user interaction behavior feature data to generate user interaction behavior habit cluster data; Step S44: performing data division processing on the user interaction behavior characteristic data into clustered user interaction behavior habit types according to the user interaction behavior habit clustering data to generate clustered user interaction behavior characteristic data.

5. The intelligent demand forecasting and analysis method for commodity auction according to claim 1 is characterized in that: Step S5 includes the following steps: Step S51: performing clustered user demand auction item conversion feature analysis based on clustered user interaction behavior feature data to generate clustered user demand auction item conversion feature data; Step S52: using clustered user interaction behavior feature data as clustered user demand item probability factors; establishing clustered user demand item classification probability mapping relationships for clustered user demand item probability factors based on a preset decision tree classification algorithm to obtain a clustered user demand item probability tree model; optimizing and adjusting the tree node probability coefficients of the clustered user demand item probability tree model through clustered user demand item conversion feature data to generate an optimized clustered user demand item probability tree model; Step S53: transmitting the user interaction behavior feature data to the optimized clustering user demand item probability tree model to perform instant user demand item probability prediction, and generate instant user demand item probability prediction data; Step S54: Transmit the probability prediction data of the real-time user demand item to the dynamic item ontology association network for intelligent analysis of the real-time user-associated demand item, generate the real-time user-associated demand item prediction data, and transmit the real-time user-associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand item.

6. An intelligent demand forecasting and analysis system for commodity auctions, characterized in that: The intelligent demand forecasting and analysis method for commodity auction according to claim 1 is used to execute the intelligent demand forecasting and analysis system for commodity auction, comprising: The item body feature analysis module is used to obtain the item body data of the online commodity auction platform; perform item body feature analysis based on the item body data to generate item body feature data; A gain feature screening module is used to perform a screening process on the gain features of the auction item itself according to the auction item itself feature data to generate the auction item itself gain feature data; A dynamic auction item ontology association module is used to establish an auction item ontology attribute matrix set through the auction item ontology gain feature data; perform a dynamic association network analysis of the auction item ontology based on the auction item ontology attribute matrix set to generate a dynamic auction item ontology association network; The clustering user interaction behavior analysis module is used to detect user interaction data in the online commodity auction platform; perform user interaction behavior feature analysis and processing based on the user interaction data to generate user interaction behavior feature data; perform clustering user interaction behavior habit type interaction behavior feature analysis and processing based on the user interaction behavior feature data to generate clustered user interaction behavior feature data; The instant user-associated demand item prediction module is used to construct an optimized demand item classification probability tree model for clustered users based on the clustered user interaction behavior feature data, and generate an optimized clustered user demand item probability tree model; transmit the user interaction behavior feature data to the optimized clustered user demand item probability tree model to perform instant user demand item probability prediction, and generate instant user demand item probability prediction data; transmit the instant user demand item probability prediction data to the dynamic item ontology association network to perform instant user-associated demand item intelligent analysis, generate instant user-associated demand item prediction data, and transmit the instant user-associated demand item prediction data to the terminal to perform the visual feedback operation of the intelligent prediction of the demand item.

Images