User big data mining method and system based on supply chain finance

Abstract

The present application relates to the field of data processing technology, and provides a user big data mining method and system based on supply chain finance. By mining and updating session vectors step by step, the embodiment of the present application can more deeply understand user behavior and accurately identify the user's operation preferences under different page interaction tasks. This method not only improves the analysis accuracy of user data, but also provides enterprises with a more scientific decision-making basis, thereby achieving efficient operation of supply chain finance business and significantly improving user satisfaction.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and specifically, to a user big data mining method and system based on supply chain finance. Background Art

[0002] In the field of supply chain finance, with the rapid development of information technology, a large amount of user behavior data has been recorded, providing companies with the opportunity to gain a deeper understanding of user needs and behavior patterns. However, how to effectively analyze and utilize this data to accurately capture user preferences and meet their personalized needs has always been a challenge facing the industry.

[0003] Traditional user behavior analysis methods can only provide a rough user portrait, and it is difficult to accurately describe the specific preferences of users in different scenarios and tasks. This leads to a lack of precision in the delivery of personalized content by enterprises, and an inability to effectively improve user experience and conversion rates. Summary of the invention

[0004] In order to improve the above problems, the present application provides a user big data mining method and system based on supply chain finance.

[0005] In a first aspect of an embodiment of the present application, a user big data mining method based on supply chain finance is provided, which is applied to a user big data mining system, and the method includes:

[0006] In the process of mining the x-th order conversation dense vector, the r-th order supply chain finance business conversation vector is feature updated to obtain a user behavior conversation vector, wherein the r-th order supply chain finance business conversation vector is the result of mining the r-th order conversation dense vector on the supply chain finance business log, and different user behavior conversation vectors are conversation vectors of the r-th order supply chain finance business conversation vector under different page interaction tasks;

[0007] Performing user preference dense vector mining on each of the user behavior session vectors to obtain operation preference dense vectors under different page interaction tasks;

[0008] Performing weighted integration on the operation preference dense vectors under different page interaction tasks to obtain an x-th order supply chain finance business session vector;

[0009] The user push demand mining is performed based on the y-th order supply chain finance business conversation vector obtained by mining the y-th order conversation dense vector to obtain the user push demand mining result of the supply chain finance business log, where y is not less than x.

[0010] Optionally, performing user preference dense vector mining on each of the user behavior session vectors to obtain operation preference dense vectors under different page interaction tasks includes:

[0011] Performing time series session dense vector mining on the user behavior session vector to obtain a time series session dense vector;

[0012] The user preference dense vector is mined on the time series session dense vector based on the primary page interaction task and the secondary page interaction task to obtain the operation preference dense vector. The primary page interaction task is the page interaction task corresponding to the user behavior session vector, and the secondary page interaction task is a page interaction task in the dynamic page interaction task that is different from the primary page interaction task.

[0013] Optionally, the performing time series session dense vector mining on the user behavior session vector to obtain the time series session dense vector includes:

[0014] Decomposing the user behavior conversation vector based on feature cores to obtain user behavior conversation sub-vectors corresponding to P feature cores, wherein the P feature cores do not overlap;

[0015] Performing session dense vector mining on the P user behavior session sub-vectors to obtain a local time series session dense vector;

[0016] The P feature cores are updated based on the feature cores, and session dense vector mining is performed on the local time series session dense vectors corresponding to the P feature cores after the feature core update to obtain the time series session dense vectors. The feature core-based update is used to adjust the distribution labels of the P feature cores.

[0017] Optionally, performing session dense vector mining on the P user behavior session sub-vectors to obtain a local time series session dense vector includes:

[0018] Performing a feature focusing operation on the P user behavior conversation sub-vectors to obtain behavior conversation focus vectors of the P user behavior conversation sub-vectors;

[0019] Performing weighted integration on the behavior conversation attention vectors of the P user behavior conversation sub-vectors to obtain a first behavior conversation weighted vector;

[0020] Performing residual processing on the first behavior session weighted vector to obtain the first behavior session residual connection vector;

[0021] A decision tree network is used to perform session dense vector mining on the first behavior as the session residual connection vector to obtain the local time series session dense vector.

[0022] Optionally, performing session dense vector mining on the local time series session dense vectors corresponding to the P feature cores updated based on the feature core to obtain the time series session dense vector includes:

[0023] Performing a feature focusing operation on the local temporal conversation dense vectors corresponding to the P feature cores updated based on the feature core to obtain the behavior conversation attention vectors corresponding to the P feature cores;

[0024] Performing weighted integration on the P behavior session attention vectors to obtain a second behavior session weighted vector;

[0025] Performing feature mapping on the second behavior session weighted vector, and performing residual processing on the feature-mapped second behavior session weighted vector to obtain a second behavior session residual connection vector;

[0026] A decision tree network is used to perform session dense vector mining on the second behavior as the session residual connection vector to obtain the time series session dense vector.

[0027] Optionally, performing a feature focusing operation on the P user behavior conversation sub-vectors to obtain the behavior conversation focus vectors of the P user behavior conversation sub-vectors includes:

[0028] A feature focusing operation is performed based on the index features, identification features and attribute features corresponding to the user behavior conversation vector to obtain P behavior conversation attention vectors of the local time series conversation dense vector.

[0029] Optionally, the session dense vector mining process includes a vector extraction process and a vector optimization process. The identification eigenvalues ​​and attribute eigenvalues ​​in the z-th order vector optimization process are obtained by fusing the identification eigenvalues ​​and attribute eigenvalues ​​in the z-1-th order vector optimization with the identification eigenvalues ​​and attribute eigenvalues ​​in the corresponding vector extraction process. The index eigenvalues ​​in the y-th order vector optimization process are the index eigenvalues ​​in the z-1-th order vector optimization.

[0030] Optionally, the performing user preference dense vector mining on the time series session dense vector based on the primary page interaction task and the secondary page interaction task to obtain the operation preference dense vector includes:

[0031] Performing weighted integration on the time series session dense vector and the distribution representation vector to obtain a first operation preference dense vector, wherein the distribution representation vector is used to indicate a location label corresponding to the user behavior session vector;

[0032] Under the primary page interaction task, using a decision tree network to perform user preference dense vector mining on the first operation preference dense vector to obtain a key operation preference dense vector;

[0033] Under the secondary page interaction task, using the decision tree network to perform user preference dense vector mining on the first operation preference dense vector to obtain a secondary key operation preference dense vector;

[0034] The key operation preference dense vector and the sub-key operation preference dense vector are weightedly integrated to obtain the operation preference dense vector.

[0035] Optionally, performing weighted integration on the operation preference dense vectors under different page interaction tasks to obtain an x-th order supply chain finance business session vector includes:

[0036] The operation preference dense vector is fused with the page interaction task vector to obtain an interaction task directional preference fusion viewpoint;

[0037] The directional preference fusion viewpoints of each interactive task are weighted and integrated to obtain the x-th order supply chain finance business conversation vector.

[0038] Optionally, performing user preference dense vector mining on each of the user behavior session vectors to obtain operation preference dense vectors under different page interaction tasks includes:

[0039] The user preference dense vector mining is performed on the user behavior session vector under each page interaction task by using the session dense vector mining model corresponding to the same model variable, so as to obtain the operation preference dense vector under different page interaction tasks.

[0040] Optionally, the session dense vector mining process includes a vector extraction process and a vector optimization process, wherein the vector extraction process includes a feature compression process for the supply chain finance business session vector, and the vector optimization process includes a feature derivation process for the supply chain finance business session vector;

[0041] The user push demand mining is performed based on the y-th order supply chain finance business session vector obtained by mining the y-th order session dense vector to obtain the user push demand mining result of the supply chain finance business log, including:

[0042] On the basis that the feature derivation reaches the initial dimension, the extracted supply chain finance business conversation vector is determined as the y-th order supply chain finance business conversation vector obtained by mining the y-th order conversation dense vector;

[0043] User push demand mining is performed based on the y-th order supply chain financial business session vector to obtain the user push demand mining result.

[0044] Optionally, the supply chain finance business log is an e-commerce business log, a blockchain transaction business log, or an Internet credit business log.

[0045] A second aspect of an embodiment of the present application provides a user big data mining system, comprising: a processor, a memory and a bus connected to the processor; the processor and the memory communicate with each other through the bus; the processor is used to call a computer program in the memory to execute the above-mentioned user big data mining method based on supply chain finance.

[0046] According to a third aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which a program is stored, and when the program is executed by a processor, the above-mentioned user big data mining method based on supply chain finance is implemented.

[0047] The user big data mining method and system based on supply chain finance provided in the embodiment of the present application can more deeply understand user behavior and accurately identify user operation preferences under different page interaction tasks by mining and updating session vectors step by step. This method not only improves the analysis accuracy of user data, but also provides enterprises with a more scientific decision-making basis, thereby achieving efficient operation of supply chain finance business and significantly improving user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 A flowchart of a user big data mining method based on supply chain finance provided in an embodiment of the present application.

[0050] Figure 2 A schematic diagram of a product module of a user big data mining system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0051] The exemplary embodiments disclosed in the present application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0052] In order to better understand the above technical scheme, the technical scheme of the present application is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, rather than limitations on the technical scheme of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0053] See also Figure 1 , which is a flowchart of a user big data mining method based on supply chain finance provided in an embodiment of the present application. The method is applied to a user big data mining system, and the specific content description of the method includes S110-S140.

[0054] S110. During the x-th order session dense vector mining process, the r-th order supply chain finance business session vector is feature updated to obtain a user behavior session vector. The r-th order supply chain finance business session vector is the result of performing r-th order session dense vector mining on the supply chain finance business log, and different user behavior session vectors are session vectors of the r-th order supply chain finance business session vector under different page interaction tasks.

[0055] Where r=x-1.

[0056] S120: Perform user preference dense vector mining on each of the user behavior session vectors to obtain operation preference dense vectors under different page interaction tasks.

[0057] S130. Perform weighted integration on the operation preference dense vectors under different page interaction tasks to obtain an x-th order supply chain finance business session vector.

[0058] S140, performing user push demand mining based on the y-th order supply chain finance business session vector obtained by mining the y-th order session dense vector, and obtaining the user push demand mining result of the supply chain finance business log, where y is not less than x.

[0059] In an exemplary supply chain finance business scenario, the user big data mining system plays a vital role. It can provide enterprises with accurate user preference and user demand information through in-depth analysis of user behavior, thereby optimizing service quality and improving user experience. The following is a detailed introduction to how the system performs a series of data mining processes.

[0060] During the interaction between the user and the supply chain finance platform, the system records a series of user behavior logs, which become the raw data for analysis by the user big data mining system. First, in S110, the user big data mining system updates the features of the r-th order supply chain finance business session vector. These session vectors are the results obtained by mining the r-th order session dense vector of the supply chain finance business log, reflecting the user's behavior pattern on the platform. The system updates these session vectors based on the user's behavior under different page interaction tasks, thereby obtaining user behavior session vectors. These user behavior session vectors more finely characterize the user's behavior characteristics under different page interaction tasks.

[0061] For example, session dense vector mining is a data analysis technique that is mainly used to extract meaningful information from the interaction session between the user and the system. In this process, each user session is converted into a vector in a high-dimensional space, where each dimension may represent an aspect of user behavior, such as clicks, browsing, purchases, etc. These vectors are dense and rich in information, so they are called "dense vectors." Mining these vectors can help understand user behavior patterns and preferences.

[0062] When users interact with the supply chain finance platform, their behaviors (such as browsing products, placing orders, checking account information, etc.) are recorded and converted into a series of data points. These data points are further processed into session vectors (supply chain finance business session vectors), which represent the user's behavioral characteristics in a specific supply chain finance business scenario.

[0063] As users continue to interact with the platform, their behavior patterns may change. Feature updating is a process where old session vectors are adjusted and optimized based on the user's latest behavior data to more accurately reflect the user's current behavior and preferences.

[0064] User behavior session vectors are vectors obtained through feature updating, which more accurately represent the user's behavior in a specific session. These vectors not only take into account the user's historical behavior, but also incorporate the latest interaction data, thus providing richer details for user portraits.

[0065] Supply chain finance business logs are detailed log files that record all user activities on the supply chain finance platform. They contain detailed information such as the timestamp of user behavior, behavior type (such as click, browse, purchase, etc.), and behavior object (such as specific goods, pages, etc.). These logs are the basis for subsequent data analysis and mining.

[0066] Page interaction tasks refer to the tasks or goals that users need to complete when interacting with each page on the platform. For example, users may browse, add to cart, or buy immediately on a product detail page. Each page interaction task generates corresponding user behavior data, which is crucial for understanding user needs and behaviors.

[0067] In detail, in S110, the user big data mining system enters a key data processing stage - the x-th order conversation dense vector mining. The core goal of this step is to conduct an in-depth analysis of the user's behavior on the supply chain finance platform and update the user's behavioral features based on these behavioral data. First, the system reviews and analyzes the r-th order supply chain finance business conversation vectors. These vectors were previously extracted from the supply chain finance business logs through the conversation dense vector mining technology, and they have captured the user's preliminary behavioral patterns on the platform. However, the user's behavior is dynamically changing, so these old conversation vectors need to be continuously updated and optimized. In order to update the features, the system collects the latest behavioral data generated by the user in each page interaction task. These page interaction tasks can include browsing products, searching for specific items, adding products to the shopping cart, and making settlements. Every operation of the user is accurately recorded and converted into data points, which are then integrated into the existing conversation vectors. During the integration process, the system not only adds new data points, but also adjusts the weights of old data points based on the user's latest behavior. For example, if a user has recently browsed a certain category of goods frequently, the weight of the data points related to this category of goods in the conversation vector will increase. This dynamic adjustment ensures that the session vector can always accurately reflect the user's current behavior and preferences. Finally, through feature updates, the system obtains user behavior session vectors. These vectors are more accurate and detailed than the previous supply chain finance business session vectors because they not only contain the user's historical behavior data, but also incorporate the user's recent interaction information. These user behavior session vectors provide a solid foundation for subsequent user preference mining and personalized service push.

[0068] Next, in S120, the user big data mining system mines user preference dense vectors for each user behavior session vector. The purpose of this step is to deeply understand the user's operation preferences, such as click habits, search habits, and purchase habits when browsing products. Through this mining, the system can obtain dense vectors of operation preferences under different page interaction tasks, which reflect the user's personalized needs in detail.

[0069] For example, the dense vector of operation preference is an important concept in the user big data mining system. It refers to a high-dimensional data representation obtained by deeply mining and analyzing the behavioral data of users in specific page interaction tasks. This vector not only contains the user's preference information for different operations (such as clicking, browsing, purchasing, etc.), but also reflects the behavioral patterns and habits of users when performing these operations. The "density" of the dense vector of operation preference is reflected in that it captures many details of user behavior, making the description of user preferences more comprehensive and accurate. In the process of user interaction with the supply chain finance platform, the system records each user's operation, including operation type, operation object, operation time, etc., and converts this information into numerical form. These numerical values ​​are further combined into a high-dimensional vector, in which each dimension represents an aspect of user behavior. For example, one dimension may reflect the user's click frequency on a certain type of product, while another dimension may represent the user's stay time when browsing the product details page. Through the combination of these dimensions, the dense vector of operation preference can meticulously characterize the user's operation preferences and behavioral characteristics in page interaction tasks.

[0070] In detail, in S120, the user big data mining system further explores the user's behavior patterns, especially their operation preferences under different page interaction tasks. The core of this step is to mine the user preference dense vector of the user behavior session vector, and the goal is to obtain the operation preference dense vector under different page interaction tasks. First, the system analyzes the user behavior session vector obtained in S110. These vectors have preliminarily captured the user's behavior characteristics under different page interaction tasks. However, in order to understand the user's operation preferences more accurately, the system needs to further mine these data. During the mining process, the system focuses on the user's specific operations on each page, such as clicks, scrolling, dwell time, etc., as well as the correlation and sequence between these operations. For example, when browsing products, users may first view product images, then read product descriptions, and finally view user reviews. This series of operation sequences and frequencies reflect the user's operation preferences. Through in-depth analysis of user behavior session vectors, the system can extract the user's operation preferences under each page interaction task and convert these preference information into operation preference dense vectors. These vectors not only contain the user's preference for specific operations, but also reflect the user's behavior patterns and habits when performing these operations. Finally, the system obtains dense vectors of operation preferences under different page interaction tasks. These vectors provide important data support for the subsequent construction of user behavior profiles and personalized service push. For example, if it is found that users prefer to use keyword search rather than category browsing when searching for products, the system can consider increasing the weight of keyword search and intelligent recommendation functions in the subsequent search function optimization.

[0071] Entering S130, the user big data mining system performs weighted integration on the dense vectors of operation preferences under different page interaction tasks. This step is to integrate the user preference information collected in different pages and tasks to form a comprehensive and unified user behavior portrait. Through this integration, the system obtains the x-th order supply chain finance business conversation vector, which is a more comprehensive and in-depth description of user behavior.

[0072] For example, weighted integration is a data processing and analysis method that refers to combining multiple data sources or data features according to certain weights to form a more comprehensive and representative data set or feature representation. In this process, "weighting" means that different data sources or features are not simply added or averaged when integrated, but are given different weights based on their importance or information content. These weights may be determined based on a variety of factors such as data reliability, relevance, information content, historical performance, etc. In user big data mining systems, weighted integration is widely used to integrate user behavior data from different channels, different time periods or different dimensions. Through weighted integration, the system can comprehensively consider various factors to obtain a more accurate and comprehensive user behavior portrait, thereby providing a more solid foundation for subsequent data analysis and decision-making.

[0073] In detail, in S130, the user big data mining system performs a key data integration process, namely, weighted integration of the dense vectors of operation preferences under different page interaction tasks. The goal of this step is to effectively integrate the various user operation preference information mined in the previous steps to form a unified and comprehensive description of user behavior. First, the system reviews the dense vectors of operation preferences corresponding to each user behavior session vector obtained in S120. These vectors capture the preference information of users under different page interaction tasks such as browsing, searching, and purchasing. However, since the importance and influence of users' behaviors in different tasks may be different, the system needs to balance these differences through weighted integration. In the process of weighted integration, the system assigns a weight to each dense vector of operation preference. This weight may be determined based on a variety of factors, such as the frequency of the task in user behavior, the influence on user decision-making, and the quality and reliability of the data. For example, if the user frequently performs search operations and the search behavior has a significant impact on the user's final purchase decision, the dense vector of operation preference under the search task may be given a higher weight. Next, the system weights and sums the dense vectors of each operation preference according to these weights to obtain a comprehensive description of user behavior. This process ensures that user preferences under each task are properly considered, while also highlighting those tasks that have a greater impact on user behavior. Finally, the x-th order supply chain finance business session vector obtained through weighted integration is a comprehensive representation of user preferences under multiple page interaction tasks. This vector not only reflects the overall behavior pattern of users, but also provides strong data support for subsequent user push demand mining. Through this integrated approach, the user big data mining system can understand user needs more accurately, thereby providing more accurate and personalized services.

[0074] Finally, in S140, the user big data mining system mines user push demand based on the y-th order supply chain financial business conversation vector obtained by mining the y-th order conversation dense vector. This step is to convert the user behavior characteristics obtained from the previous analysis into actual user needs, and to push accurate content accordingly. For example, if a user frequently browses a certain category of goods, the system will push new products or promotional information of this category of goods to the user. Here, y is not less than x, which means that when the system mines push demand, it may consider higher-order conversation vectors than user behavior conversation vectors to ensure the accuracy and pertinence of the push.

[0075] For example, user push demand mining refers to identifying and predicting users' potential needs and interests for specific content or services by analyzing users' behavioral data on digital platforms in order to push accurate information. In the user big data mining system, this link may involve in-depth analysis of user behavior data, including but not limited to browsing history, purchase records, search queries, click behaviors, etc., to reveal user preferences and predict their future information needs based on this. In this way, the system can provide users with more personalized content recommendations that meet their interests and needs, thereby enhancing the user experience. The results of user push demand mining are specific conclusions or data outputs obtained through the above mining process. These results may include users' preference for different types of content, potential purchase intentions, and the categories of products or services that are most likely to be of interest. These data can be presented in the form of lists, charts, or reports for reference by corporate decision makers. Based on these results, companies can adjust their marketing strategies, optimize product recommendation algorithms, or improve the user interface of websites and applications to better meet user needs and improve user satisfaction and loyalty.

[0076] In detail, in S140, the user big data mining system uses the session dense vector obtained in the previous step to mine user push demand. This step is a key link in the entire data mining process, which is directly related to whether the enterprise can accurately capture user needs and provide personalized service push accordingly. First, the system reviews and analyzes the supply chain financial business session vectors obtained by mining the y-th order session dense vector. These session vectors integrate the behavioral characteristics of users under multiple page interaction tasks and are a comprehensive and in-depth description of user behavior. Since y is not less than x, this means that when the system mines user push demand, it may consider more complex user behavior patterns than the x-th order, thereby ensuring the accuracy and foresight of the mining results. Next, the system identifies the user's potential needs and interests based on these session vectors. For example, by analyzing the user's browsing and purchasing behavior in different time periods, the system may find that the user's interest in a certain type of goods or services is rising, or that the user has a significant response to certain promotional activities. These findings are important bases for mining user push demand. Then, the system predicts its future information or product needs based on the mined user needs. This process can involve complex algorithms and models, such as machine learning or deep learning models, which can predict future trends and behaviors based on historical data. Finally, the system integrates all the mining results into a user push demand mining report or data set for the company's marketing team or decision makers. These results can not only help companies develop more accurate marketing strategies, but also optimize product recommendation systems and improve user experience and satisfaction. It can be seen that step S140 is a crucial part of the user big data mining system. It provides companies with the opportunity to stand out in the fierce market competition by converting users' complex behavioral data into valuable business insights.

[0077] Through the above process, the user big data mining system not only deeply analyzes the behavioral characteristics of users on the supply chain finance platform, but also provides users with personalized service push based on these characteristics. This not only improves the user experience, but also brings higher user stickiness and conversion rate to enterprises.

[0078] The embodiment of the present application updates the features of the r-th order supply chain finance business conversation vector in the process of mining the x-th order conversation dense vector, and finely captures the behavioral features of users under different page interaction tasks, generating a more representative user behavior conversation vector. This improvement not only deepens the understanding of user behavior, but also provides a more accurate data basis for subsequent user preference analysis.

[0079] Furthermore, by mining the dense vector of user preferences for user behavior session vectors, the embodiment of the present application can accurately identify the user's operation preferences under different page interaction tasks. This fine mining of preference information helps enterprises understand user needs more comprehensively and provides strong support for personalized services.

[0080] In addition, the embodiment of the present application also adopts a weighted integration method to comprehensively process the dense vectors of operation preferences under different page interaction tasks, and obtain a more comprehensive and accurate x-order supply chain financial business session vector. This step enhances the integration and availability of user behavior data, making the decisions based on these data more scientific and effective.

[0081] Finally, by mining user push demand based on high-order conversation dense vectors, the embodiment of the present application can accurately predict and meet the personalized needs of users, and improve the pertinence and effectiveness of user push content. This push strategy oriented to user demand not only improves user experience, but also brings higher user stickiness and conversion rate to enterprises, thereby achieving a dual improvement in supply chain financial business efficiency and user satisfaction.

[0082] Under some preferred technical ideas, the user preference dense vector mining is performed on each of the user behavior session vectors to obtain the operation preference dense vectors under different page interaction tasks, including: performing time series session dense vector mining on the user behavior session vector to obtain the time series session dense vector; performing user preference dense vector mining on the time series session dense vector based on the primary page interaction task and the secondary page interaction task to obtain the operation preference dense vector, the primary page interaction task is the page interaction task corresponding to the user behavior session vector, and the secondary page interaction task is a page interaction task in the dynamic page interaction task that is different from the primary page interaction task.

[0083] Based on this technical idea, the user big data mining system performs a key step: in-depth user preference dense vector mining of user behavior session vectors. The purpose of this process is to more accurately understand the user's operating habits and preferences under different page interaction tasks.

[0084] First, the system mines the user behavior conversation vector into a time series conversation dense vector. This step is mainly to capture the user's behavior pattern in the time series. By analyzing the user's behavior data at each time point, the system can generate a time series conversation dense vector. This vector not only reflects the sequence of user behavior, but also includes the correlation and conversion probability between behaviors, thus providing richer information for mining user preferences.

[0085] Next, the system further mines user preference dense vectors based on the primary page interaction tasks and secondary page interaction tasks. Here, the primary page interaction tasks refer to the page interaction tasks that directly correspond to the user behavior session vector, such as the user's behavior when browsing the product details page. The secondary page interaction tasks refer to other tasks in the dynamic page interaction tasks that are different from the primary page interaction tasks, such as the user's behavior of jumping to the shopping cart or payment page after browsing the product details page.

[0086] By performing this hierarchical mining of time-series session dense vectors, the system can more finely identify users’ operation preferences under different page interaction tasks. For example, the system may find that some users prefer to buy directly after browsing the product details page, while other users prefer to add products to the shopping cart before checking out. These findings are crucial for the formulation of personalized recommendations and marketing strategies.

[0087] Finally, through this series of mining steps, the user big data mining system can output dense vectors of operation preferences under different page interaction tasks. These vectors not only reveal the personalized needs and behavior patterns of users, but also provide an important basis for enterprises to optimize service processes and improve user experience.

[0088] In this way, by mining user behavior session vectors in time series and mining hierarchical user preferences, the user big data mining system can more accurately capture and predict user operation preferences under different page interaction tasks. This refined user behavior analysis not only helps improve the accuracy of personalized recommendations, but also helps companies develop more targeted marketing strategies, thereby improving user satisfaction and loyalty, and achieving a dual improvement in business efficiency and user experience.

[0089] Under some preferred technical ideas, the user behavior conversation vector is subjected to time series conversation dense vector mining to obtain a time series conversation dense vector, including: feature core-based decomposition of the user behavior conversation vector to obtain user behavior conversation sub-vectors corresponding to P feature cores, wherein the P feature cores do not overlap; conversation dense vector mining is performed on the P user behavior conversation sub-vectors to obtain a local time series conversation dense vector; feature core-based updating is performed on the P feature cores, and conversation dense vector mining is performed on the local time series conversation dense vectors corresponding to the P feature cores after the feature core-based update to obtain a time series conversation dense vector, wherein the feature core-based update is used to adjust the distribution labels of the P feature cores.

[0090] Based on this design idea, the user big data mining system performs a complex process of mining the time series conversation dense vector of the user behavior conversation vector to obtain the time series conversation dense vector. The following is a detailed example explanation of this process.

[0091] First, the user big data mining system decomposes the user behavior conversation vector based on feature cores. Feature cores are the basic units that constitute the user behavior conversation vector, and each feature core corresponds to a specific aspect of user behavior. The system separates these feature cores through decomposition and obtains P user behavior conversation sub-vectors corresponding to feature cores without overlap. This means that each sub-vector focuses on describing a unique dimension of user behavior, such as browsing time, click frequency, etc., thus avoiding information redundancy and confusion.

[0092] Next, the system performs session dense vector mining on each of the P user behavior session sub-vectors. The purpose of this step is to deeply explore the user behavior features represented by each sub-vector and capture the changes of these features in the time series. Through mining, the system can generate local time series session dense vectors, which reflect the dynamic evolution of each feature core on the time axis, providing a rich data foundation for subsequent analysis.

[0093] Then, the system performs an update based on the feature core. The purpose of this step is to adjust the distribution labels of the P feature cores to ensure that they can more accurately reflect the actual situation of user behavior. Through the update, the system can optimize the division and definition of the feature cores so that each feature core can better capture a specific aspect of user behavior. This update is based on the previous mining results, so it has the characteristics of iterative optimization and can continuously improve the accuracy and effectiveness of the feature cores.

[0094] Finally, the system performs session dense vector mining again on the local time series session dense vectors corresponding to the P updated feature kernels based on the feature kernel. This mining is performed on the basis of the optimized feature kernel, so it can more accurately reveal the global patterns and trends of user behavior in the time series. Through this step, the system finally obtains time series session dense vectors, which not only contain rich details of user behavior, but also reflect the dynamic changes of user behavior over time.

[0095] It can be seen that the user big data mining system successfully captures the precise patterns and dynamic changes of user behavior in time series through a series of complex steps such as feature core decomposition, local time series session dense vector mining, feature core update, and global time series session dense vector mining. This in-depth user behavior analysis provides enterprises with valuable insights and decision support, helps improve the accuracy and timeliness of personalized services, and thus significantly improves user satisfaction and business efficiency.

[0096] Under some preferred technical ideas, the performing session dense vector mining on the P user behavior conversation sub-vectors to obtain a local time series conversation dense vector includes: performing feature focusing operation on the P user behavior conversation sub-vectors to obtain the behavior conversation attention vectors of the P user behavior conversation sub-vectors; performing weighted integration on the behavior conversation attention vectors of the P user behavior conversation sub-vectors to obtain a first behavior conversation weighted vector; performing residual processing on the first behavior conversation weighted vector to obtain a first behavior conversation residual connection vector; and performing session dense vector mining on the first behavior conversation residual connection vector using a decision tree network to obtain the local time series conversation dense vector.

[0097] In the embodiment of the present application, the user big data mining system performs refined conversation dense vector mining on P user behavior conversation sub-vectors to obtain a local time series conversation dense vector. The following is a detailed example explanation of the process.

[0098] First, the system performs feature focusing on the P user behavior conversation sub-vectors. The purpose of this step is to highlight the key features of each sub-vector, that is, those features that have the greatest impact on user behavior. Through feature focusing, the system can generate behavior conversation attention vectors for the P user behavior conversation sub-vectors. These attention vectors not only retain the important information of the original sub-vectors, but also further emphasize the key features, providing a more accurate data basis for subsequent analysis.

[0099] Next, the system performs weighted integration on the P behavior conversation attention vectors. In this process, the system assigns corresponding weights to each attention vector based on its importance and information content. Through weighted integration, the system can comprehensively consider the contribution of each attention vector and obtain a more comprehensive and accurate first behavior conversation weighted vector. This weighted vector integrates the information of multiple attention vectors, providing richer data input for subsequent mining.

[0100] Then, the system performs residual processing on the first-line conversation weighted vector. The purpose of residual processing is to eliminate redundant information and noise that may exist in the weighted vector and further purify the data. Through residual processing, the system can obtain a cleaner and more accurate first-line conversation residual connection vector. This residual connection vector not only retains the main information of the weighted vector, but also removes the interference factors, providing more reliable data support for subsequent mining.

[0101] Finally, the system uses a decision tree network to mine the session dense vectors for the first line of the session residual connection vector. The decision tree network is a powerful machine learning model that can deeply mine the potential patterns and associations in the data. Through this step of mining, the system finally obtains the local time series session dense vectors. These vectors not only capture the time series characteristics of user behavior, but also reflect the user's behavior patterns and preferences at different time points.

[0102] Therefore, the user big data mining system successfully obtains local time series conversation dense vectors through a series of refined steps such as feature focus, weighted integration, residual processing, and decision tree network mining. These vectors not only provide strong support for in-depth analysis of user behavior, but also provide an important basis for enterprises to formulate personalized recommendations and marketing strategies. In this way, the system can more accurately understand user needs and behavior patterns, thereby improving user experience and business efficiency.

[0103] Under some preferred technical ideas, the local temporal conversation dense vectors corresponding to the P feature cores updated based on the feature core are subjected to conversation dense vector mining to obtain the temporal conversation dense vector, including: performing feature focusing operation on the local temporal conversation dense vectors corresponding to the P feature cores updated based on the feature core to obtain the behavior conversation attention vectors corresponding to the P feature cores; performing weighted integration on the P behavior conversation attention vectors to obtain a second behavior conversation weighted vector; performing feature mapping on the second behavior conversation weighted vector, and performing residual processing on the second behavior conversation weighted vector after feature mapping to obtain a second behavior conversation residual connection vector; performing conversation dense vector mining on the second behavior conversation residual connection vector using a decision tree network to obtain the temporal conversation dense vector.

[0104] Under some preferred technical ideas, the user big data mining system further mines the local time-series conversation dense vectors corresponding to the P feature cores updated based on the feature core to obtain the time-series conversation dense vectors. The following is a detailed example explanation of the process.

[0105] First, the system performs a feature focusing operation on the local time-series conversation dense vectors corresponding to the updated P feature cores. The purpose of this step is to highlight the key features in each local time-series conversation dense vector, that is, those features that have the greatest impact on the user's behavior conversation. Through feature focusing, the system can generate behavioral conversation attention vectors corresponding to the P feature cores. These attention vectors not only retain the important information of the original local time-series conversation dense vectors, but also further focus on the key features, providing a more accurate data basis for subsequent analysis.

[0106] Next, the system performs weighted integration on these P behavior conversation attention vectors. In this process, the system assigns corresponding weights to each attention vector based on its importance and information content. The purpose of weighted integration is to comprehensively consider the contribution of each attention vector, so as to obtain a more comprehensive and accurate second behavior conversation weighted vector. This weighted vector integrates the information of multiple attention vectors, providing richer data input for subsequent mining.

[0107] Then, the system performs feature mapping and residual processing on the second behavior conversation weighted vector. The purpose of feature mapping is to transform the weighted vector from the original feature space to a new feature space to better reveal the potential structure and association in the data. The purpose of residual processing is to eliminate the redundant information and noise that may be generated during the feature mapping process, so as to obtain a purer and more accurate second behavior conversation residual connection vector.

[0108] Finally, the system uses a decision tree network to mine the session dense vector for the second behavior session residual connection vector. The decision tree network is a powerful machine learning model that can deeply mine potential patterns and associations in data. Through this step of mining, the system finally obtains the time series session dense vector. These vectors not only capture the time series characteristics of user behavior, but also reflect the user's behavior patterns and preferences at different time points, providing enterprises with valuable insights into user behavior.

[0109] It can be seen that through a series of refined steps such as feature focusing, weighted integration, feature mapping, residual processing and decision tree network mining, the user big data mining system successfully obtained time series conversation dense vectors. These vectors provide enterprises with in-depth user behavior analysis, which helps to improve the accuracy and timeliness of personalized services, thereby significantly improving user satisfaction and business efficiency. At the same time, this mining method also has good scalability and flexibility, and can adapt to user behavior analysis tasks under different scenarios and needs.

[0110] Under some preferred technical ideas, the feature focusing operation is performed on the P user behavior conversation sub-vectors to obtain the behavior conversation attention vectors of the P user behavior conversation sub-vectors, including: performing feature focusing operation based on the index features, identification features and attribute features corresponding to the user behavior conversation vectors to obtain the behavior conversation attention vectors of the P local time series conversation dense vectors. In detail, the conversation dense vector mining process includes a vector refinement process and a vector optimization process. The identification feature value and attribute feature value in the z-th order vector optimization process are obtained by merging the identification feature value and attribute feature value in the z-1th order vector optimization with the identification feature value and attribute feature value in the corresponding vector refinement process. The index feature value in the y-th order vector optimization process is the index feature value in the z-1th order vector optimization.

[0111] Based on this design idea, the user big data mining system performs feature focusing operations on P user behavior conversation sub-vectors to obtain the behavior conversation attention vectors of these sub-vectors. The following is a detailed example explanation of this process.

[0112] The system first performs feature focusing based on multiple features of the user behavior session vector. These features include index features, identification features, and attribute features. Index features may refer to the time or sequence information of user behavior, identification features may represent the unique identifier of the user or behavior, and attribute features may describe the specific content or nature of the user behavior.

[0113] When performing feature focusing, the system weights or selectively emphasizes these features to highlight the most important or representative features of the user's behavior session. Through this step, the system is able to generate behavior session attention vectors for P user behavior session sub-vectors. These attention vectors are refined representations of the original sub-vectors, focusing more on the key features that reflect the user's behavior.

[0114] Next, the system conducts the session dense vector mining process, which includes two stages: vector extraction and vector optimization. In the vector extraction process, the system focuses on extracting useful information from the original data and forming preliminary session dense vectors. In the vector optimization process, the system is committed to improving and perfecting these vectors to make them more accurate and effective in representing user behavior sessions.

[0115] For example, during the z-th order vector optimization process, the system updates the identification eigenvalues ​​and attribute eigenvalues. These new eigenvalues ​​are obtained by fusing the eigenvalues ​​in the z-1-th order vector optimization with the eigenvalues ​​in the corresponding vector refinement process. This fusion may involve weighted averaging, maximum selection, or other data fusion techniques, aiming to combine information from different stages to obtain a more comprehensive feature representation.

[0116] It is worth noting that the index feature value in the z-th order vector optimization process remains unchanged and is directly inherited from the index feature value in the z-1-th order vector optimization. This is because the index feature can represent the order or time information of the behavior, which is stable during the optimization process and should not be changed.

[0117] Through the above process, the user big data mining system can gradually refine and optimize the vector representation of user behavior sessions, and finally obtain a time-series session dense vector that can accurately reflect user behavior patterns and preferences.

[0118] In this way, through feature focus operations and multi-stage vector extraction and optimization processes, the temporal characteristics and key information of user behavior can be captured more accurately. This not only helps companies to understand user needs and behavioral habits more deeply, but also provides strong data support for application scenarios such as personalized recommendations and precision marketing. At the same time, the flexibility and scalability of the solution also enable it to adapt to changes in demand in different industries and scenarios.

[0119] In some other exemplary embodiments, the user preference dense vector mining of the time series session dense vector based on the primary page interaction task and the secondary page interaction task to obtain the operation preference dense vector includes: weighted integration of the time series session dense vector and the distribution representation vector to obtain a first operation preference dense vector, and the distribution representation vector is used to indicate the location label corresponding to the user behavior session vector; under the primary page interaction task, using a decision tree network to mine the user preference dense vector of the first operation preference dense vector to obtain a key operation preference dense vector; under the secondary page interaction task, using the decision tree network to mine the user preference dense vector of the first operation preference dense vector to obtain a sub-key operation preference dense vector; weighted integration of the key operation preference dense vector and the sub-key operation preference dense vector to obtain the operation preference dense vector.

[0120] Based on this embodiment, the user big data mining system further utilizes the time series conversation dense vector to mine the user's operation preference dense vector.

[0121] First, the system performs weighted integration on the time-series session dense vector and the distribution representation vector to obtain the first operation preference dense vector. The distribution representation vector here is an important data element, which is used to indicate the location label corresponding to the user behavior session vector. The location label can be understood as the page or functional area where the user is located during the interaction process, such as the homepage of a shopping website, the product details page, etc. Through weighted integration, the system comprehensively considers the user behavior information in the time-series session dense vector and the location information in the distribution representation vector, forming a more comprehensive representation of user operation preferences.

[0122] Next, the system uses the decision tree network to mine the user preference dense vector of the first operation preference dense vector under the primary page interaction task and the secondary page interaction task. The primary page interaction task can refer to the user's interaction behavior on the main page (such as the website homepage), while the secondary page interaction task refers to the user's interaction behavior on the secondary page (such as the product details page, shopping cart page, etc.).

[0123] In the primary page interaction task, the system mines the key operation preference dense vector through the decision tree network. This vector reflects the user's main operation preference and behavior pattern on the primary page. At the same time, in the secondary page interaction task, the system also uses the decision tree network to mine the secondary key operation preference dense vector, which reflects the user's operation preference on the secondary page.

[0124] Finally, the system performs weighted integration of the key operation preference dense vector and the sub-key operation preference dense vector to obtain the final operation preference dense vector. This vector integrates the user's operation preference information under different page interaction tasks and can more comprehensively reflect the user's overall operation habits and preferences.

[0125] Through this series of operations, the user big data mining system successfully mined the dense vector of user operation preferences. This not only helps companies to understand users' operation habits and preferences more deeply, but also provides strong data support for optimizing page design and improving user experience.

[0126] In this way, by comprehensively considering the user's operation behavior under different page interaction tasks, the user's operation preferences and needs can be captured more accurately. At the same time, the use of decision tree networks for mining also improves the efficiency and accuracy of data processing. Ultimately, these mining results will provide companies with more accurate user behavior insights, helping to improve user satisfaction and business conversion rates.

[0127] In some optional technical solutions, the operation preference dense vectors under different page interaction tasks are weighted integrated to obtain the x-th order supply chain finance business session vector, including: fusing the operation preference dense vector with the page interaction task vector to obtain the interaction task directional preference fusion viewpoint; weighted integration of each of the interaction task directional preference fusion viewpoints to obtain the x-th order supply chain finance business session vector.

[0128] Based on this technical solution, the user big data mining system further performs weighted integration processing on the dense vectors of operation preferences under different page interaction tasks to obtain the x-th order supply chain finance business conversation vector.

[0129] First, the system fuses the dense operation preference vector with the page interaction task vector. The page interaction task vector here is a data representation used to describe the user's behavioral characteristics and preferences under a specific page interaction task. The fusion process can be understood as combining the dense operation preference vector with the page interaction task vector in some form, such as through weighted averaging, splicing or other feature fusion techniques, to form a new data representation, namely, the interaction task directional preference fusion view. This fusion view contains both the user's operation preference information and the specific page interaction task characteristics, so it can more comprehensively reflect the user's behavior pattern under a specific task.

[0130] Next, the system performs weighted integration of the directional preference fusion opinions of each interactive task. This means that the system assigns corresponding weights to each fusion opinion according to its importance and information content, and sums them up to obtain a comprehensive and global representation, namely the x-order supply chain finance business session vector. This session vector integrates the user's behavioral preference information under multiple different page interaction tasks, so it can more accurately characterize the user's overall behavioral characteristics and needs in the supply chain finance business scenario.

[0131] In this way, the user big data mining system successfully uses the dense vector of operation preferences and the vector of page interaction tasks to build a comprehensive and detailed user behavior model. This model can not only help enterprises to have a deeper understanding of the behavior patterns and preferences of users in the supply chain finance business, but also provide strong data support for personalized services and precision marketing. In other words, by integrating the user operation preference information under different page interaction tasks, a more global and comprehensive user behavior session vector is generated. This vector can more accurately reflect the overall needs and preferences of users in the supply chain finance business, thereby providing enterprises with more accurate and personalized service strategies. This not only helps to improve user satisfaction and loyalty, but also effectively improves the business efficiency and market competitiveness of enterprises.

[0132] In some alternative embodiments, performing user preference dense vector mining on each of the user behavior session vectors to obtain the operation preference dense vectors under different page interaction tasks includes: respectively utilizing the session dense vector mining models corresponding to the same model variables to perform user preference dense vector mining on the user behavior session vectors under each page interaction task to obtain the operation preference dense vectors under different page interaction tasks.

[0133] Based on this embodiment, the user big data mining system adopts another method to mine user preference dense vectors for user behavior session vectors.

[0134] The system first uses the session dense vector mining model corresponding to the same model variable to mine user preference dense vectors for each page interaction task. The "model variable" here refers to the features or parameters used in the mining process, and the "session dense vector mining model" is an algorithm model specifically used to extract user preferences from user behavior session vectors.

[0135] Specifically, for each page interaction task, the system will apply a session dense vector mining model corresponding to the task. Although these models use the same model variables, they will be fine-tuned according to the characteristics and data distribution of different page interaction tasks to ensure the accuracy and effectiveness of the mining results.

[0136] During the mining process, the system inputs the user behavior session vector into the corresponding mining model, and extracts the user's operation preference information under the page interaction task through model calculation and analysis. This information will be encoded into an operation preference dense vector for subsequent analysis and application.

[0137] Since user behavior patterns and preferences under different page interaction tasks may vary greatly, the system can more accurately capture the user's real preferences and needs under each task by adopting targeted mining models.

[0138] Finally, the system obtains a series of dense vectors of operation preferences under different page interaction tasks. These vectors not only reflect the user's operating habits and preferences on each page, but also provide valuable user behavior insights for enterprises, which helps to optimize product design, improve user experience and promote business innovation. In this way, the user big data mining system can more effectively mine and utilize valuable information in user behavior data, and provide enterprises with more accurate and personalized service strategies. This not only helps to improve user satisfaction and loyalty, but also improves the market competitiveness of enterprises, and achieves business growth and profitability.

[0139] In some other alternative embodiments, the conversation dense vector mining process includes a vector extraction process and a vector optimization process, the vector extraction process includes a feature compression process for the supply chain finance business conversation vector, and the vector optimization process includes a feature derivation process for the supply chain finance business conversation vector; the user push demand mining is performed on the yth-order supply chain finance business conversation vector obtained by mining the yth-order conversation dense vector to obtain the user push demand mining result of the supply chain finance business log, including: on the basis of the feature derivation reaching the initial dimension, the extracted supply chain finance business conversation vector is determined as the yth-order supply chain finance business conversation vector obtained by mining the yth-order conversation dense vector; user push demand mining is performed based on the yth-order supply chain finance business conversation vector to obtain the user push demand mining result.

[0140] Based on this embodiment, the session dense vector mining process of the user big data mining system is further refined into a vector refinement process and a vector optimization process. These two processes play a key role in improving data quality and mining accuracy.

[0141] First, the vector extraction process mainly involves feature compression of supply chain finance business conversation vectors. Feature compression refers to the use of certain algorithms and technical means to merge, simplify or remove redundant or highly correlated features in conversation vectors, thereby retaining the most representative and distinguishing feature information. This process can reduce the dimension and complexity of the data while retaining the main features and structure of the data, providing a more refined and efficient data set for subsequent data analysis and mining. In the user big data mining system, feature compression can more effectively extract key information of supply chain finance business conversations, laying a solid foundation for subsequent user preference mining and precise push.

[0142] Secondly, the vector optimization process is to derive features from the supply chain finance business session vector. Feature derivation refers to the generation of new, more expressive and discriminative features based on the original features to enrich the expressiveness of the data and improve the predictive performance of the model. In the user big data mining system, feature derivation may involve operations such as combining, transforming or introducing external information on the original features to generate new features that are more targeted and discriminative. This process helps capture deeper patterns and rules of user behavior, providing strong support for accurate user portraits and personalized services.

[0143] When the initial feature derivation reaches a certain dimension, the system determines the supply chain finance business conversation vector extracted at this time as the y-order supply chain finance business conversation vector obtained by mining the y-order conversation dense vector. This step is a key link to ensure data quality and mining accuracy, which ensures that the conversation vector used for subsequent user push demand mining is high-quality data that has been optimized and refined.

[0144] Finally, based on the y-th order supply chain finance business conversation vector obtained above, the user big data mining system conducts user push demand mining. This step aims to deeply analyze the user's behavior patterns and preferences to accurately predict and meet the user's personalized needs. By mining the user push demand results, the company can more accurately push content and services that meet the user's interests and needs to improve the user experience and satisfaction.

[0145] In this way, this refined conversation dense vector mining process not only improves the refinement and expressiveness of the data, but also provides more accurate and comprehensive data support for subsequent user push demand mining. Through this technical solution, the user big data mining system can have a deeper understanding of user needs and behavior patterns, provide enterprises with more accurate and personalized service strategies, and thus enhance market competitiveness and achieve business growth.

[0146] In some examples, the supply chain finance business log is an e-commerce business log, a blockchain transaction business log, or an Internet credit business log.

[0147] Based on this example, the supply chain finance business logs processed by the user big data mining system can be various types of business logs, such as e-commerce business logs, blockchain transaction business logs, or Internet credit business logs. These different types of logs contain rich user behavior data and business information, which are of great analytical value to the user big data mining system.

[0148] First, when processing e-commerce business logs, the system mainly focuses on users' shopping behaviors, browsing records, search histories, and payment information on e-commerce platforms. By analyzing these data, the system can gain an in-depth understanding of users' consumption habits, purchasing preferences, and price sensitivity, thereby providing personalized product recommendations, promotion strategies, and customer service for e-commerce platforms.

[0149] Secondly, for blockchain transaction business logs, the user big data mining system focuses on analyzing transaction data, smart contract execution, and user interaction in the blockchain network. Such logs are highly transparent and traceable, which helps the system discover potential risk points, abnormal transaction behaviors, and market trends, and provide risk warnings and decision-making support for financial institutions.

[0150] Finally, in terms of Internet credit business logs, the system mainly focuses on users' borrowing and lending behaviors, repayment records, credit scores, and overdue situations. These data are of great significance for assessing users' credit risks, formulating reasonable credit policies, and optimizing risk control models. By analyzing these logs, the system can help financial institutions more accurately identify high-quality customers and potential risk customers, thereby improving the efficiency and security of credit business.

[0151] Therefore, the user big data mining system can provide enterprises with all-round and multi-dimensional user behavior analysis and business insights by processing different types of supply chain finance business logs. This not only helps enterprises to have a deeper understanding of user needs and market dynamics, but also provides strong data support for their strategic decision-making and business innovation. By mining and utilizing the valuable information in these log data, enterprises can optimize product design, improve service quality, and reduce operating costs, thus standing out in the fierce market competition.

[0152] Please refer to Figure 2 The embodiment of the present application also provides a user big data mining system 100, including a processor 111, and a memory 112 and a bus 113 connected to the processor 111. The processor 111 and the memory 112 communicate with each other through the bus 113. The processor 111 is used to call the program instructions in the memory 112 to execute the above-mentioned user big data mining method based on supply chain finance.

[0153] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity or system including the element.

[0154] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0155] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A user big data mining method based on supply chain finance, characterized in that: Applied to a user big data mining system, the method comprises: In the process of mining the x-th order conversation dense vector, the r-th order supply chain finance business conversation vector is feature updated to obtain a user behavior conversation vector, wherein the r-th order supply chain finance business conversation vector is the result of mining the r-th order conversation dense vector on the supply chain finance business log, and different user behavior conversation vectors are conversation vectors of the r-th order supply chain finance business conversation vector under different page interaction tasks; Performing user preference dense vector mining on each of the user behavior session vectors to obtain operation preference dense vectors under different page interaction tasks; Performing weighted integration on the operation preference dense vectors under different page interaction tasks to obtain an x-th order supply chain finance business session vector; Perform user push demand mining based on the y-th order supply chain finance business session vector obtained by mining the y-th order session dense vector, and obtain the user push demand mining result of the supply chain finance business log, where y is not less than x, and r=x-1; The performing user preference dense vector mining on each of the user behavior session vectors to obtain the operation preference dense vectors under different page interaction tasks includes: respectively using the session dense vector mining models corresponding to the same model variables to perform user preference dense vector mining on the user behavior session vectors under each page interaction task to obtain the operation preference dense vectors under different page interaction tasks; Among them, the conversation dense vector mining process includes a vector extraction process and a vector optimization process, the vector extraction process includes a feature compression process for the supply chain finance business conversation vector, and the vector optimization process includes a feature derivation process for the supply chain finance business conversation vector; the user push demand mining is performed on the y-th order supply chain finance business conversation vector obtained by mining the y-th order conversation dense vector to obtain the user push demand mining result of the supply chain finance business log, including: on the basis of the feature derivation reaching the initial dimension, the extracted supply chain finance business conversation vector is determined as the y-th order supply chain finance business conversation vector obtained by mining the y-th order conversation dense vector; user push demand mining is performed based on the y-th order supply chain finance business conversation vector to obtain the user push demand mining result; Among them, the supply chain finance business log is an e-commerce business log, a blockchain transaction business log or an Internet credit business log.

2. The method according to claim 1, characterized in that The user preference dense vector mining is performed on each of the user behavior session vectors to obtain the operation preference dense vectors under different page interaction tasks, including: Performing time series session dense vector mining on the user behavior session vector to obtain a time series session dense vector; The user preference dense vector is mined on the time series session dense vector based on the primary page interaction task and the secondary page interaction task to obtain the operation preference dense vector. The primary page interaction task is the page interaction task corresponding to the user behavior session vector, and the secondary page interaction task is a page interaction task in the dynamic page interaction task that is different from the primary page interaction task.

3. The method according to claim 2, characterized in that The performing time series session dense vector mining on the user behavior session vector to obtain the time series session dense vector includes: Decomposing the user behavior conversation vector based on feature cores to obtain user behavior conversation sub-vectors corresponding to P feature cores, wherein the P feature cores do not overlap; Performing session dense vector mining on the P user behavior session sub-vectors to obtain a local time series session dense vector; The P feature cores are updated based on the feature cores, and session dense vector mining is performed on the local time series session dense vectors corresponding to the P feature cores after the feature core update to obtain the time series session dense vectors. The feature core-based update is used to adjust the distribution labels of the P feature cores.

4. The method according to claim 3, characterized in that The performing session dense vector mining on the P user behavior session sub-vectors to obtain a local time series session dense vector includes: Performing a feature focusing operation on the P user behavior conversation sub-vectors to obtain behavior conversation focus vectors of the P user behavior conversation sub-vectors; Performing weighted integration on the behavior conversation attention vectors of the P user behavior conversation sub-vectors to obtain a first behavior conversation weighted vector; Performing residual processing on the first behavior session weighted vector to obtain the first behavior session residual connection vector; A decision tree network is used to perform session dense vector mining on the first behavior as the session residual connection vector to obtain the local time series session dense vector.

5. The method according to claim 3, characterized in that: The performing session dense vector mining on the local time series session dense vectors corresponding to the P feature cores updated based on the feature core to obtain the time series session dense vectors includes: Performing a feature focusing operation on the local temporal conversation dense vectors corresponding to the P feature cores updated based on the feature core to obtain the behavior conversation attention vectors corresponding to the P feature cores; Performing weighted integration on the P behavior session attention vectors to obtain a second behavior session weighted vector; Performing feature mapping on the second behavior session weighted vector, and performing residual processing on the feature-mapped second behavior session weighted vector to obtain a second behavior session residual connection vector; A decision tree network is used to perform session dense vector mining on the second behavior as the session residual connection vector to obtain the time series session dense vector.

6. The method according to claim 4, characterized in that The performing a feature focusing operation on the P user behavior conversation sub-vectors to obtain the behavior conversation focus vectors of the P user behavior conversation sub-vectors includes: Performing a feature focusing operation based on the index feature, identification feature, and attribute feature corresponding to the user behavior conversation vector to obtain P behavior conversation attention vectors of the local time series conversation dense vector; Among them, the conversation dense vector mining process includes a vector extraction process and a vector optimization process. The identification eigenvalue and attribute eigenvalue in the z-th order vector optimization process are obtained by fusing the identification eigenvalue and attribute eigenvalue in the z-1-th order vector optimization with the identification eigenvalue and attribute eigenvalue in the corresponding vector extraction process. The index eigenvalue in the y-th order vector optimization process is the index eigenvalue in the z-1-th order vector optimization.

7. The method according to claim 2, characterized in that The performing user preference dense vector mining on the time series session dense vector based on the primary page interaction task and the secondary page interaction task to obtain the operation preference dense vector includes: Performing weighted integration on the time series session dense vector and the distribution representation vector to obtain a first operation preference dense vector, wherein the distribution representation vector is used to indicate a location label corresponding to the user behavior session vector; Under the primary page interaction task, using a decision tree network to perform user preference dense vector mining on the first operation preference dense vector to obtain a key operation preference dense vector; Under the secondary page interaction task, using the decision tree network to perform user preference dense vector mining on the first operation preference dense vector to obtain a secondary key operation preference dense vector; The key operation preference dense vector and the sub-key operation preference dense vector are weightedly integrated to obtain the operation preference dense vector.

8. The method according to any one of claims 1 to 7, characterized in that: The weighted integration of the operation preference dense vectors under different page interaction tasks to obtain the x-th order supply chain financial business session vector includes: The operation preference dense vector is fused with the page interaction task vector to obtain an interaction task directional preference fusion viewpoint; The directional preference fusion viewpoints of each interactive task are weighted and integrated to obtain the x-th order supply chain finance business conversation vector.

9. A user big data mining system, characterized in that: It comprises a processor, a memory and a bus connected to the processor; the processor and the memory communicate with each other via the bus; The processor is used to call the computer program in the memory to execute the user big data mining method based on supply chain finance as described in any one of claims 1-8.

Images