Knowledge graph-based bidding subject matter semantic matching analysis method and system

Abstract

The application discloses a knowledge graph-based bidding subject matter semantic matching analysis method and system, and relates to the technical field of semantic matching analysis. The method comprises the following steps: obtaining a subject matter list and a subject matter knowledge graph of a to-be-bid project; based on the semantic path distance between each subject matter node in the subject matter knowledge graph, analyzing and obtaining the delivery capacity matching value of each supplier for each subject matter; based on historical bid-winning records, calculating the capacity coverage difference of each supplier; based on the bid-winning frequency and bid-winning time attenuation characteristics of each subject matter in the historical bid-winning records of each supplier, calculating the historical active concentration degree of each supplier; based on the delivery capacity matching value, the capacity coverage difference and the historical active concentration degree, calculating the comprehensive matching correlation degree of each supplier, and obtaining the subject matter semantic correlation analysis result of the to-be-bid project and each supplier. The application solves the technical problem that the existing bidding matching method cannot mine the implied semantic correlation between subject matters.

Description

Technical Field

[0001] This invention relates to the field of semantic matching analysis technology, specifically to a semantic matching analysis method and system for bidding documents based on knowledge graphs. Background Technology

[0002] In bidding activities, projects released by the bidding party typically include multiple items, requiring the selection of the most capable partner from numerous bidding suppliers. Existing supplier matching methods mainly rely on manual experience to compare suppliers' historical bidding records item by item, or use keyword matching to search whether a supplier has undertaken projects with the same name.

[0003] This type of method has the following shortcomings: The name of the subject matter can be expressed in multiple ways in the bidding documents; the same type of subject matter may appear in different bidding projects with different conceptual names at different levels. For example, "server" and "computing equipment" belong to different expressions at the same semantic level, and text matching alone cannot identify this implicit semantic relationship. Factors such as the range of project types undertaken by the supplier and the level of project execution activity are difficult to comprehensively assess manually, leading to the final matching result relying on the subjective experience of the evaluators and lacking a systematic quantitative analysis method. Summary of the Invention

[0004] The invention provides a semantic matching analysis method and system for bidding and tendering targets based on knowledge graphs, which solves the technical problem that existing bidding and tendering matching methods are based only on text matching and human experience evaluation, and cannot use the semantic hierarchical relationship between the targets for systematic quantitative analysis.

[0005] In a first aspect, the present invention provides a semantic matching analysis method for bidding documents based on knowledge graphs, the method comprising: Obtain a list of target items for the project to be tendered, and obtain a target item knowledge graph, wherein the target item knowledge graph contains semantic hierarchical associations between target item nodes; Obtain the historical bidding records of each supplier, and analyze and obtain the delivery capability matching value of each supplier for each target based on the semantic path distance between each target node in the target knowledge graph. Based on the historical bidding records, calculate the capability coverage difference of each supplier; Based on the bidding frequency and bidding time decay characteristics of each target item in the historical bidding records of each supplier, the historical active focus of each supplier is calculated; Based on the delivery capability matching value, the capability coverage difference, and the historical active focus, the comprehensive matching correlation degree of each supplier is calculated, and the semantic correlation analysis results of the target items between the project to be tendered and each supplier are obtained.

[0006] Secondly, the present invention also provides a semantic matching and analysis system for bidding and tendering objects based on knowledge graphs, the system comprising: The graph acquisition module is used to acquire a list of target items for the project to be tendered and to acquire a target item knowledge graph, wherein the target item knowledge graph contains the semantic hierarchical association relationship between target item nodes. The capability matching calculation module is used to obtain the historical bidding records of each supplier, and analyze and obtain the delivery capability matching value of each supplier for each target based on the semantic path distance between each target node in the target knowledge graph. The coverage difference calculation module is used to calculate the capability coverage difference of each supplier based on the historical bidding records. The active focus calculation module is used to calculate the historical active focus of each supplier based on the bidding frequency and bidding time decay characteristics of each target item in the historical bidding records of each supplier. The comprehensive correlation calculation module is used to calculate the comprehensive matching correlation of each supplier based on the delivery capability matching value, the capability coverage difference, and the historical active focus, and to obtain the semantic correlation analysis results of the target objects between the project to be tendered and each supplier.

[0007] One or more technical solutions provided in this invention have at least the following technical effects or advantages: First, this invention constructs a knowledge graph of the target, forming a semantic hierarchical association structure based on the superior, subordinate, and peer relationships between target nodes. It quantifies the semantic similarity between the target items in the supplier's historical bids and the target items of the project to be bid based on semantic path distance to obtain a delivery capability matching value. Compared with existing methods that only rely on text matching and manual experience to compare the target item names one by one, this invention can uncover the implicit semantic associations between concept names at different levels, thus expanding the scope of the supplier's deliverable capability assessment.

[0008] Second, this invention calculates the supplier's capability coverage difference, using the ratio of the types of goods covered by the supplier's historical winning bids to the types of goods required in this demand as explicit coverage, and the weighted sum of delivery capability matching values ​​as the solution delivery capability matching value. The deviation between the two reflects the supplier's potential for cross-domain capability expansion. Compared with existing methods that only count the number of historical winning bids, this invention can more comprehensively measure the supplier's actual coverage capability for this demand.

[0009] Third, this invention calculates the historical activity focus of each supplier by the frequency of winning bids and a decay factor based on the time decay index. Recent winning bid records receive a higher time decay weight, while the weight of long-term records decays accordingly. Compared with existing methods that treat all historical records equally, this invention can reflect the current activity level of suppliers in the target product field and avoids evaluation bias caused by the accumulation of long-term performance. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating the semantic matching analysis method for bidding targets based on knowledge graphs provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the semantic matching analysis method for bidding targets based on knowledge graphs provided in this embodiment of the invention. Figure 3 This is a structural diagram of the knowledge graph-based semantic matching analysis system for bidding and tendering objects provided in this embodiment of the invention; The diagram is labeled as follows: Graph acquisition module 11, Ability matching calculation module 12, Coverage difference calculation module 13, Activity and focus calculation module 14, and Comprehensive correlation calculation module 15. Detailed Implementation

[0012] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0013] Example 1, as Figure 1 As shown, this invention provides a flowchart illustrating a semantic matching analysis method for bidding documents based on knowledge graphs; as... Figure 2 As shown, this invention provides a logical diagram of a semantic matching analysis method for bidding and tendering targets based on knowledge graphs. The method includes: S100: Obtain a list of target items for the project to be tendered, and obtain a target item knowledge graph, wherein the target item knowledge graph contains semantic hierarchical associations between target item nodes; In bidding activities, the tender documents issued by the tendering party typically contain multiple items, with the names of each item recorded in text form. Suppliers need to demonstrate their historical delivery capabilities for each item when submitting their bids. However, the names of items in a supplier's past winning bids may not be entirely consistent with the wording in the current tender document. If only text matching is used to compare item names, a large number of semantically highly related but worded items will be missed, leading to an incomplete assessment of the supplier's delivery capabilities. Item knowledge graphs, by constructing hierarchical relationships (superordinate, subordinate, and equivalent) between item nodes, organically link item names scattered across different bidding projects according to semantic hierarchy.

[0014] Step S100 provided in this embodiment of the invention includes: Extract a list of subject matter names from the tender documents of the project to be tendered, and use it as the subject matter list, wherein the subject matter list contains at least one subject matter name; From the target knowledge graph, obtain the target node corresponding to each target name in the target list, and the semantic hierarchical association between each target node.

[0015] The semantic hierarchical relationships include superordinate relationships, subordinate relationships, and peer relationships. The superordinate relationship indicates that one target node is a general concept of another target node. The subordinate relationship indicates that one target node is a specific subclass of another target node. The peer relationship indicates that two target nodes belong to parallel subclasses under the same superordinate concept.

[0016] The specific implementation method is as follows: First, a list of item names is extracted from the tender documents of the project to be tendered, forming the item list. The tender documents are the official procurement documents issued by the tendering party, containing basic project information and technical specifications. The item name list is extracted from the item list section or technical specifications of the tender documents. The extraction method involves text parsing of the tender documents, identifying the item list tables or item name paragraphs, and extracting each item name one by one. After extraction, the item names are standardized, removing redundant spaces and punctuation marks, and conforming to the standardized name format used in the item knowledge graph. For example, the tender documents of a project to be tendered list three items: a computing server, storage devices, and a network switch. After standardization, the item list contains the names of these three items.

[0017] Then, the target item nodes corresponding to each target item name in the target item list are obtained from the target item knowledge graph. The target item knowledge graph is a directed acyclic graph structure built with target item names from historical bidding announcements as nodes and semantic hierarchical relationships as edges. Each target item node stores the standardized name of the target item. Each target item name in the target item list is matched with the node name in the target item knowledge graph using exact string matching. For a successfully matched target item name, the corresponding target item node and its position in the graph are obtained; for a target item name that fails to match, the name is temporarily stored as a new target item node for subsequent knowledge graph updates.

[0018] The semantic hierarchical relationships between successfully matched target nodes are retrieved. These relationships are searched along the edges between nodes in the target knowledge graph, and include three types: superior relationships, subordinate relationships, and homologous relationships. A superior relationship indicates that one target node is a general concept of another target node; for example, "server" is a superior concept of "computing server." A subordinate relationship indicates that one target node is a specific subclass of another target node; for example, "computing server" is a subordinate concept of "server." A homologous relationship indicates that two target nodes belong to parallel subclasses under the same superior concept; for example, "computing server" and "storage server" are both subordinate concepts of "server," and they are homologous. These semantic hierarchical relationships constitute the connectivity paths between nodes in the target knowledge graph, and subsequent steps calculate the semantic path distance based on these connectivity paths.

[0019] For example, in the target knowledge graph, the parent node of the target node "Computing Server" is "Server", the lower node is empty, and the sibling nodes are "Storage Server" and "Network Server". The parent node of the target node "Storage Device" is "Data Storage Device", and the lower nodes include "Disk Array" and "Solid State Storage Device". The parent node of the target node "Network Switch" is "Network Device", and the lower nodes include "Core Switch" and "Access Switch".

[0020] Step S100 provided in this embodiment of the invention further includes: The construction of the target knowledge graph includes: Collect the names of the subject matter from historical bidding announcements to form a set of subject matter names; Entity recognition and semantic hierarchical classification are performed on each object name in the object name set, and the superior, inferior and co-existing relationships between each object are extracted; Using the names of each target as nodes and the superior, subordinate, and peer relationships as edges, a directed acyclic graph structure is constructed as the target knowledge graph. Each target node stores the standardized name of the target, and each edge stores the type of semantic hierarchical association.

[0021] The specific implementation method is as follows: First, collect the names of the tendered items from historical bidding announcements to form a set of tendered item names. These announcements are sourced from publicly available bidding information platforms or internal bidding management systems. Extract all tendered item names from the list of tendered items in each historical announcement, standardize them by removing redundant spaces and punctuation, and merge different spellings of the same item. Finally, deduplicate all standardized tendered item names and compile them into a single set of tendered item names.

[0022] Entity identification and semantic hierarchical classification are performed on each item name in the set of item names. Entity identification employs a combination of rule matching and statistical feature extraction to extract core terms and modifiers from the item names. Core terms represent the semantic category of the item, while modifiers represent attributes such as specifications, uses, and brands. Semantic hierarchical classification assigns item names to a pre-defined classification system based on the hierarchical relationships between core terms. The hierarchical structure of the classification system is determined with reference to the classification standards for bidding items, such as the classification hierarchy in the government procurement item classification catalog.

[0023] Extract the hierarchical, subordinate, and equivalent relationships between the target items. A hierarchical or subordinate relationship exists between two target item nodes if and only if one item's classification level is one level higher than the other's, and they have a direct inclusion relationship in their core terms. For example, in a classification hierarchy, "server" is a first-level category, and "computing server" is a second-level category; a hierarchical edge (from computing server to server) and a subordinate edge (from server to computing server) are established between them. An equivalent relationship exists between two target item nodes if and only if both items belong to the same hierarchical category and are at the same classification level within that category. For example, "computing server" and "storage server" both belong to the second-level category of "server," and an equivalent edge is established between them.

[0024] Using the names of each target item as nodes and superior, subordinate, and peer relationships as edges, a directed acyclic graph (DAG) structure is constructed as the target item knowledge graph. Each target item node stores the standardized name of the target item, and each node can include attributes such as the frequency of the target item's appearance in historical bidding announcements and the most recent appearance time. Each edge stores the type of semantic hierarchical relationship, i.e., superior, subordinate, or peer relationship. The constraints of the DAG ensure that the semantic hierarchical relationships between targets do not form closed loops, guaranteeing the correctness of subsequent semantic path distance calculations.

[0025] For example, the target knowledge graph contains several target nodes. The target node "Server" has subordinate edges connecting to "Computing Server," "Storage Server," and "Network Server," while its superior edges are empty. The target node "Computing Server" has a superior edge connecting to "Server," and a corresponding edge connecting to "Storage Server" and "Network Server." The target node "Data Storage Device" has subordinate edges connecting to "Disk Array," "Solid State Storage Device," and "Storage Server," while its superior edges are empty. The target node "Storage Server" has a superior edge connecting to both "Server" and "Data Storage Device," indicating that "Storage Server" belongs to the subordinate concept of two superior categories simultaneously.

[0026] The following technical effects were achieved through this step: By constructing a knowledge graph of the subject matter, the scattered subject matter names in historical bidding announcements are organized into a directed acyclic graph structure according to hierarchical, subordinate, and co-positional relationships. This establishes semantic association paths between subject matter at different levels and with different expressions, providing a quantifiable semantic distance metric basis for the subsequent calculation of delivery capability matching values.

[0027] S200: Obtain the historical bidding records of each supplier, and based on the semantic path distance between each target node in the target knowledge graph, analyze and obtain the delivery capability matching value of each supplier for each target. This step starts with the target item nodes in the target item list and ends with the winning target item nodes in the historical bidding records of each supplier. It calculates the shortest semantic path distance in the target item knowledge graph and converts the semantic path distance into a delivery capability matching value, providing a semantic-level quantitative indicator of delivery capability for subsequent comprehensive matching correlation calculation.

[0028] Step S200 provided in this embodiment of the invention includes: Obtain the historical bidding records of each supplier, wherein the historical bidding records include the project number and the name of the subject matter of the historical bidding projects of the supplier; For each item in the list of items, in the item knowledge graph, starting from the item node and ending at each winning item node in the historical bidding records of each supplier, the number of edges traversed by the shortest semantic path between the starting point and each ending point is calculated as the semantic path distance. Based on the semantic path distance, the single-point matching value of the supplier for the subject matter of the project to be tendered is calculated and obtained, wherein when the semantic path distance is zero, the single-point matching value is taken as the preset maximum matching value; Based on the single-point matching value of the target product, the supplier's delivery capability matching value for the target product is calculated and obtained.

[0029] The specific implementation method is as follows: First, obtain the historical bidding records of each supplier. These records are sourced from the bidding management system or the bidding announcement database. Each historical bidding record contains the supplier's unique identifier, the project number of the historical bidding project, and the name of the historical bidding item. Group all historical bidding records of each supplier by supplier identifier, with each supplier corresponding to a list of historical bidding item names. For example, supplier S1's historical bidding records contain item names such as "server," "disk array," "network switch," and "computing device," while supplier S2's historical bidding records contain item names such as "storage server," "core switch," and "solid-state storage device."

[0030] For each item in the item list, the corresponding item node is determined in the item knowledge graph, serving as the starting point for semantic path calculation. For each supplier, the list of historically won item names is traversed, and the item node corresponding to each historically won item name is found in the item knowledge graph, serving as the endpoint set for semantic path calculation.

[0031] In the object knowledge graph, the shortest semantic path distance is calculated by taking the starting point as the source and each endpoint as the target. The object knowledge graph is a directed acyclic graph, where nodes represent object names and edges represent hierarchical, subordinate, and sibling relationships. The calculation of the shortest semantic path uses a breadth-first search algorithm, expanding outwards layer by layer from the starting point, visiting adjacent nodes directly connected to the current node until the target endpoint is reached for the first time. The number of edges traversed when the target endpoint is reached for the first time is the shortest semantic path distance. In a directed acyclic graph, the breadth-first search algorithm guarantees that the path visited for the first time is the shortest path.

[0032] Semantic path distance reflects the semantic proximity between the starting and ending targets. A semantic path distance of zero indicates that the supplier's historically won bids and the target of this tender are from the same node, meaning the supplier has completely relevant delivery experience. A semantic path distance of 1 indicates that the supplier's historically won bids and the target of this tender have a direct superior, subordinate, or equivalent relationship, meaning the supplier has highly relevant delivery experience. The larger the semantic path distance, the greater the semantic distance between the two, and the weaker the supplier's ability to transfer delivery experience.

[0033] For example, the list of items in this tender includes "Computing Server". In the knowledge graph of the items, the parent node of the "Computing Server" node is "Server", and the sibling nodes are "Storage Server" and "Network Server". Supplier S1's historical bidding records include "Server". In the knowledge graph, the semantic path between "Server" and "Computing Server" is: Computing Server → Parent Relationship → Server, passing through 1 edge, with a semantic path distance of 1. Supplier S2's historical bidding records also include "Storage Server". In the knowledge graph, the semantic path between "Storage Server" and "Computing Server" is: Storage Server → Sibling Relationship → Computing Server, passing through 1 edge, with a semantic path distance of 1. Supplier S2's historical bidding records also include "Core Switch". The parent node of "Core Switch" is "Network Switch", and the parent node of "Network Switch" is "Network Device". The relationship between "Network Device" and "Computing Server" requires multiple edges, with a semantic path distance of 4.

[0034] Based on semantic path distance, the single-point matching value of the supplier's bid for the project is calculated. The single-point matching value is a quantitative measure of the degree of matching between the supplier's delivery capability in a historically won bid and in this current tender. The single-point matching value is calculated as 1 ÷ (D + ε), where D is the semantic path distance and ε is a preset positive decimal to prevent the denominator from being zero; ε is set to 0.01. When the semantic path distance D is zero, the single-point matching value is set to the preset maximum matching value. This preset maximum matching value is reasonably set based on the single-point matching value when D=1, and is set to 2 ÷ (1 + ε). The rationale for this setting is that D being zero indicates that the supplier's historically won bid is completely consistent with the current tender, and the supplier has perfectly matching delivery experience. Therefore, the matching value should be higher than the highly semantically related case when D=1, but not excessively higher, to maintain a reasonable gradient in matching values ​​between different semantic distances. When D=1, the single-point matching value is 1÷(1+ε), and the preset maximum matching value is 2÷(1+ε), which is about twice that when D=1. This can reflect the advantage of fully matching delivery experience without over-amplifying the weight in subsequent calculations.

[0035] For example, for supplier S1's historically won bid item "server" and the current bid item "computing server", the semantic path distance D=1, and the single-point matching value of the item = 1÷(1+0.01)≈0.990. For supplier S2's historically won bid item "storage server" and the current bid item "computing server", the semantic path distance D=1, and the single-point matching value of the item = 1÷(1+0.01)≈0.990. For supplier S2's historically won bid item "core switch" and the current bid item "computing server", the semantic path distance D=4, and the single-point matching value of the item = 1÷(4+0.01)≈0.249.

[0036] Based on the single-point matching value of the subject matter, the supplier's delivery capability matching value for the subject matter is calculated. The delivery capability matching value is a comprehensive measure of the supplier's overall delivery capability for a specific subject matter in this tender. If the supplier has won multiple bids for subject matters that are semantically related to the subject matter in the past, multiple single-point matching values ​​will be obtained. The maximum value among all single-point matching values ​​is taken as the supplier's delivery capability matching value for that subject matter. The rationale for taking the maximum value is that the more relevant the supplier's delivery experience is for a particular subject matter in the past, the more it proves that the supplier has the core capability to undertake the subject matter in this tender. The sum of multiple related but irrelevant performance records should not exceed a single highly relevant performance record.

[0037] For example, the single-point matching values ​​for each historical winning bid item of "computing server" by supplier S1 are as follows: server 0.990, disk array 0.200, network switch 0.167, and computing device 0.500. Taking the maximum value of 0.990, supplier S1's delivery capability matching value for "computing server" is 0.990. The single-point matching values ​​for each historical winning bid item of "computing server" by supplier S2 are as follows: storage server 0.990, core switch 0.249, and solid-state storage device 0.167. Taking the maximum value of 0.990, supplier S2's delivery capability matching value for "computing server" is 0.990. The fact that the two suppliers have the same delivery capability matching value indicates that both have highly relevant delivery experience in the computing server field.

[0038] The following technical effects were achieved through this step: Instead of traditional text matching, the semantic path distance of the shortest path in the target object knowledge graph is used to quantify the semantic similarity between the target object in the supplier's past winning bids and the target object in this tender as a comparable number of path edges. The delivery capability matching value is taken as the maximum value among all single-point matching values. The target object in the supplier's past winning bids with the closest correlation is used as the evaluation benchmark for its core delivery capability, avoiding the interference of multiple scattered performance records with low correlation on the matching results.

[0039] S300: Based on the historical bidding records, calculate the capability coverage difference of each of the suppliers; Bidding projects typically involve multiple different types of items. A supplier's overall delivery capability depends not only on its performance on individual items but also on the breadth of its historical bidding experience covering the various types of items required in this project. Looking only at the matching value of a single item may overlook the supplier's coverage of the overall demand. Even if a supplier's historical bidding record includes all types of items, the distribution of delivery capability matching values ​​for each item may be uneven, requiring a comprehensive evaluation of its coverage through quantitative methods. This step measures the supplier's capability coverage from two dimensions: explicit coverage and solution delivery capability matching values. By calculating the deviation between these two dimensions, the capability coverage difference is obtained, providing a reliable basis for subsequent adjustments to the overall matching correlation.

[0040] Step S300 provided in this embodiment of the invention includes: Obtain the historical bidding records of each supplier, and count the total number of duplicate bid items of each supplier after deduplication using the bid item knowledge graph, as the capability coverage breadth of the supplier. The total number of object types in the object list is counted as the total number of required objects; The ratio of the capability coverage breadth to the total number of the required targets is calculated as the explicit coverage. Obtain the weighted sum of the delivery capability matching values ​​of each of the suppliers for each of the items in all the item lists, and use it as the solution delivery capability matching value of each of the suppliers; The deviation between the explicit coverage and the matching value of the solution delivery capability is calculated as the capability coverage difference.

[0041] The specific implementation method is as follows: First, obtain the historical bidding records of each supplier. The source of these historical bidding records is the same as in step S200, extracted from the bidding management system or the bidding announcement database. Map the names of the historical bidding items of each supplier to the item knowledge graph to obtain the corresponding item nodes for each supplier's historical bidding items. Deduplicate the set of historical bidding item nodes for each supplier, removing duplicate item nodes, and count the total number of item types after deduplication, which serves as the supplier's capability coverage breadth. Capability coverage breadth reflects the range of item categories covered by the supplier's historical delivery experience; the wider the range, the stronger the supplier's ability to handle different types of items.

[0042] The total number of item types in the list of items to be tendered is taken as the total number of required items. The total number of required items reflects the diversity of the requirements for the items in this tender project.

[0043] The ratio of the breadth of capability coverage to the number of target species is used as the explicit coverage. Explicit Coverage = Breadth of Capability Coverage ÷ Total Number of Target Species. The explicit coverage ranges from zero to one. The closer the explicit coverage is to one, the higher the direct matching ratio between the types of target species in the supplier's historical bids and the types of target species in the current demand. Explicit coverage only considers whether the target species have a direct matching relationship with the same name or the same node, and does not involve semantic level expansion, reflecting the supplier's coverage at the textual level.

[0044] Obtain the weighted sum of the delivery capability matching values ​​of each supplier for each item in the entire item list, and use this sum as the solution delivery capability matching value for each supplier. The delivery capability matching value is calculated in step S200 and reflects the supplier's semantic-level delivery capability for each item. Solution delivery capability matching value = Σ(w i ×Delivery Capacity Matching Value), where i ranges from 1 to the total number of required items, and wi is the weighting coefficient of the i-th item. The weighting coefficient of each item is determined based on its importance in this bidding project. The method for determining the weighting coefficient is as follows: obtain the budget amount or quantity percentage of each item in the bidding documents, normalize the percentage of each item to the range of zero to one, and use it as the weighting coefficient. If the bidding documents do not specify the differences in importance of each item, then the weighting coefficient of each item is 1 divided by the total number of required items, i.e., w. i =1 ÷ Total number of required items.

[0045] For example, the weight coefficients for the three target items—the computing server, storage device, and network switch—are all taken as 1 ÷ 3 ≈ 0.333. Assuming that step S200 calculates the matching values ​​of supplier S1's delivery capability for the three target items as 0.990, 0.500, and 0.333 respectively, the solution delivery capability matching value = 0.333 × 0.990 + 0.333 × 0.500 + 0.333 × 0.333 = 0.330 + 0.167 + 0.111 = 0.608. The matching values ​​of supplier S2's delivery capability for the three target items as 0.500, 0.990, and 0.800 respectively, the solution delivery capability matching value = 0.333 × 0.500 + 0.333 × 0.990 + 0.333 × 0.800 = 0.167 + 0.330 + 0.266 = 0.763.

[0046] Finally, the deviation between explicit coverage and solution delivery capability matching value is calculated as the capability coverage difference. The capability coverage difference measures the degree of difference between the supplier's textual coverage and semantic coverage. Capability coverage difference = |Explicit Coverage - Solution Delivery Capability Matching Value| ÷ [(Explicit Coverage + Solution Delivery Capability Matching Value) ÷ 2]. The closer the capability coverage difference is to zero, the more consistent the supplier's explicit coverage and solution delivery capability matching value are, and the more reliable the capability assessment results. A larger capability coverage difference indicates a significant difference between explicit coverage and solution delivery capability matching value, suggesting that the supplier's actual capabilities may be underestimated or overestimated, requiring correction in subsequent comprehensive matching correlation calculations.

[0047] For example, supplier S1 has an explicit coverage of 1.0, a solution delivery capability matching value of 0.608, and a capability coverage discrepancy of |1.0-0.608|÷[(1.0+0.608)÷2]=0.392÷0.804≈0.488. Supplier S2 has an explicit coverage of 0.667, a solution delivery capability matching value of 0.763, and a capability coverage discrepancy of |0.667-0.763|÷[(0.667+0.763)÷2]=0.096÷0.715≈0.134. Supplier S2's capability coverage discrepancy of 0.134 is significantly smaller than supplier S1's 0.488, indicating that supplier S2's explicit coverage is more consistent with its solution delivery capability matching value, and its semantic-level delivery capability assessment results are more reliable. Although supplier S1 has an explicit coverage of 1.0, its solution delivery capability matching value is low and the capability coverage varies greatly. Therefore, its overall delivery capability may be subject to high uncertainty.

[0048] The following technical effects were achieved through this step: By constructing a two-dimensional evaluation index of explicit coverage and solution delivery capability matching value, explicit coverage reflects the degree of direct matching between the types of goods won by the supplier in the past and the types of goods required, while solution delivery capability matching value reflects the supplier's coverage capability at the semantic level based on semantic path distance. Together, they constitute a complete evaluation of the supplier's capability coverage. The capability coverage difference is obtained by calculating the deviation between explicit coverage and solution delivery capability matching value. The closer the deviation is to zero, the more reliable the assessment of the supplier's actual delivery capability, providing a quantitative basis for subsequent correction of the comprehensive matching correlation.

[0049] S400: Calculate the historical active focus of each supplier based on the bidding frequency and bidding time decay characteristics of each target item in the historical bidding records of each supplier; This step uses each item in the list of items as the unit of analysis, counts the total number of times each supplier has won the bid for that item within a preset historical time range as the winning bid frequency, calculates the time elapsed between each winning bid and the current time and converts it into a time decay index, calculates the decay factor for each winning bid based on the time decay index, and obtains the historical activity focus by weighting and summing the decay factors for the same item. This provides a time-sensitive indicator that reflects the supplier's activity level in the current item field for subsequent comprehensive matching correlation calculation.

[0050] Step S400 provided in this embodiment of the invention includes: Obtain the historical bidding records of each supplier. For each item in the list of items, count the total number of times the supplier has won the bid for that item within a preset historical time range, and use this count as the supplier's bidding frequency for that item. Calculate the time elapsed between each supplier's winning bid and the current time, and divide by the base time decay unit to obtain the time decay index for each winning bid. Based on the time decay index, the decay factor of each winning bid of each supplier is calculated and obtained, and the decay factors of the same target are weighted and summed to obtain the historical active focus of the supplier on the target.

[0051] The specific implementation method is as follows: First, obtain the historical bidding records of each supplier. The source of the historical bidding records is the same as in step S200. Each historical bidding record includes the supplier identifier, the historical bidding project number, the name of the historical bidding item, and the date of the bidding for that project. For each item in the item list, determine the corresponding item node in the item knowledge graph, as well as all related nodes that have a superior, subordinate, or sibling relationship with that node. For each supplier, retrieve the supplier's historical bidding records, and filter out the bidding records whose item node corresponding to the historical bidding item name belongs to the above-mentioned set of related nodes, thus forming the supplier's set of related bidding records for that item.

[0052] The total number of times a supplier wins a bid for a given item within a preset historical time frame is counted as the supplier's winning frequency for that item. The preset historical time frame is a fixed period calculated backwards from the current moment, determined based on the effective continuity of a supplier's delivery capacity in the bidding market. The preset historical time frame is determined by obtaining the time interval between two consecutive winning bids for the same item from all suppliers' historical bidding records, and taking the 80th percentile of all time intervals as the preset historical time frame. The 80th percentile means that approximately 80% of the time intervals between two consecutive winning bids do not exceed this value, and this time frame covers the supplier's continuous active period under normal business conditions.

[0053] For example, 2000 samples were collected from the historical bidding records of all suppliers, showing the time interval between two consecutive bidding records of the same supplier in the same subject area. The 80th percentile is 5 years, or 1825 days, and the preset historical time range is 1825 days. Supplier S1 won bids for the "Computing Server" related subject 3 times in the last 1825 days, with a bidding frequency of 3. Supplier S2 won bids for the "Computing Server" related subject 2 times in the last 1825 days, with a bidding frequency of 2.

[0054] Then, calculate the time elapsed between each supplier's winning bid and the current time. For each winning bid record in the associated winning bid record set, obtain the winning bid date for that record and calculate the number of days between that winning bid date and the current time. This number is taken as the time elapsed between the winning bid and the current time. The longer the time elapsed between the winning bid and the current time, the further back in time the winning bid occurred. Divide the time elapsed between each winning bid and the current time by the baseline time decay unit to obtain the time decay index for each winning bid. The baseline time decay unit is a benchmark scale for normalizing the winning bid duration, taken as one-tenth of a preset historical time range. Baseline time decay unit = preset historical time range ÷ 10. Taking one-tenth is to ensure that the range of values ​​for the time decay index roughly corresponds to a distribution from 0 to 10, satisfying the numerical characteristics requirements of the decay factor formula.

[0055] For example, the preset historical time range is 1825 days, and the base time decay unit = 1825 ÷ 10 ≈ 183 days. Supplier S1 has 3 historical bidding records in the associated targets of the computing server. The time elapsed since the winning bid date is as follows: the first winning bid is 120 days ago, the second winning bid is 450 days ago, and the third winning bid is 1100 days ago. The time decay indices for each winning bid are: 120 ÷ 183 ≈ 0.656, 450 ÷ 183 ≈ 2.459, and 1100 ÷ 183 ≈ 6.011, respectively.

[0056] Next, based on the time decay index, the decay factor for each supplier's winning bid is calculated. The decay factor = 1 ÷ (1 + α), where α is the time decay index for that winning bid. The decay factor ranges from 0 to 1. The closer the decay factor is to 1, the more recent the winning bid, and the greater its contribution to current activity; the closer it is to 0, the more distant the winning bid, and the smaller its contribution to current activity. For example, the decay factors for supplier S1's three winning bids in the server-related targets are as follows: First decay factor = 1 ÷ (1 + 0.656) = 1 ÷ 1.656 ≈ 0.604, Second decay factor = 1 ÷ (1 + 2.459) = 1 ÷ 3.459 ≈ 0.289, Third decay factor = 1 ÷ (1 + 6.011) = 1 ÷ 7.011 ≈ 0.143.

[0057] Finally, the attenuation factors for the same target are weighted and summed to obtain the supplier's historical active focus on that target. The weights of the weighted sum are determined based on the project size corresponding to each winning bid, with project size measured by the winning bid amount or the total contract price. The weight of each winning bid = the project amount of that winning bid ÷ the sum of all winning bid amounts for that supplier in the associated winning bid record set for that target. If the bidding management system or the winning bid announcement database does not store winning bid amount information, then the weights of each winning bid are equal, i.e., weight = 1 ÷ winning bid frequency.

[0058] For example, supplier S1 won three bids in the computing server related project, with the amounts being RMB 820,000, RMB 650,000, and RMB 430,000 respectively, totaling RMB 1,900,000. The weights of each bid are: 820,000 ÷ 1,900,000 ≈ 0.432, 650,000 ÷ 1,900,000 ≈ 0.342, and 430,000 ÷ 1,900,000 ≈ 0.226. Historical active focus = 0.432 × 0.604 + 0.342 × 0.289 + 0.226 × 0.143 = 0.261 + 0.099 + 0.032 = 0.392. Supplier S2 won two bids in the computing server related project, with the amounts being RMB 950,000 (200 days ago) and RMB 550,000 (800 days ago). The time decay indices are 200 ÷ 183 ≈ 1.093 and 800 ÷ 183 ≈ 4.372 respectively. The attenuation factors are 1÷(1+1.093)=1÷2.093≈0.478 and 1÷(1+4.372)=1÷5.372≈0.186, respectively. The weights are 95÷150≈0.633 and 55÷150≈0.367, respectively. Historical active focus = 0.633×0.478+0.367×0.186=0.303+0.068=0.371. The historical active focus of supplier S1 is 0.392, and the historical active focus of supplier S2 is 0.371. Supplier S1 has a higher winning frequency and more recent winning records in the computing server field, and its historical active focus is slightly higher than that of supplier S2. This indicates that supplier S1 has a deeper investment and higher activity in this target field. For example, Table 1 is an example data table of supplier historical bidding records provided in the embodiments of the present invention, showing the historical bidding records of supplier S1 and supplier S2 and the corresponding bidding time and bidding amount data.

[0059] Table 1: Example Data Table of Supplier's Historical Bidding Records

[0060] The following technical effects were achieved through this step: By introducing a time decay mechanism, the time elapsed between each successful bid and the current moment is converted into a time decay index. The decay factor reflects the differentiated contribution of each successful bid to the current activity level, with recent successful bids contributing significantly more than those from longer periods. The historical activity focus is obtained by weighting and summing the frequency of successful bids and the decay factor, comprehensively reflecting the supplier's continuous investment depth and current activity level in the target asset sector.

[0061] S500: Based on the delivery capability matching value, the capability coverage difference, and the historical active focus, calculate the comprehensive matching correlation degree of each of the suppliers, and obtain the semantic correlation analysis results of the target object between the project to be tendered and each of the suppliers.

[0062] This step involves weighted summation of the historical active focus of each supplier on all targets to obtain the historical active focus of the solution. This is then weighted and fused with the solution delivery capability matching value to obtain the initial comprehensive matching degree. After correction with the capability coverage difference, the comprehensive matching correlation degree is obtained. The semantic correlation analysis results are then output in descending order.

[0063] Step S500 provided in this embodiment of the invention includes: Based on the delivery capability matching value, the capability coverage difference, and the historical active focus, the comprehensive matching correlation degree of each supplier is calculated, and the semantic association analysis results of the target asset between the project to be tendered and each supplier are obtained, including: Obtain the weighted sum of the historical active focus of each supplier on each target in the target list, as the supplier's scheme historical active focus; The supplier's solution delivery capability matching value and the historical active focus of the solution are weighted and fused to obtain the initial comprehensive matching degree of the supplier; The initial comprehensive matching degree is corrected using the capability coverage difference to obtain the comprehensive matching correlation degree of the supplier; The suppliers are sorted in descending order of their comprehensive matching relevance, and the semantic association analysis results of the target items between the project to be tendered and each supplier are generated. The analysis results include the comprehensive matching relevance ranking, delivery capability matching value, historical active focus, and capability coverage difference of each supplier.

[0064] The specific implementation method is as follows: First, obtain the weighted sum of the historical active focus of each supplier on each item in the target list, which will be used as the supplier's historical active focus. Target Historical Active Focus = Σ(w i ×Historical active focus), where w iThe weighting coefficient for the i-th target is the same as the weighting coefficient used in calculating the solution delivery capability matching value in step S300. The solution historical activity focus integrates the supplier's current activity level across all demand targets.

[0065] The initial overall matching degree of the supplier is obtained by weighted fusion of the solution delivery capability matching value and the solution's historical activity and focus. The solution delivery capability matching value has been calculated in step S300. The weighted fusion calculation method is: initial overall matching degree = w1 × solution delivery capability matching value + w2 × solution's historical activity and focus, where w1 is the weight coefficient of the solution delivery capability matching value and w2 is the weight coefficient of the solution's historical activity and focus.

[0066] The methods for determining w1 and w2 are as follows: A standard sample set is obtained by manually confirming the winning suppliers from historical bidding projects. For each standard sample, the solution delivery capability matching value calculated in step S300 and the solution historical activity focus calculated in this step are obtained respectively. Logistic regression is used to fit the binary classification result of whether the supplier won the bid. The ratio of the absolute value of the regression coefficient of the solution delivery capability matching value in the logistic regression model to the sum of the absolute values ​​of the two regression coefficients is taken as w1, and the ratio of the absolute value of the regression coefficient of the solution historical activity focus to the sum of the absolute values ​​of the two regression coefficients is taken as w2.

[0067] For example, supplier S1's delivery capability matching value is 0.608, its historical active focus is 0.239, and its initial overall matching degree is 0.668 × 0.608 + 0.332 × 0.239 = 0.406 + 0.079 = 0.485. Supplier S2's delivery capability matching value is 0.763, its historical active focus is 0.380, and its initial overall matching degree is 0.668 × 0.763 + 0.332 × 0.380 = 0.510 + 0.126 = 0.636.

[0068] The initial comprehensive matching degree is corrected using the capability coverage difference degree to obtain the supplier's comprehensive matching relevance. The capability coverage difference degree has already been calculated in step S300. The correction method is as follows: calculate 1 and subtract the capability coverage difference degree; the difference is used as the coverage reliability factor. Coverage reliability factor = 1 - capability coverage difference degree. The initial comprehensive matching degree is multiplied by the coverage reliability factor, and the product is used as the supplier's comprehensive matching relevance. When the calculation result is greater than 1, it is taken as 1.

[0069] For example, supplier S1's capability coverage difference is 0.488, coverage confidence factor = 1 - 0.488 = 0.512, and overall matching relevance = 0.485 × 0.512 ≈ 0.248. Supplier S2's capability coverage difference is 0.134, coverage confidence factor = 1 - 0.134 = 0.866, and overall matching relevance = 0.636 × 0.866 ≈ 0.551.

[0070] Suppliers are ranked according to their overall matching relevance from highest to lowest, generating semantic association analysis results between the projects to be tendered and each supplier's target assets. The analysis results include the overall matching relevance ranking of each supplier, the matching value of their solution delivery capabilities, the historical activity level of their solutions, and the difference in their capability coverage.

[0071] The following technical effects were achieved through this step: The solution delivery capability matching value and the solution's historical activity and focus are weighted and integrated. The weighting coefficients are determined by fitting historical bidding data through logistic regression, so that the overall matching degree objectively reflects the contribution ratio of the two indicators to the bidding result. The initial overall matching degree is corrected by the coverage credibility factor. The supplier with greater capability coverage difference has a larger correction degree, so that the final overall matching correlation degree reflects both the supplier's delivery capability and activity level, while also taking into account the reliability of the supplier's capability assessment.

[0072] Preferably, embodiments of the present invention also provide an adaptive update function for the target knowledge graph, including: Obtain the delivery capability matching value of each supplier for each item in the target list during the current analysis process; The delivery capability matching value of each supplier for each target is compared with a preset map update threshold. The supplier whose delivery capability matching value exceeds the map update threshold and the corresponding target are used as semantic matching update samples. For each semantic matching update sample, obtain the name of the winning bidder in the supplier's historical bidding records that has the shortest semantic path distance to the corresponding bidder on the bidder knowledge graph; When at least two suppliers simultaneously have a delivery capability matching value between the winning bid node and the corresponding node in the list of bid items that is greater than or equal to a preset graph update threshold, a new co-position relationship edge is added between the winning bid node and the corresponding node in the list of bid items. If a co-position relationship edge already exists between the two, the weight value of the co-position relationship edge is increased by a preset increment. The specific implementation method is as follows: After outputting the semantic association analysis results in step S500, the knowledge graph of the target items is adaptively updated using the delivery capability matching values ​​of each supplier for each target item calculated during this analysis. The initial version of the knowledge graph is constructed based on the target item names in historical bidding announcements, and the semantic hierarchical associations in the graph depend on entity recognition and semantic hierarchical classification of the target item names. As bidding activities continue, new target item names constantly emerge, and the actual bidding behavior of suppliers continuously reveals the practical associations between different target items. By analyzing the delivery capability matching values ​​of suppliers for the target items, implicit semantic associations that were not identified during the knowledge graph construction phase can be mined from actual business data, allowing the knowledge graph to gradually become richer and more complete with increased usage frequency.

[0073] The delivery capability matching value of each supplier for each target item is compared with a preset graph update threshold. The delivery capability matching value is calculated in step S200 and reflects the supplier's semantic-level delivery capability for a certain target item. The preset graph update threshold is used to filter supplier-target item pairs with high delivery correlation as trigger conditions for knowledge graph updates. The preset graph update threshold is determined as follows: obtain the delivery capability matching values ​​of all suppliers for all targets items in the current analysis process, take the median and maximum value of all delivery capability matching values, and the preset graph update threshold = median + (maximum value - median) × preset scaling factor. The preset scaling factor is set to 0.5, so that the threshold value is located at the midpoint between the median and the maximum value, ensuring that only supplier-target item pairs with high delivery capability matching values ​​can trigger graph updates.

[0074] For example, the median of all delivery capability matching values ​​in the current analysis is 0.35, and the maximum is 0.99. The preset map update threshold = 0.35 + (0.99 - 0.35) × 0.5 = 0.35 + 0.32 = 0.67. Supplier S1's delivery capability matching value for the computing server is 0.99, exceeding the threshold of 0.67. Therefore, supplier S1 and its corresponding target computing server are used as semantic matching update samples. Supplier S2's delivery capability matching value for the storage device is 0.99, exceeding the threshold of 0.67, and is also used as a semantic matching update sample.

[0075] For each semantic matching update sample, the name of the winning bid item that forms a semantic matching update sample with the target item in the supplier's historical bidding records is obtained. This winning bid item name is the name of the historical winning bid item that enabled the supplier to obtain the highest single-point matching value for the target item when calculating the semantic path distance in step S200. For example, for the semantic matching update sample "Supplier S1 - Computing Server", the name of the historical winning bid item corresponding to the highest single-point matching value obtained by Supplier S1 for the Computing Server in step S200 is "Server", that is, the name of the winning bid item corresponding to this semantic matching update sample is "Server". For the semantic matching update sample "Supplier S2 - Storage Device", the name of the historical winning bid item corresponding to the highest single-point matching value obtained by Supplier S2 for the Storage Device in step S200 is "Disk Array", that is, the name of the winning bid item corresponding to this semantic matching update sample is "Disk Array".

[0076] A knowledge graph update operation is performed when at least two suppliers simultaneously have delivery capability matching values ​​greater than or equal to a preset knowledge graph update threshold for these two items. The two items refer to two items in the current semantic matching update sample: one is the item node in the item list, and the other is the corresponding winning item node in the supplier's historical bidding record. The fact that at least two suppliers simultaneously have delivery capability matching values ​​exceeding the threshold for both items indicates that these two items are highly interchangeable in actual bidding activities.

[0077] For example, in this round of analysis, in addition to supplier S1 forming a matching relationship for "computing server-server" that exceeds the threshold, there is also supplier S3 whose delivery capability matching value for "computing server" is 0.85, which also exceeds the threshold of 0.67, and the historical winning bid item corresponding to its highest single-point matching value is also "server". At this time, the knowledge graph is updated.

[0078] In the target knowledge graph, a new corresponding edge is added between the "Server" node and the "Computation Server" node. This new corresponding edge indicates that the two targets are semantically related as parallel subclasses; that is, "Server" and "Computation Server" can be considered targets at the same semantic level in actual bidding processes. When a corresponding edge already exists between them, its weight is increased by a preset increment. The initial weight of the corresponding edge is 1, increasing by a preset increment of 1 each time. A larger weight indicates a more stable corresponding relationship between the two targets revealed by actual business data, and this weight information can be used for distance weighting adjustments during subsequent semantic path distance calculations. For example, upon inspection, there is currently no corresponding edge between the "Server" node and the "Computation Server" node (only a superior-ranking edge exists), therefore a new corresponding edge is added between them, with an initial weight of 1.

[0079] The updated target knowledge graph is stored for subsequent rounds of target semantic matching analysis. The updated knowledge graph includes new semantic associations revealed by the supplier's actual bidding behavior. When performing the target semantic matching analysis next time, step S100 will read the updated and richer knowledge graph, enabling the semantic path distance calculation to utilize the newly added semantic association information, thereby improving the accuracy and coverage of the supplier's delivery capability matching value calculation.

[0080] Example 2, as Figure 3 As shown, based on the same inventive concept provided in Embodiment 1, this embodiment of the invention also provides a knowledge graph-based semantic matching analysis system for bidding and tendering targets, the system comprising: The graph acquisition module 11 is used to acquire a list of target items for the project to be tendered and to acquire a target item knowledge graph, wherein the target item knowledge graph contains semantic hierarchical associations between target item nodes. The capability matching calculation module 12 is used to obtain the historical bidding records of each supplier, and analyze and obtain the delivery capability matching value of each supplier for each target based on the semantic path distance between each target node in the target knowledge graph. The coverage difference calculation module 13 is used to calculate the capability coverage difference of each of the suppliers based on the historical bidding records; The active focus calculation module 14 is used to calculate the historical active focus of each supplier based on the bidding frequency and bidding time decay characteristics of each target in the historical bidding records of each supplier. The comprehensive correlation calculation module 15 is used to calculate the comprehensive matching correlation of each supplier based on the delivery capability matching value, the capability coverage difference, and the historical active focus, and to obtain the semantic correlation analysis results of the target items between the project to be tendered and each supplier. A list of target item names is extracted from the tender documents of the project to be tendered as the target item list, wherein the target item list contains at least one target item name; From the target knowledge graph, obtain the target node corresponding to each target name in the target list, and the semantic hierarchical association between each target node.

[0081] The semantic hierarchical relationships include superordinate relationships, subordinate relationships, and peer relationships. The superordinate relationship indicates that one target node is a general concept of another target node. The subordinate relationship indicates that one target node is a specific subclass of another target node. The peer relationship indicates that two target nodes belong to parallel subclasses under the same superordinate concept.

[0082] Collect the names of the subject matter from historical bidding announcements to form a set of subject matter names; Entity recognition and semantic hierarchical classification are performed on each object name in the object name set, and the superior, inferior and co-existing relationships between each object are extracted; Using the names of each target as nodes and the superior, subordinate, and peer relationships as edges, a directed acyclic graph structure is constructed as the target knowledge graph. Each target node stores the standardized name of the target, and each edge stores the type of semantic hierarchical association.

[0083] In one embodiment, the map acquisition module 11 is further configured to extract a list of subject matter names from the tender documents of the project to be tendered, as the subject matter list, wherein the subject matter list contains at least one subject matter name; From the target knowledge graph, obtain the target node corresponding to each target name in the target list, and the semantic hierarchical association between each target node.

[0084] The semantic hierarchical relationships include superordinate relationships, subordinate relationships, and peer relationships. The superordinate relationship indicates that one target node is a general concept of another target node. The subordinate relationship indicates that one target node is a specific subclass of another target node. The peer relationship indicates that two target nodes belong to parallel subclasses under the same superordinate concept.

[0085] The construction of the target object knowledge graph includes: Collect the names of the subject matter from historical bidding announcements to form a set of subject matter names; Entity recognition and semantic hierarchical classification are performed on each object name in the object name set, and the superior, inferior and co-existing relationships between each object are extracted; Using the names of each target as nodes and the superior, subordinate, and peer relationships as edges, a directed acyclic graph structure is constructed as the target knowledge graph. Each target node stores the standardized name of the target, and each edge stores the type of semantic hierarchical association.

[0086] In one embodiment, the capability matching calculation module 12 is further configured to obtain the historical bidding records of each supplier, wherein the historical bidding records include the project number of the supplier's historical bidding projects and the name of the historical bidding target; For each item in the list of items, in the item knowledge graph, starting from the item node and ending at each winning item node in the historical bidding records of each supplier, the number of edges traversed by the shortest semantic path between the starting point and each ending point is calculated as the semantic path distance. Based on the semantic path distance, the single-point matching value of the supplier for the subject matter of the project to be tendered is calculated and obtained, wherein when the semantic path distance is zero, the single-point matching value is taken as the preset maximum matching value; Based on the single-point matching value of the target product, the supplier's delivery capability matching value for the target product is calculated and obtained.

[0087] In one embodiment, the coverage difference calculation module 13 is further used to obtain the historical bidding records of each supplier and count the total number of duplicate target items of each supplier's historical bidding items after deduplication of the target item knowledge graph, as the capability coverage breadth of the supplier. The total number of object types in the object list is counted as the total number of required objects; The ratio of the capability coverage breadth to the total number of the required targets is calculated as the explicit coverage. Obtain the weighted sum of the delivery capability matching values ​​of each of the suppliers for each of the items in all the item lists, and use it as the solution delivery capability matching value of each of the suppliers; The deviation between the explicit coverage and the matching value of the solution delivery capability is calculated as the capability coverage difference.

[0088] In one embodiment, the active focus calculation module 14 is further configured to obtain the historical bidding records of each of the suppliers, and for each item in the list of items, to count the total number of times the supplier has won the bid for that item within a preset historical time range, as the bidding frequency of that supplier for that item. Calculate the time elapsed between each supplier's winning bid and the current time, and divide by the base time decay unit to obtain the time decay index for each winning bid. Based on the time decay index, the decay factor of each winning bid of each supplier is calculated and obtained, and the decay factors of the same target are weighted and summed to obtain the historical active focus of the supplier on the target.

[0089] In one embodiment, the comprehensive correlation calculation module 15 is further configured to obtain the weighted sum of the historical active focus of each supplier on each target in the target list, as the supplier's scheme historical active focus; The supplier's solution delivery capability matching value and the historical active focus of the solution are weighted and fused to obtain the initial comprehensive matching degree of the supplier; The initial comprehensive matching degree is corrected using the capability coverage difference to obtain the comprehensive matching correlation degree of the supplier; The suppliers are sorted in descending order of their comprehensive matching relevance, and the semantic association analysis results of the target items between the project to be tendered and each supplier are generated. The analysis results include the comprehensive matching relevance ranking, delivery capability matching value, historical active focus, and capability coverage difference of each supplier.

[0090] Obtain the delivery capability matching value of each supplier for each item in the target list during the current analysis process; The delivery capability matching value of each supplier for each target is compared with a preset map update threshold. The supplier whose delivery capability matching value exceeds the map update threshold and the corresponding target are used as semantic matching update samples. For each semantic matching update sample, obtain the name of the winning bidder in the supplier's historical bidding records that has the shortest semantic path distance to the corresponding bidder on the bidder knowledge graph; When at least two suppliers simultaneously have a delivery capability matching value between the winning bid node and the corresponding node in the list of bid items that is greater than or equal to a preset graph update threshold, a new co-position relationship edge is added between the winning bid node and the corresponding node in the list of bid items. If a co-position relationship edge already exists between the two, the weight value of the co-position relationship edge is increased by a preset increment. The updated object knowledge graph is stored for subsequent rounds of object semantic matching analysis.

Claims

1. A semantic matching analysis method for bidding documents based on knowledge graphs, characterized in that, include: Obtain a list of target items for the project to be tendered, and obtain a target item knowledge graph, wherein the target item knowledge graph contains semantic hierarchical associations between target item nodes; Obtain the historical bidding records of each supplier, and based on the semantic path distance between each target node in the target knowledge graph, analyze and obtain the delivery capability matching value of each supplier for each target, wherein the delivery capability matching value is used to characterize the supplier's delivery capability for a certain target. Based on the historical bidding records, the capability coverage difference of each supplier is calculated, wherein the capability coverage difference is used to characterize the degree of deviation between the capability coverage breadth of each supplier and the delivery capability matching value; Based on the bidding frequency and bidding time decay characteristics of each target item in the historical bidding records of each supplier, the historical active focus of each supplier is calculated. The bidding time decay characteristics are used to characterize the decay characteristics of the contribution of the historical bidding records to the active state over time, and the historical active focus is used to characterize the depth of continuous investment of each supplier in the target item and the current active state. Based on the delivery capability matching value, the capability coverage difference, and the historical active focus, the comprehensive matching correlation degree of each supplier is calculated, and the semantic correlation analysis results of the target items between the project to be tendered and each supplier are obtained.

2. The semantic matching analysis method for bidding objects based on knowledge graphs according to claim 1, characterized in that, The process involves obtaining a list of target items for the projects to be tendered and acquiring a target item knowledge graph. This target item knowledge graph contains semantic hierarchical relationships between target item nodes, including: Extract a list of subject matter names from the tender documents of the project to be tendered, and use it as the subject matter list, wherein the subject matter list contains at least one subject matter name; From the target knowledge graph, obtain the target node corresponding to each target name in the target list, and the semantic hierarchical association between each target node.

3. The semantic matching analysis method for bidding documents based on knowledge graphs according to claim 2, characterized in that, The semantic hierarchical relationships include superordinate relationships, subordinate relationships, and peer relationships. The superordinate relationship indicates that one target node is a general concept of another target node. The subordinate relationship indicates that one target node is a specific subclass of another target node. The peer relationship indicates that two target nodes belong to parallel subclasses under the same superordinate concept.

4. The semantic matching analysis method for bidding objects based on knowledge graphs according to claim 3, characterized in that, The construction of the target object knowledge graph includes: Collect the names of the subject matter from historical bidding announcements to form a set of subject matter names; Entity recognition and semantic hierarchical classification are performed on each object name in the object name set, and the superior, inferior and co-existing relationships between each object are extracted; Using the names of each target as nodes and the superior, subordinate, and peer relationships as edges, a directed acyclic graph structure is constructed as the target knowledge graph. Each target node stores the standardized name of the target, and each edge stores the type of semantic hierarchical association.

5. The semantic matching analysis method for bidding documents based on knowledge graphs according to claim 1, characterized in that, The step of obtaining historical bidding records of each supplier and analyzing the semantic path distance between nodes of each target item in the target item knowledge graph to obtain the delivery capability matching value of each supplier for each target item includes: Obtain the historical bidding records of each supplier, wherein the historical bidding records include the project number and the name of the subject matter of the historical bidding projects of the supplier; For each item in the list of items, in the item knowledge graph, starting from the item node and ending at each winning item node in the historical bidding records of each supplier, the number of edges traversed by the shortest semantic path between the starting point and each ending point is calculated as the semantic path distance. Based on the semantic path distance, the single-point matching value of the supplier for the subject matter of the project to be tendered is calculated and obtained, wherein when the semantic path distance is zero, the single-point matching value is taken as the preset maximum matching value; Based on the single-point matching value of the target product, the supplier's delivery capability matching value for the target product is calculated and obtained.

6. The semantic matching analysis method for bidding documents based on knowledge graphs according to claim 1, characterized in that, The calculation of the capability coverage difference of each supplier based on the historical bidding records includes: Obtain the historical bidding records of each supplier, and count the total number of duplicate bid items of each supplier after deduplication using the bid item knowledge graph, as the capability coverage breadth of the supplier. The total number of object types in the object list is counted as the total number of required objects; The ratio of the capability coverage breadth to the total number of the required targets is calculated as the explicit coverage. Obtain the weighted sum of the delivery capability matching values ​​of each of the suppliers for each of the items in all the item lists, and use it as the solution delivery capability matching value of each of the suppliers; The deviation between the explicit coverage and the matching value of the solution delivery capability is calculated as the capability coverage difference.

7. The semantic matching analysis method for bidding objects based on knowledge graphs according to claim 1, characterized in that, The calculation of the historical active focus of each supplier, based on the frequency and time decay characteristics of each bid in the historical bid-winning records of each bidder, includes: Obtain the historical bidding records of each supplier. For each item in the list of items, count the total number of times the supplier has won the bid for that item within a preset historical time range, and use this count as the supplier's bidding frequency for that item. Calculate the time elapsed between each supplier's winning bid and the current time, and divide by the base time decay unit to obtain the time decay index for each winning bid. Based on the time decay index, the decay factor of each winning bid of each supplier is calculated and obtained, and the decay factors of the same target are weighted and summed to obtain the historical active focus of the supplier on the target.

8. The semantic matching analysis method for bidding objects based on knowledge graphs according to claim 1, characterized in that, Based on the delivery capability matching value, the capability coverage difference, and the historical active focus, the comprehensive matching correlation degree of each supplier is calculated, and the semantic association analysis results of the target asset between the project to be tendered and each supplier are obtained, including: Obtain the weighted sum of the historical active focus of each supplier on each target in the target list, as the supplier's scheme historical active focus; The supplier's solution delivery capability matching value and the historical active focus of the solution are weighted and fused to obtain the initial comprehensive matching degree of the supplier; The initial comprehensive matching degree is corrected using the capability coverage difference degree to obtain the comprehensive matching correlation degree of the supplier. Specifically, 1 is subtracted from the capability coverage difference degree, and the difference is used as the coverage confidence factor. The initial comprehensive matching degree is multiplied by the coverage confidence factor, and the product is used as the comprehensive matching correlation degree of the supplier. The suppliers are sorted in descending order of their comprehensive matching relevance, and the semantic association analysis results of the target items between the project to be tendered and each supplier are generated. The analysis results include the comprehensive matching relevance ranking, delivery capability matching value, historical active focus, and capability coverage difference of each supplier.

9. The semantic matching analysis method for bidding objects based on knowledge graphs according to claim 1, characterized in that, Also includes: Obtain the delivery capability matching value of each supplier for each item in the target list during the current analysis process; The delivery capability matching value of each supplier for each target is compared with a preset map update threshold. The supplier whose delivery capability matching value exceeds the map update threshold and the corresponding target are used as semantic matching update samples. For each semantic matching update sample, obtain the name of the winning bidder in the supplier's historical bidding records that has the shortest semantic path distance to the corresponding bidder on the bidder knowledge graph; When at least two suppliers simultaneously have a delivery capability matching value between the winning bid node and the corresponding node in the list of bid items that is greater than or equal to a preset graph update threshold, a new co-position relationship edge is added between the winning bid node and the corresponding node in the list of bid items. If a co-position relationship edge already exists between the two, the weight value of the co-position relationship edge is increased by a preset increment. The updated object knowledge graph is stored for subsequent rounds of object semantic matching analysis.

10. A semantic matching and analysis system for bidding and tendering targets based on knowledge graphs, characterized in that, The system is used to implement the knowledge graph-based semantic matching analysis method for bidding documents according to any one of claims 1 to 9, the system comprising: The graph acquisition module is used to acquire a list of target items for the project to be tendered and to acquire a target item knowledge graph, wherein the target item knowledge graph contains the semantic hierarchical association relationship between target item nodes. The capability matching calculation module is used to obtain the historical bidding records of each supplier, and analyze and obtain the delivery capability matching value of each supplier for each target based on the semantic path distance between each target node in the target knowledge graph. The coverage difference calculation module is used to calculate the capability coverage difference of each supplier based on the historical bidding records. The active focus calculation module is used to calculate the historical active focus of each supplier based on the bidding frequency and bidding time decay characteristics of each target item in the historical bidding records of each supplier. The comprehensive correlation calculation module is used to calculate the comprehensive matching correlation of each supplier based on the delivery capability matching value, the capability coverage difference, and the historical active focus, and to obtain the semantic correlation analysis results of the target objects between the project to be tendered and each supplier.

Images