Hierarchical information deep mining and matching method applied to procurement service system

Abstract

The present application provides a hierarchical information deep mining and matching method applied to a procurement service system, which relates to the field of procurement management technology. The method includes: executing data interaction with the procurement service system to establish a multi-source data set; calling a surface information recognition layer to perform data preprocessing and recognition to establish an initial matching candidate library; inputting the preprocessed multi-source data set and the initial matching candidate library into a multi-layer matching channel, configuring the multi-layer matching channel with historical data, performing procurement task matching reconstruction, and generating matching reconstruction results; and performing procurement management based on the matching reconstruction results. This application solves the technical problem in the prior art that it is difficult to accurately identify the best match between procurement tasks and procurement products, resulting in inaccurate procurement decisions, thereby affecting procurement efficiency. Through hierarchical information deep mining and matching, accurate matching of procurement services is performed, thereby improving procurement efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of procurement management, and in particular to a hierarchical information deep mining and matching method applied to a procurement service system. Background Art

[0002] Procurement service systems are tools that manage a company's procurement processes and activities through information technology. Currently, many procurement service systems rely on traditional rule-based matching or simple filtering algorithms, failing to deeply analyze the complex relationships between various variables (such as market demand, seasonality, and historical purchasing records). This results in low matching accuracy. They perform poorly in handling diverse and complex procurement tasks and are unable to dynamically adjust procurement strategies to meet evolving needs and market conditions. This leads to inaccurate matching results and significantly reduces procurement efficiency.

[0003] In summary, the existing technology has a technical problem in that it is difficult to accurately identify the best match between procurement tasks and procurement products, resulting in inaccurate procurement decisions, thereby affecting procurement efficiency. Summary of the Invention

[0004] The purpose of this application is to provide a hierarchical information deep mining and matching method applied to the procurement service system, so as to solve the technical problem in the existing technology that it is difficult to accurately identify the best match between procurement tasks and procurement products, resulting in inaccurate procurement decisions and thus affecting procurement efficiency.

[0005] In view of the above problems, the present application provides a hierarchical information deep mining and matching method applied to a procurement service system, wherein the hierarchical information deep mining and matching method applied to a procurement service system includes: executing data interaction with the procurement service system to establish a multi-source data set; calling the surface information recognition layer to perform data preprocessing and identification of the multi-source data set to establish an initial matching candidate library, wherein the initial matching candidate library is a mapping matching database between procurement tasks and procurement products, and the initial matching candidate library includes a matching library, a candidate library and a taboo library; inputting the preprocessed multi-source data set and the initial matching candidate library into a multi-layer matching channel, configuring the multi-layer matching channel using historical data in the preprocessed multi-source data set, performing procurement task matching reconstruction based on the initial matching candidate library, and generating a matching reconstruction result, wherein the multi-layer matching channel includes a semantic matching layer, a behavioral analysis layer, and a transition decision analysis layer; and performing procurement management according to the matching reconstruction result.

[0006] Optionally, obtain historical data from the preprocessed multi-source data set; send the historical data to the semantic matching layer, perform synonymous semantic conversion using the semantic matching layer, and then perform procurement description reconstruction; use the semantic embedding comparison network, structural label path comparison network, and knowledge graph logic comparison network in the semantic matching layer to perform semantic mapping alignment of the procurement description reconstruction results; adapt and reconstruct the procurement tasks in the initial matching candidate library according to the semantic mapping alignment results, establish a semantic adaptation reconstruction result, and send the semantic adaptation reconstruction result to the transition decision analysis layer.

[0007] Optionally, the historical data is sent to the behavior analysis layer, and the behavior analysis layer is used to reconstruct the procurement chain path diagram; the behavioral preferences and behavioral patterns of each procurement task are extracted based on the procurement chain path diagram; behavioral analysis clustering is performed according to the procurement tasks, behavioral preferences, and behavioral patterns, and a behavior analysis clustering result is established; the procurement tasks in the initial matching candidate library are adapted and reconstructed using the behavior analysis clustering result, and a behavior adaptation reconstruction result is established, and the behavior adaptation reconstruction result is sent to the transition decision analysis layer.

[0008] Optionally, a transition judgment threshold of the transition decision analysis layer is configured based on the initial matching candidate library, and the transition judgment threshold is the transition threshold from the candidate library and the taboo library to the matching library in the mapping matching process; after receiving the semantic adaptation reconstruction result and the behavioral adaptation reconstruction result, the matching value reconstruction of the initial matching candidate library is performed, and the transition judgment threshold is used to perform the reconstructed matching value transition judgment; and the procurement task matching reconstruction is completed according to the transition judgment result.

[0009] Optionally, the standard deviation of the matching value of the initial matching candidate library is calculated to generate a calculation result; a first dynamic adjustment factor and a second dynamic adjustment factor are configured, the first dynamic adjustment factor is the transition adjustment factor from the candidate library to the matching library, and the second dynamic adjustment factor is the transition adjustment factor from the taboo library to the matching library; a matching sliding window is established, and the matching sliding window is used to perform matching attention extraction on the initial matching candidate library to establish a matching attention extraction result; the transition judgment threshold of the transition decision analysis layer is configured with the matching attention extraction result, the first dynamic adjustment factor, the second dynamic adjustment factor, and the calculation result.

[0010] Optionally, graph feature extraction is performed on the procurement chain path diagram to establish a graph feature extraction result; feature criticality is identified using the graph feature extraction result to establish key decision features; and after associating the key decision features with the tasks in the procurement chain path diagram, behavioral preferences and behavioral patterns are generated.

[0011] Optionally, the procurement tasks are time-sorted to establish time-sorting results, and time series associations are established based on the time-sorting results; the procurement tasks are clustered based on the time series associations, and behavior analysis clustering based on the procurement task clustering results is performed based on behavior preferences and behavior patterns to establish behavior analysis clustering results.

[0012] Optionally, a collection evaluation mapped to the matching and reconstruction result is established, and a feedback database is generated using the collection evaluation; and channel optimization management of the multi-layer matching channel is performed using the feedback database.

[0013] Optionally, an identification rule for abnormal tasks is established, the historical data is screened based on the identification rule, and an abnormal task identifier is established; data diversion of the historical data is performed according to the abnormal task identifier, and multi-layer matching channel configuration is completed according to the data diversion result.

[0014] Optionally, a matching warning identification is performed on the matching reconstruction result to establish an abnormality early warning output; a new extended matching suggestion is generated based on the abnormality early warning output, and the new extended matching suggestion is fed back to the management user.

[0015] The technical solution provided in this application has at least the following beneficial effects:

[0016] By performing data interaction with the procurement service system, a multi-source dataset is established; the surface information recognition layer is called to perform data preprocessing and recognition of the multi-source dataset, and an initial matching candidate library is established. The initial matching candidate library is a mapping and matching database between procurement tasks and procurement products, and the initial matching candidate library includes a matching library, a candidate library, and a taboo library; the preprocessed multi-source dataset and the initial matching candidate library are input into a multi-layer matching channel, and after configuring the multi-layer matching channel using historical data in the preprocessed multi-source dataset, procurement task matching reconstruction based on the initial matching candidate library is performed to generate a matching reconstruction result. The multi-layer matching channel includes a semantic matching layer, a behavioral analysis layer, and a transition decision analysis layer; procurement management is performed based on the matching reconstruction result. In other words, by collecting multi-source datasets, establishing an initial matching candidate library, and configuring a multi-layer matching channel, procurement tasks and products are accurately matched from multiple dimensions, the matching results are optimized, and the final procurement decision is generated, thereby improving procurement efficiency.

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, which can be implemented in accordance with the contents of the description, and to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are specifically listed below. It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easy to understand through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in this application or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely exemplary, and a person of ordinary skill in the art can obtain other drawings based on the provided drawings without creative work.

[0019] Figure 1 This is a flowchart of the hierarchical information deep mining and matching method applied to the procurement service system in this application.

[0020] Figure 2 A flowchart illustrating the process of establishing behavioral adaptation and reconstruction results in the hierarchical information deep mining and matching method applied to the procurement service system in this application. DETAILED DESCRIPTION

[0021] This application addresses the existing technical problem of inaccurate procurement decisions, which hinders procurement efficiency, by providing a hierarchical information deep mining and matching method for procurement service systems. This method collects multi-source datasets, establishes an initial matching candidate library, and configures multi-layer matching channels. This method accurately matches procurement tasks and products from multiple dimensions, optimizes the matching results, and generates the final procurement decision, thereby improving procurement efficiency.

[0022] Below, the technical solutions in this application will be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this application, rather than all the embodiments of this application. It should be understood that this application is not limited to the example embodiments described herein. Based on the embodiments of this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application. It should also be noted that, for the convenience of description, only the parts related to this application, rather than all of them, are shown in the accompanying drawings.

[0023] For examples, please see the attached Figure 1The present application provides a hierarchical information deep mining and matching method applied to a procurement service system, wherein the hierarchical information deep mining and matching method applied to a procurement service system specifically includes the following steps:

[0024] S100: Execute data interaction with the procurement service system and establish a multi-source data set.

[0025] Specifically, the procurement service system is an enterprise-level information system designed to support digital management of the entire process, from procurement requirements to product warehousing, sales, transfers, inventory management, and financial settlement. It includes multiple subsystems, such as the decision analysis subsystem, quotation management subsystem, warehouse management subsystem, basic archive subsystem, procurement management subsystem, incoming goods management subsystem, sales management subsystem, transfer management subsystem, inventory management subsystem, accounts receivable management subsystem, and a cash register backend. Data is exchanged between the various subsystems of the procurement service system to establish multi-source datasets, including procurement data, supplier data, market data, and logistics data. Within the procurement service system's architecture, various modules (such as procurement management, inventory management, and quotation management) interact with each other through APIs, message queues, or direct database connections. For example, the procurement management module and the supplier management module share real-time data such as supplier inventory status and product prices through data interfaces.

[0026] When constructing a multi-source dataset, data consistency and integrity must be ensured. This is accomplished through steps such as data synchronization, data cleansing, and data standardization, ultimately integrating the data into a unified multi-source dataset. For example, suppose Company A is engaged in procurement. A multi-source dataset is constructed based on the following data sources: Procurement data: Company A's purchase order data from the past year, including product name, quantity, purchase date, and unit price (e.g., Order 1: Product A, quantity 1,000, unit price 5 yuan); Supplier data: Supplier B's quote and delivery capabilities; e.g., Supplier B's quote for Product A is 4.8 yuan, with a lead time of 7 days; Market data: Product A's market price trends, with price fluctuations ranging from 4.5 yuan to 5.5 yuan over the past six months; External API data: Supplier B's logistics company's delivery time, averaging 5 days. Data from these diverse sources is interacted, pre-processed, and integrated through APIs and database interfaces to create a multi-source dataset. Multi-source datasets provide rich data, integrating multiple perspectives, from procurement records to supplier performance and market trends. This allows the procurement service system to comprehensively consider various factors and achieve more accurate matching.

[0027] S200: Calling the surface information recognition layer to perform data preprocessing and recognition of the multi-source data set, and establishing an initial matching candidate library. The initial matching candidate library is a mapping matching database between procurement tasks and procurement products, and the initial matching candidate library includes a matching library, a candidate library, and a taboo library.

[0028] Specifically, the surface information recognition layer is a module or stage in the data processing process that is responsible for extracting preliminary information from multi-source data sets. It mainly focuses on the preliminary filtering, classification, and organization of data, and usually includes operations such as data cleaning and feature extraction. The surface information recognition layer cleans the raw data obtained from multi-source data sets, including removing redundant data, correcting missing values, and removing invalid or erroneous data. Data from different sources are converted into a unified format for subsequent operations. For example, product prices provided by different suppliers are converted to the same unit of measurement (such as yuan / unit), or product names are standardized. Features related to matching the procurement task are extracted from the data, such as product price, delivery time, quality rating, and other information.

[0029] After data preprocessing, the surface information recognition layer uses a series of recognition rules or algorithms to classify the processed data and identify matching candidates. For example, products are initially screened based on their characteristics (such as specifications, functions, and price) to determine which products are likely to be suitable for the current procurement task. This series of recognition rules or algorithms includes product specification matching rules, price matching rules, supplier evaluation rules, market trend rules, risk management rules, quality control rules, and compliance rules. These predefined rules are typically encoded into the procurement service system's algorithms, allowing the system to automatically perform data matching and analysis.

[0030] Leveraging historical data, the system identifies products that are highly compatible with the procurement task, potential candidates, and completely incompatible with the requirements, and labels them accordingly. A preliminary matching analysis is performed on the preprocessed data to establish an initial matching candidate library, consisting of a matching library, a candidate library, and a taboo library, storing different types of matching relationships. Products that are highly compatible with the procurement task are screened to form the matching library. Products in the matching library closely align with the procurement task requirements in terms of price, delivery time, quality, and other aspects, meeting the optimal selection criteria. Products that have a certain degree of compatibility but may not fully meet all criteria are assigned to the candidate library. Products in the candidate library may not be included in the matching library due to certain factors (such as slightly higher price or longer delivery time), but they remain viable candidates for further evaluation. Products that completely mismatch the procurement task are automatically excluded and assigned to the taboo library. For example, products whose specifications differ significantly from the procurement task requirements or whose supplier's credit score does not meet the requirements are removed from the candidate library.

[0031] After the initial matching candidate library is established, it is continuously updated. That is, during the actual procurement process, the initial matching candidate library is regularly optimized and adjusted based on new procurement task data, supplier updated data, and market dynamics. For example, if a product exhibits poor delivery timeliness during the actual procurement process, it will be removed from the matching library or reclassified into a candidate library or taboo library. The initial matching candidate library refers to a preliminary candidate library generated based on multi-source data sets through certain preprocessing and recognition rules. It contains preliminary mapping relationships between procurement tasks and procurement products. However, these mapping relationships are not the final matching results, but rather a candidate set that has undergone certain screening, classification, and induction.

[0032] For example, suppose Company A needs to purchase 1,000 smartphones. The procurement task includes the following requirements: price: no more than 3,000 yuan per unit, delivery time: no more than 10 days, and quality rating: 4 stars or higher. Through preprocessing and recognition at the surface information recognition layer, data processing is performed on the multi-source dataset, extracting the following information: Supplier 1: Product A (quote price 2,800 yuan, delivery time 10 days, quality rating 4.5 stars); Supplier 2: Product B (quote price 3,200 yuan, delivery time 12 days, quality rating 4.2 stars); Supplier 3: Product C (quote price 2,900 yuan, delivery time 10 days, quality rating 3.8 stars); Supplier 4: Product D (quote price 2,500 yuan, delivery time 7 days, quality rating 4.9 stars). The products are then categorized according to the procurement task requirements: a matching pool: Product D from Supplier 4; a candidate pool: Product A from Supplier 1 (both price and quality rating meet requirements, but the delivery time is slightly longer and requires further evaluation); and a taboo pool: Product B from Supplier 2 (the quote price and delivery time do not meet the requirements) and Product C from Supplier 3 (the quality rating does not meet the requirements).

[0033] By establishing an initial matching candidate library, we can accurately identify the best match between procurement tasks and products, avoid subjective errors in manual screening, and quickly provide the best matching products and alternative products, reducing manual intervention and decision-making time. By eliminating the taboo library, we can effectively eliminate products that do not meet the conditions, thereby reducing invalid selections and focusing on making reasonable procurement decisions.

[0034] S300: Input the preprocessed multi-source data set and the initial matching candidate library into a multi-layer matching channel. After configuring the multi-layer matching channel using historical data in the preprocessed multi-source data set, perform procurement task matching reconstruction based on the initial matching candidate library to generate a matching reconstruction result. The multi-layer matching channel includes a semantic matching layer, a behavioral analysis layer, and a transition decision analysis layer.

[0035] Furthermore, the present application S300 includes:

[0036] S310: Establishing identification rules for abnormal tasks, screening the historical data based on the identification rules, and establishing abnormal task identification; S320: Executing data diversion of historical data according to the abnormal task identification, and completing multi-layer matching channel configuration according to the data diversion result.

[0037] Specifically, historical data is extracted from the preprocessed multi-source data set, including past procurement task records, supplier information, market prices, etc. Abnormal task identification rules are a set of logic or algorithms used to identify tasks or behaviors that are different from normal patterns from historical data. Abnormal tasks usually refer to tasks that show significant deviations in certain key characteristics, such as abnormal procurement volume, price deviation, unreasonable delivery time, etc., which may be caused by system errors, data input problems, or changes in the market environment. For example, if the price of a procurement task is significantly different from the historical average price, it can be marked as a price abnormality task; if the supplier's delivery time exceeds the expected normal range, it can be marked as a delivery time abnormality task; if the procurement volume suddenly increases or decreases, exceeding the normal procurement fluctuation range, it can be marked as a procurement volume abnormality task.

[0038] According to the preset abnormal task identification rules, historical data is screened to identify which tasks meet the abnormal conditions. When an abnormal task is identified, it will be marked with an abnormal flag to ensure that it receives special treatment in the subsequent process. For example, suppose the price of a historical procurement task is 10,000 yuan, and the price fluctuation range of the same product in history is usually between 3,000 and 5,000 yuan. This task is identified as an abnormal task and marked as a price abnormality. A unique abnormal task identifier is assigned to each identified abnormal task to quickly identify the corresponding task. The abnormal task identifier is a label or mark assigned to it by the procurement service system after it identifies an abnormal task. Based on the abnormal task identifier, it is possible to identify which tasks are normal and which tasks are abnormal.

[0039] Based on abnormal task identification, historical data is triaged. Abnormal tasks are isolated and processed separately, while normal tasks continue to enter the matching pipeline according to the normal process. The goal of data triage is to ensure that abnormal tasks do not interfere with the matching and decision-making process for normal tasks. For example, abnormal tasks may require additional review and optimization steps to ensure that they do not affect the accuracy of the overall purchasing decision. Different matching pipelines are configured for normal and abnormal tasks based on the data characteristics. The multi-layer matching pipeline includes a semantic matching layer, a behavioral analysis layer, and a transition decision analysis layer. The semantic matching layer is responsible for semantic matching and conversion of text, descriptions, or tags in the data. By mapping the descriptions in the raw data into a standardized semantic space, different representations can be uniformly understood. The behavioral analysis layer analyzes user purchasing preferences and behavioral patterns based on historical purchasing behavior, identifies similar historical purchasing tasks, and optimizes the selection of candidate products based on these behavioral patterns. The transition decision analysis layer makes the final matching decision based on the analysis results of the first two layers, determining which candidate products should be transitioned to the final matching pool and generating the final matching results.

[0040] After the multi-layer matching channel is configured, procurement task matching reconstruction is performed based on the initial matching candidate library and the preprocessed multi-source dataset. The matching degree between each procurement task and multiple candidate products is re-evaluated, and the matching score is adjusted based on the results of semantic matching and behavioral analysis. Based on the procurement task matching reconstruction, a matching reconstruction result is generated, including the final matching score between each procurement task and the corresponding product. These scores reflect the best matching product for each procurement task. The matching reconstruction result is the output result after executing the procurement task matching reconstruction. It represents the relationship between each procurement task and the most matching procurement product after multi-level analysis. It is the final output after calculation of the multi-layer matching channel and is used to guide subsequent procurement decisions. Through the multi-layer matching channel, the matching results are optimized from the two aspects of semantic understanding and behavioral analysis, reducing matching errors caused by inconsistent descriptions or different behavioral preferences.

[0041] Furthermore, the present application further comprises the following steps:

[0042] S330: Acquire the historical data in the pre-processed multi-source data set; S340: Send the historical data to the semantic matching layer, use the semantic matching layer to perform synonymous semantic conversion, and then perform procurement description reconstruction; S350: Use the semantic embedding comparison network, structural label path comparison network, and knowledge graph logic comparison network in the semantic matching layer to perform semantic mapping alignment of the procurement description reconstruction results; S360: Adapt and reconstruct the procurement tasks in the initial matching candidate library according to the semantic mapping alignment results, establish a semantic adaptation reconstruction result, and send the semantic adaptation reconstruction result to the transition decision analysis layer.

[0043] Specifically, relevant historical data is extracted from pre-processed multi-source data sets, including historical procurement task descriptions, product information, supplier information, etc. The historical data is sent to the semantic matching layer, which first performs synonymous semantic conversion to ensure that the different words and expressions in the procurement task description can be unified. For example, "procurement", "purchase", "buy", "purchase", "purchase", "buy" and "purchase" are converted into the same standardized description "procurement", and "sell", "sell", "sell" and "sell on sale" are converted into the same standardized description "sell", thereby eliminating matching deviations caused by language differences. Synonymous semantic conversion is the conversion of different but similar words or phrases into standardized semantic expressions, ensuring that the meaning expressed is the same in different contexts or expressions.

[0044] Synonymous semantic conversion is a crucial step in the semantic matching layer of procurement service systems. Converting synonyms or near-synonyms in procurement descriptions into standardized terms helps eliminate matching errors caused by lexical discrepancies and improves the accuracy of matching between procurement tasks and purchased products. A dictionary containing mappings between synonyms and standardized terms is typically automatically generated from an existing knowledge base. Before performing synonymous semantic conversion, the procurement descriptions must be preprocessed, including removing extraneous spaces and punctuation, converting the text to lowercase, and performing word segmentation. For each term in the procurement description, a synonym dictionary is searched for a corresponding synonym. If a synonym is found, the term is replaced with a standardized term from the dictionary. In some cases, the meaning of a term may depend on its context. Therefore, the context of the term is analyzed to ensure the accuracy of synonymous semantic conversion. In certain procurement domains, unique terminology and expressions may exist. Integrating this domain-specific knowledge ensures the accuracy of synonymous semantic conversion. The synonym dictionary and conversion rules are continuously optimized based on user feedback.

[0045] After synonymous semantic conversion, procurement description reconstruction is performed. This is a structured process designed to extract standardized information from inconsistent procurement task descriptions. For example, "Purchase 10,000 laptops, model X, delivery next month" is converted into the standard format: "Product: Laptops; Quantity: 10,000; Model: X; Delivery: Next Month." Procurement description reconstruction not only standardizes the description format but also ensures the clear presentation of key elements such as product name, quantity, and delivery time. The result of procurement description reconstruction is a structured, standardized information format that enables procurement tasks from different sources and representations to be processed consistently on the same platform.

[0046] The semantic matching layer includes a semantic embedding comparison network, a structural label path comparison network, and a knowledge graph logical comparison network. The semantic embedding comparison network is a deep learning model used to compare different semantic expressions. It converts different texts (such as purchase descriptions) into vector representations (semantic embeddings) and measures the similarity between different texts by comparing these embedding vectors. The structural label path comparison network is a comparison model designed for processing structured data (such as graphs or label paths). It compares the label paths of structured information (such as products, quantities, and delivery times) in purchase descriptions. It analyzes and compares the structured components of the data, ensuring that the different hierarchical relationships and label structures in the purchase task descriptions can be identified. Label paths refer to the structured paths of labeled attributes and associations in the data. The knowledge graph logical comparison network is a logical reasoning model based on the knowledge graph. It compares different task descriptions through logical relationships, graphically representing various entities and their relationships. The logical comparison network understands the semantics of different purchase descriptions by comparing entities and their relationships in the graph, ensuring that the logical relationships between tasks can be identified and accurately matched.

[0047] The semantic embedding comparison network converts procurement descriptions (such as product name, quantity, and supplier) into vector representations. Each task description is mapped into a multidimensional semantic space, forming a semantic vector. By calculating the similarity between these vectors, the semantic embedding comparison network automatically identifies and matches identical or similar parts of the description, ensuring the correct understanding of the core information of the procurement task. For example, synonyms such as "laptop" and "notebook PC" are mapped to the same semantic space.

[0048] Structural information (such as product category, quantity, and delivery time) is extracted from purchase descriptions and compared using a structured label path comparison network. This network considers not only the textual content but also the hierarchical relationships between different labels. This ensures that even if the order or presentation of the descriptions differs, the relationships between the elements can still be accurately identified, allowing for effective semantic comparison of purchase tasks.

[0049] The knowledge graph logical comparison network is used to compare entities and their relationships within procurement task descriptions. Knowledge graphs provide a structured representation of different entities (such as products, suppliers, and delivery times) and their relationships. For example, the "product → model → quantity" relationship within the graph can be used to infer interdependencies within procurement tasks. By comparing entities and logical relationships within the graph, the knowledge graph logical comparison network ensures an understanding of the logical relationships and hierarchical structure between tasks, thereby improving matching accuracy.

[0050] Using a semantic embedding comparison network, a structural label path comparison network, and a knowledge graph logic comparison network, the procurement description reconstruction results are semantically mapped and aligned, unifying each procurement task description into a standardized semantic space. Through the multi-layered work of the comparison network, descriptions from different sources and representations are aligned into a consistent standard format.

[0051] After completing the semantic mapping alignment, the procurement tasks in the initial matching candidate library are adapted and reconstructed. The semantic mapping alignment results are input into the initial matching candidate library for adaptive reconstruction. Adaptive reconstruction refers to adjusting or optimizing the candidate tasks based on the aligned semantic information to better match them with the new procurement tasks. The purpose of adaptive reconstruction is to ensure a higher degree of match between each procurement task description and its corresponding product or service by adjusting the procurement task descriptions in the initial candidate library. The reconstructed tasks will better meet actual procurement needs and avoid any previous matching errors. The semantic adaptive reconstruction results obtained after reconstruction include an optimized matching relationship between procurement tasks and products.

[0052] For example, suppose Task 1: Purchase 10,000 Model A laptops, with a delivery date of next month; Task 2: Purchase 15,000 Model B laptops, with a delivery date of next month. The initial matching candidate database contains multiple matching items, including Candidate 1: Model A laptops, price 5,300, inventory 100,000, and delivery date of next month; Candidate 2: Model B laptops, price 6,100, inventory 50,000, and delivery date of next month; and Candidate 3: Model C laptops, price 3,200, inventory 12,000, and delivery date of next month. Based on the model, inventory, and delivery date in the task descriptions, Task 1 is matched with Candidate 1, while Task 2 is matched with Candidate 2.

[0053] The semantic adaptation reconstruction results are sent to the transition decision analysis layer, which performs more complex decision analysis based on the optimized task and product matching results. Synonymous semantic conversion and a semantic embedding comparison network eliminate differences between different representations, ensuring the consistency of procurement task descriptions and accurate matching. Semantic adaptation reconstruction ensures that procurement tasks are optimized based on actual conditions, more closely matching tasks in historical data, thereby improving the quality of procurement decisions.

[0054] Further, as attached Figure 2 As shown, this application also includes the following steps:

[0055] S331: Send the historical data to the behavior analysis layer, and use the behavior analysis layer to reconstruct the procurement chain path diagram; S332: Extract the behavioral preferences and behavioral patterns of each procurement task based on the procurement chain path diagram; S333: Perform behavioral analysis clustering based on procurement tasks, behavioral preferences, and behavioral patterns, and establish behavioral analysis clustering results; S334: Use the behavioral analysis clustering results to adapt and reconstruct the procurement tasks in the initial matching candidate library, establish behavioral adaptation reconstruction results, and send the behavioral adaptation reconstruction results to the transition decision analysis layer.

[0056] Furthermore, this application S332 includes:

[0057] Graph feature extraction is performed on the procurement chain path diagram to establish a graph feature extraction result; feature criticality is identified using the graph feature extraction result to establish key decision features; and after associating the key decision features with tasks in the procurement chain path diagram, behavioral preferences and behavioral patterns are generated.

[0058] Furthermore, this application S333 includes:

[0059] The procurement tasks are time-sorted, a time-sorting result is established, and a time series association is established based on the time-sorting result; the procurement tasks are clustered based on the time series association, and the procurement task clustering result is used to perform behavior analysis clustering based on behavior preferences and behavior patterns to establish a behavior analysis clustering result.

[0060] Specifically, historical data is sent to the behavioral analysis layer, a module dedicated to analyzing historical data and behavioral patterns. By analyzing data such as user behavior and the execution of procurement tasks, it identifies underlying patterns, preferences, and insights. The procurement chain path diagram is a visual representation of procurement tasks and their associated behaviors, decisions, and processes. It depicts the entire process, from procurement demand generation, task assignment, product selection, supplier evaluation, to final purchase completion, as well as the connections and impacts between each step. The behavioral analysis layer reconstructs a procurement chain path diagram based on historical data, analyzing each step of the procurement task execution process, such as demand generation, supplier selection, product screening, pricing decisions, delivery time selection, and payment methods. The procurement chain path diagram accurately understands the relationships between each step and identifies dependencies between tasks and decisions. The procurement chain path diagram is a visual representation of procurement tasks and their associated behaviors, decisions, and processes. It depicts the entire process, from procurement demand generation, task assignment, product selection, supplier evaluation, to final purchase completion, as well as the connections and impacts between each step.

[0061] A procurement chain path graph consists of multiple nodes (such as task nodes, supplier nodes, and product nodes) and edges (such as the selection relationship between tasks and products). Based on the procurement chain path graph, graph feature extraction is performed to extract features relevant to procurement decisions. Graph feature extraction involves extracting meaningful feature data from the procurement chain path graph. These can be nodes, edges, paths, or other graph structural features related to procurement tasks. The graph feature extraction result is a set of structured data obtained through the graph feature extraction method, which reflects the key elements in the procurement chain path graph and the relationships between them. For example, a node may represent a procurement task, while the edges connected to it may represent the selection relationship between the task and a specific product or supplier. Graph feature extraction extracts key information from these nodes and edges, such as node degree (indicating the connectivity of nodes in the graph), path density (indicating the density of connections between nodes), and node attributes (such as product price, delivery time, and supplier rating).

[0062] Based on the results of graph feature extraction, feature criticality is identified, and graph features are screened to identify which features are most critical to procurement decisions. Features with higher criticality have a significant impact on decision-making and may involve factors such as price, delivery time, and supplier quality. A series of characteristic elements are extracted from the procurement chain path diagram, such as the time node of the procurement task, supplier information, product specifications, and price. The extracted features are preprocessed, including data cleaning, formatting, and conversion operations. Statistical methods (such as correlation analysis) are used to evaluate the importance of each feature to the procurement decision. Based on the results of the feature importance assessment, features that have a significant impact on procurement decisions are screened to obtain key decision features. Key decision features refer to features that have a significant impact on task matching, supplier selection, product pricing, and other aspects in procurement decisions, and can usually significantly improve the accuracy and efficiency of procurement decisions.

[0063] Associate key decision features with tasks in the procurement chain path diagram to identify which tasks involve these features in their decision-making processes. By associating tasks and features, the behavioral preferences and patterns of each task can be further analyzed. For example, some procurement tasks may prefer suppliers with short delivery times, while other tasks may prefer suppliers with long-term cooperation. Based on the association between key decision features and procurement tasks, the behavioral preferences and behavioral patterns of each task are generated. Behavioral preferences are tendencies exhibited during the execution of procurement tasks, such as supplier selection and price sensitivity; behavioral patterns are recurring behavioral sequences or patterns during the execution of procurement tasks (such as always submitting orders within a specific time period).

[0064] For example, suppose a company is purchasing office equipment. The historical data includes the following: Task A: Purchase 100 desktop computers, budget 50,000 yuan, and require a 7-day delivery time. Supplier X: Unit price 450 yuan, delivery time 5 days, 90% credibility; Supplier Y: Unit price 430 yuan, delivery time 10 days, 85% credibility. Task B: Purchase 20 high-end monitors, budget 10,000 yuan, and require a 10-day delivery time. Supplier X: Unit price 520 yuan, delivery time 12 days, 95% credibility; Supplier Z: Unit price 480 yuan, delivery time 10 days, 88% credibility. Feature extraction and criticality identification analysis reveal that delivery time and supplier reputation have a significant impact on procurement decisions. In Task A, due to budget constraints and high delivery time requirements, Supplier X is preferred. In Task B, Supplier Z, despite its slightly lower credibility, is preferred based on delivery time and budget. Through graph feature extraction and key decision feature identification, we can accurately identify which features are critical to the procurement decision, thereby improving matching accuracy. By analyzing the behavioral preferences and patterns of tasks, identifying key characteristics and task patterns, and effectively reducing the risks and costs caused by inaccurate procurement decisions.

[0065] Sorting procurement tasks by time means sorting all procurement tasks by their execution time, typically in ascending or descending timestamp order. Time sorting can help understand the order in which procurement tasks occur and uncover underlying patterns, which is particularly important when dynamically analyzing task execution. Each procurement task has an execution time or scheduled time. Task timestamps are used to obtain task timing information. Procurement tasks are sorted by timestamp using common sorting algorithms such as quick sort or merge sort. The result of this time sorting is a time series that reflects the order in which each procurement task occurred.

[0066] Based on the time sorting results, we further analyze the temporal relationships between tasks and identify temporal dependencies between them. This is accomplished by analyzing the time intervals, overlaps, or sequential order of tasks. The completion time of certain procurement tasks can affect the execution of subsequent tasks. For example, the delivery date of Task A can affect the scheduled execution of Task B. We can also identify whether there are cyclical patterns between tasks or demand fluctuations within specific time periods. By establishing temporal relationships, we can identify temporal dependencies between different procurement tasks and uncover underlying patterns in the execution of procurement tasks.

[0067] Procurement tasks are clustered based on temporal associations. Similar tasks are grouped together based on task timing, characteristics, and other information. Clustering algorithms typically use algorithms like K-means and DBSCAN. Tasks with a high frequency of occurrence during specific time periods, such as the beginning of each month or the end of each quarter, are grouped together. Furthermore, clustering can be performed based on other procurement task characteristics, such as budget range, demand volume, and task type.

[0068] Based on the behavioral characteristics and patterns of procurement tasks, behavioral analysis clustering is performed to further subdivide procurement tasks into different behavioral pattern groups. This is accomplished by analyzing the behavioral characteristics of procurement tasks within each cluster, such as supplier selection and price sensitivity. The main purpose of behavioral analysis clustering is to discover behavioral patterns and preferences in procurement tasks. For example, whether there is a tendency to choose low-priced products under budget constraints, or whether products from reputable suppliers are selected under time constraints. Some tasks may always be completed within a certain time period, while others may be executed under certain specific conditions (such as when the budget reaches a certain amount).

[0069] Behavioral clustering, based on historical data analysis of procurement tasks, identifies behavioral characteristics exhibited by different tasks during execution (such as task completion speed and supplier selection preferences), and groups tasks according to these characteristics. The results of behavioral clustering generate distinct task categories, each representing a potential behavioral pattern. For example, a company's annual office equipment procurement process identified the following tasks: Task A: Purchase 100 desktop computers in January of each year, with a budget of 50,000 yuan and a delivery time of 15 days; Task B: Purchase 50 printers in March of each year, with a budget of 30,000 yuan and a delivery time of 20 days; Task C: Purchase 200 laptops in July of each year, with a budget of 100,000 yuan and a delivery time of 10 days. By sorting these tasks in time, tasks A, B, and C are concentrated in different time periods: January, March, and July, respectively. Temporal correlation analysis reveals a strong temporal dependency between the execution times of tasks A and B, as both tasks are part of the annual supplemental office equipment procurement process. Clustering procurement tasks based on time and task characteristics revealed that Tasks A and B had high similarities in budget range and equipment type, and therefore could be grouped together. However, Task C, with its different budget and procurement scale, fell into a different category. During the behavioral analysis clustering phase, the behavioral preferences of Tasks A, B, and C were analyzed. It was found that Tasks A and B tended to favor low-price suppliers, while Task C preferred products from reputable suppliers. The following behavioral analysis clustering results were generated: Category 1: Limited Budget (A, B): Choose low-price suppliers; Category 2: High Budget (C): Choose reputable suppliers.

[0070] Through behavioral analysis and clustering, the behavioral preferences and patterns of each procurement task are identified. For example, some tasks may prefer low-cost suppliers, others may prioritize delivery time, and some tasks may even prioritize the supplier's service quality or product quality during the procurement process. Based on the behavioral analysis clustering results, the procurement tasks in the initial matching candidate library are adapted and reconstructed to create behavioral adaptation and reconstruction results. Based on the behavioral analysis clustering results, suppliers that match the behavioral patterns of specific tasks are screened. Based on the demand characteristics of each cluster category, the matching of procurement tasks and products in the candidate library is optimized. Based on the matching rules and the optimized candidate library, tasks are re-matched with products and suppliers. The results of behavioral adaptation and reconstruction are saved and a new procurement task list is generated, effectively reducing unnecessary procurement risks and improving procurement efficiency. For example, for a high-priority, budget-constrained procurement task, priority will be given to matching suppliers with low prices and short delivery times to ensure cost control and timely delivery of the task.

[0071] The behavioral adaptation and reconstruction results are fed into the transition decision analysis layer, the core decision-making layer of the procurement service system. This layer is responsible for integrating all data and optimization results to make the final procurement decision. The transition decision analysis layer ultimately generates the optimal procurement plan and outputs this information to the procurement personnel or the execution layer of the procurement service system. By analyzing historical data, the behavioral characteristics and requirements of each procurement task are identified, ensuring that the task is matched with the most suitable supplier and product. This significantly improves the efficiency and accuracy of procurement task matching and avoids the errors and inefficiencies inherent in traditional matching.

[0072] Furthermore, this application S334 includes:

[0073] A transition judgment threshold of the transition decision analysis layer is configured based on the initial matching candidate library, and the transition judgment threshold is the transition threshold from the candidate library and the taboo library to the matching library in the mapping matching process; after receiving the semantic adaptation reconstruction result and the behavioral adaptation reconstruction result, the matching value reconstruction of the initial matching candidate library is performed, and the transition judgment threshold is used to perform the reconstructed matching value transition judgment; and the procurement task matching reconstruction is completed according to the transition judgment result.

[0074] Perform matching value standard deviation calculation on the initial matching candidate library to generate a calculation result; configure a first dynamic adjustment factor and a second dynamic adjustment factor, wherein the first dynamic adjustment factor is a transition adjustment factor from the candidate library to the matching library, and the second dynamic adjustment factor is a transition adjustment factor from the taboo library to the matching library; establish a matching sliding window, use the matching sliding window to perform matching focus extraction on the initial matching candidate library, and establish a matching focus extraction result; configure a transition determination threshold of a transition decision analysis layer using the matching focus extraction result, the first dynamic adjustment factor, the second dynamic adjustment factor, and the calculation result.

[0075] Specifically, the matching values ​​of all procurement tasks and products are extracted from the initial matching candidate pool, and the average of all matching values ​​is calculated as the central tendency indicator of the matching value distribution. For each matching value, the square of the difference between the matching value and the average is calculated, then summed, and finally divided by the total number of matching values. The square root is taken to obtain the standard deviation, which reflects the degree of dispersion of the matching value distribution. The purpose of calculating the standard deviation is to measure the degree of dispersion of these matching values, that is, the consistency of the matching results. If the standard deviation of the matching values ​​is low, it means that the matching results of tasks and products are relatively stable; if the standard deviation is high, it means that the matching stability is poor, and the matching strategy or threshold may need to be adjusted. For example, suppose there are five matching values: 0.9, 0.85, 0.92, 0.88, and 0.91, with a mean of 0.892 and a calculated standard deviation of 0.02.

[0076] Configure the first dynamic adjustment factor and the second dynamic adjustment factor. The first dynamic adjustment factor is used to adjust the transition process from the candidate library to the matching library, that is, according to the stability of the current matching value, dynamically adjust how the tasks in the candidate library are moved to the matching library, and determine the sensitivity of the candidate tasks entering the matching library; the second dynamic adjustment factor is used to adjust the transition process from the taboo library to the matching library. Tasks in the taboo library are temporarily not considered for some reasons, and the second dynamic adjustment factor determines the possibility of reconsideration of these tasks.

[0077] Establish a matching sliding window. This is a dynamic data processing method used to gradually analyze and adjust data. The matching sliding window is used to extract matching points of interest from the initial matching candidate pool, focusing on the system's current state. The threshold is flexibly configured based on system performance. The matching sliding window scans the matching results in the initial matching candidate pool, extracts key features or points of interest related to the matching task, calculates matching point extraction results based on the current matching state, and adjusts the threshold based on these results.

[0078] Configure the transition decision analysis layer's transition judgment threshold based on the matching attention extraction results, the first dynamic adjustment factor, the second dynamic adjustment factor, and the matching value standard deviation calculation results. The transition judgment threshold is a parameter set in the transition decision analysis layer to determine whether a task or product can transition from the candidate library or taboo library to the matching library. The transition judgment threshold is dynamically adjusted and configured based on the previous matching attention extraction results, dynamic adjustment factors, and calculation results. For example, in a certain calculation, if the matching attention extraction results indicate that the candidate task has a good matching status and the first dynamic adjustment factor is large, set the transition judgment threshold to a lower value (for example, 0.75) to ensure that the candidate task can more easily transition to the matching library.

[0079] After receiving the semantic adaptation reconstruction results and the behavioral adaptation reconstruction results, the matching values ​​in the initial matching candidate pool are recalculated based on these results. By re-evaluating the matching score between each candidate and the procurement task, the matching candidates are determined to be worthy of consideration as final matches. The reconstructed matching values ​​are evaluated using the configured transition judgment threshold. If a matching item's reconstructed score exceeds the threshold, it is considered a valid match and enters the matching pool, becoming the final matching product for the procurement task. If the score falls below the threshold, it remains in the candidate pool or taboo pool and does not enter the final matching results. After the transition judgment, the matching reconstruction of the procurement task is completed, generating the final matching relationship between the procurement task and the procurement product, i.e., the matching reconstruction result. Through semantic adaptation and behavioral adaptation reconstruction, the procurement task is precisely matched with the appropriate product, improving matching accuracy. Flexible configuration of the transition judgment threshold allows dynamic adjustment of matching criteria based on different matching situations, making procurement decisions more intelligent and personalized.

[0080] S400: Procurement management is performed according to the matching and reconstruction result.

[0081] Specifically, the procurement management system optimizes the entire procurement process based on the matching reconstruction results. The relationship between each procurement task and the best-matching product has been clearly defined, enabling procurement managers to make more accurate and efficient decisions. Based on the matching reconstruction results, each procurement task is associated with its best-matching product, further broken down into specific purchase orders, and scheduled based on factors such as supplier, inventory status, and delivery time. The procurement management system automatically allocates tasks based on product compatibility and task priority to ensure efficient completion of procurement tasks. The matching reconstruction results also help the procurement management system select the supplier that best meets procurement requirements based on behavioral analysis of historical data. Supplier selection is optimized based on past performance (e.g., on-time delivery, quality, and price), ensuring the best match between procurement tasks and products, thereby improving supply chain efficiency and reliability. The matching reconstruction results accurately identify the best-matching products and suppliers, enabling the generation of appropriate procurement contracts and ensuring that the terms between supplier and purchaser best meet the needs of both parties.

[0082] During the procurement process, the matching reconstruction results will also affect inventory management, and dynamic resource allocation will be carried out according to the priority of the procurement task, the supply of products, and the existing inventory level. If the matching degree of certain products is very high, these products need to be purchased in advance to prevent insufficient inventory, while ensuring that there is not too much backlog of resources in the inventory. Procurement management is not just about executing procurement tasks, but also includes the continuous monitoring and optimization of the procurement process. Based on the matching reconstruction results, the execution status of each procurement task is monitored in real time through data analysis, and adjustments are made to the links where problems occur. For example, if the matching degree of a procurement task is low, the supplier or product is re-evaluated to avoid wrong purchases. Through automated decision-making based on matching reconstruction results, the procurement management system can significantly improve the efficiency and accuracy of procurement decisions and reduce manual intervention and errors.

[0083] Furthermore, the present application S400 includes:

[0084] Establishing a collection evaluation mapped to the matching and reconstruction result, generating a feedback database using the collection evaluation; and performing channel optimization management of the multi-layer matching channel using the feedback database.

[0085] Specifically, evaluations of the matching reconstruction results are collected after application. The actual performance of each match is recorded, and evaluation results are generated, including quantitative indicators of matching accuracy and user satisfaction surveys. Scores are generated for each evaluation item and recorded in a feedback database. As a place to store and manage evaluation results, the feedback database provides important data support for subsequent operations, helping to identify areas of optimization within the matching process. For example, a procurement task requires 500 laptops with processors of i7 or higher, 16GB of RAM, and screen sizes of 14-15 inches. The matching products are: Product A: i7, 16GB of RAM, 14-inch screen, price $900; Product B: i7, 8GB of RAM, 15-inch screen, price $850; Product C: i5, 16GB of RAM, 14-inch screen, price $800. The matching reconstruction results recommend Product A and Product B, with Product A having the highest match (90%), but Product B having a lower price (15% discount). Using evaluation indicators (matching accuracy, time efficiency, cost-effectiveness, and user satisfaction), the matching degree of Product A was 90%, and the matching degree of Product B was 85%. It took 2 minutes to match the candidate products. Users were satisfied with the quality and price of Product A (scoring 9 / 10), but gave a higher evaluation of the cost-effectiveness of Product B (scoring 8 / 10).

[0086] The feedback database stores feedback from various stages (such as matching reconstruction, supplier selection, and procurement tasks). It records historical data and dynamically updates it. All evaluation data regarding the matching process is recorded in real time in the feedback database, serving as a basis for subsequent optimization. Evaluation results are automatically updated with each procurement task. Feedback from each procurement task is recorded in real time, gradually improving matching accuracy.

[0087] Optimize the multi-layer matching pipeline based on data from the feedback database. If the semantic matching layer's matching accuracy is low, improve its performance by adjusting the algorithm model or re-evaluating the weights of semantic features. Similarly, the parameters of the behavioral analysis layer or transition decision analysis layer can be dynamically adjusted based on feedback data to better suit current procurement needs. Optimization management is not limited to algorithm adjustments but can also involve dynamic configuration adjustments. For example, the characteristics of certain procurement tasks may be more similar to behavioral patterns in historical data. Automatically adjust the matching algorithm based on these changes to improve matching efficiency.

[0088] Supported by the feedback database, multiple matching channels (such as the semantic matching layer, behavioral analysis layer, and transition decision analysis layer) can collaborate and share optimization information in real time. For example, optimizations in the semantic matching layer can be combined with the results of the behavioral analysis layer to jointly adjust the matching process. Through continuous feedback collection and optimization, the multi-layer matching channel can develop a certain degree of self-learning ability, continuously improving matching efficiency.

[0089] Furthermore, the present application further comprises the following steps:

[0090] Perform matching warning identification on the matching reconstruction result and establish an abnormal early warning output; generate a new extended matching suggestion based on the abnormal early warning output, and feed back the new extended matching suggestion to the management user.

[0091] Specifically, the matching reconstruction results are analyzed to identify potential anomalies or unexpected matching results. Warnings may arise from issues such as low product matching, excessive costs, or long delivery times, typically requiring further analysis and optimization. When anomalies are detected in the matching reconstruction results, an alert is issued through the early warning mechanism. These anomalies may manifest as poor matching quality, mismatch with procurement requirements, or failure to meet procurement timelines. For example, a procurement requirement is to purchase 100 high-performance servers, requiring brand A, 16GB of memory, 4TB of hard drives, and a delivery time of no more than three weeks. Based on the matching reconstruction results, Supplier B and Supplier C are recommended. However, Supplier B's delivery time is four weeks, exceeding the procurement timeline requirements. Supplier C's server brand does not meet the requirements (it is not Brand A). Anomaly warnings are issued, indicating that the matching results do not meet the procurement timeline and brand requirements, Supplier B's delivery time exceeds the procurement requirements, and Supplier C's brand does not meet the requirements. Recommendations include searching for new suppliers or expanding the matching scope to find other suppliers that meet the brand and delivery requirements.

[0092] Based on the results of anomaly alerts, new matching suggestions—additional and expanded matching suggestions—are generated. These suggestions may include selecting alternative potential suppliers or product types, or modifying procurement requirements to address deficiencies in the current matching process and ensure the smooth completion of procurement tasks. These new and expanded matching suggestions are then communicated to management or decision-makers to help adjust procurement strategies or make informed decisions. This early warning mechanism effectively identifies anomalies in the matching process, such as delivery delays and brand discrepancies, and promptly notifies management.

[0093] In summary, the hierarchical information deep mining and matching method applied to the procurement service system provided by this application has the following beneficial effects:

[0094] By performing data interaction with the procurement service system, a multi-source dataset is established; the surface information recognition layer is called to perform data preprocessing and recognition of the multi-source dataset, and an initial matching candidate library is established. The initial matching candidate library is a mapping and matching database between procurement tasks and procurement products, and the initial matching candidate library includes a matching library, a candidate library, and a taboo library; the preprocessed multi-source dataset and the initial matching candidate library are input into a multi-layer matching channel, and after configuring the multi-layer matching channel using historical data in the preprocessed multi-source dataset, procurement task matching reconstruction based on the initial matching candidate library is performed to generate a matching reconstruction result. The multi-layer matching channel includes a semantic matching layer, a behavioral analysis layer, and a transition decision analysis layer; procurement management is performed based on the matching reconstruction result. In other words, by collecting multi-source datasets, establishing an initial matching candidate library, and configuring a multi-layer matching channel, procurement tasks and products are accurately matched from multiple dimensions, the matching results are optimized, and the final procurement decision is generated, thereby improving procurement efficiency.

[0095] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present application. Various modifications to these embodiments will be readily apparent to one 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 application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

[0096] Obviously, for those skilled in the art, several improvements and modifications can be made to the present application without departing from the principles of the present application, and these improvements and modifications also fall within the scope of protection of the present application.

Claims

1. A hierarchical information deep mining and matching method applied to a procurement service system, characterized in that: include: Perform data interaction with procurement service systems to build multi-source data sets; Calling the surface information recognition layer to perform data preprocessing and recognition of the multi-source data set to establish an initial matching candidate library, which is a mapping matching database between procurement tasks and procurement products. The initial matching candidate library includes a matching library, a candidate library, and a taboo library; Inputting the preprocessed multi-source dataset and the initial matching candidate library into a multi-layer matching channel, configuring the multi-layer matching channel using historical data in the preprocessed multi-source dataset, performing procurement task matching reconstruction based on the initial matching candidate library, and generating a matching reconstruction result, the multi-layer matching channel including a semantic matching layer, a behavior analysis layer, and a transition decision analysis layer; Performing procurement management based on the matching and reconstruction results; Performing procurement task matching reconstruction based on the initial matching candidate library includes: Acquire preprocessed historical data from the multi-source data set; Sending the historical data to the semantic matching layer, performing synonymous semantic conversion using the semantic matching layer, and then performing procurement description reconstruction; Using the semantic embedding comparison network, the structural label path comparison network, and the knowledge graph logic comparison network in the semantic matching layer to perform semantic mapping alignment of the procurement description reconstruction results; Adapting and reconstructing the procurement tasks in the initial matching candidate library according to the semantic mapping alignment result, establishing a semantic adaptation and reconstruction result, and sending the semantic adaptation and reconstruction result to the transition decision analysis layer; Executing procurement task matching reconstruction based on the initial matching candidate library also includes: Sending the historical data to the behavior analysis layer to reconstruct a procurement chain path diagram using the behavior analysis layer; Extracting behavioral preferences and behavioral patterns for each procurement task based on the procurement chain path diagram; Conduct behavioral analysis and clustering based on purchasing tasks, behavioral preferences, and behavioral patterns, and establish behavioral analysis clustering results; Adapting and reconstructing the procurement tasks in the initial matching candidate library using the behavior analysis clustering results, establishing a behavior adaptation and reconstruction result, and sending the behavior adaptation and reconstruction result to the transition decision analysis layer; After the behavior adaptation reconstruction result is sent to the transition decision analysis layer, the method includes: Configuring a transition determination threshold of a transition decision analysis layer based on the initial matching candidate library, wherein the transition determination threshold is a transition threshold from a candidate library and a taboo library to a matching library during a mapping and matching process; After receiving the semantic adaptation reconstruction result and the behavioral adaptation reconstruction result, performing matching value reconstruction of the initial matching candidate library, and performing transition determination of the reconstructed matching value using the transition determination threshold; Complete the procurement task matching reconstruction based on the transition judgment results.

2. The hierarchical information deep mining and matching method applied to a procurement service system according to claim 1, characterized in that: The configuring the transition determination threshold of the transition decision analysis layer based on the initial matching candidate library includes: Calculating the standard deviation of matching values ​​for the initial matching candidate library to generate a calculation result; Configure a first dynamic adjustment factor and a second dynamic adjustment factor, wherein the first dynamic adjustment factor is a transition adjustment factor from the candidate library to the matching library, and the second dynamic adjustment factor is a transition adjustment factor from the taboo library to the matching library; Establishing a matching sliding window, extracting matching interests from an initial matching candidate library using the matching sliding window, and establishing a matching interest extraction result; The transition determination threshold of the transition decision analysis layer is configured using the matching attention extraction result, the first dynamic adjustment factor, the second dynamic adjustment factor, and the calculation result.

3. The hierarchical information deep mining and matching method applied to a procurement service system according to claim 1, characterized in that: Extracting the behavioral preferences and behavioral patterns of each procurement task based on the procurement chain path diagram includes: Performing graph feature extraction on the procurement chain path graph to establish a graph feature extraction result; Identify feature criticality based on the graph feature extraction results to establish key decision features; After associating the key decision features with the tasks in the procurement chain path diagram, behavioral preferences and behavioral patterns are generated.

4. The hierarchical information deep mining and matching method applied to a procurement service system according to claim 3, characterized in that: The behavior analysis clustering is performed based on the purchasing tasks, behavior preferences, and behavior patterns, and the behavior analysis clustering results are established, including: sorting the procurement tasks by time, establishing a time sorting result, and establishing a time sequence association based on the time sorting result; The procurement tasks are clustered according to the temporal association, and the procurement task clustering results are used to perform behavioral analysis clustering under behavioral preferences and behavioral patterns to establish behavioral analysis clustering results.

5. The hierarchical information deep mining and matching method applied to a procurement service system according to claim 1, characterized in that: The performing procurement management according to the matching and reconstruction result includes: Establishing a collection evaluation mapped to the matching and reconstruction result, and generating a feedback database using the collection evaluation; The feedback database is used to perform channel optimization management of the multi-layer matching channel.

6. The hierarchical information deep mining and matching method applied to a procurement service system according to claim 1, characterized in that: The method of configuring a multi-layer matching channel using the pre-processed historical data in the multi-source data set includes: Establishing an identification rule for abnormal tasks, screening the historical data based on the identification rule, and establishing an abnormal task identifier; Data diversion of historical data is performed according to the abnormal task identifier, and multi-layer matching channel configuration is completed according to the data diversion result.

7. The hierarchical information deep mining and matching method applied to a procurement service system according to claim 1, characterized in that: The performing procurement management according to the matching reconstruction result further includes: Perform matching warning identification on the matching reconstruction results and establish abnormality early warning output; A new extended matching suggestion is generated based on the abnormal early warning, and the new extended matching suggestion is fed back to the management user.

Images