CN-121997920-A - Bid file personalized compiling system based on large model

CN121997920ACN 121997920 ACN121997920 ACN 121997920ACN-121997920-A

Abstract

The invention relates to the technical field of bidding document compiling, and discloses a large-model-based bidding document personalized compiling system which comprises an analysis module, a determination module, a construction module and a verification module, wherein the analysis module is used for determining a plurality of demand characteristic indexes of bidding documents, calculating demand priorities of the demand characteristic indexes, constructing a demand knowledge graph according to the demand priorities, the determination module is used for constructing a real-time response knowledge graph according to an enterprise database, determining a real-time data set based on the demand knowledge graph and the real-time response knowledge graph, the construction module is used for constructing a large language model, inputting the real-time data set into the large model to obtain initial bidding documents, the verification module is used for carrying out quality detection and competition simulation on the initial bidding documents, judging whether to generate an optimization instruction according to detection results and simulation results, improving matching efficiency and matching accuracy of enterprise resources and bidding demands, timely finding and optimizing defects in the initial bidding documents, and increasing bidding probability.

Inventors

LEI WENJUN
WU ZHIYING
WANG DONGLIANG

Assignees

华能招采数字科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251201

Claims (10)

1. A large model-based bid document personalization system comprising: the analysis module is used for determining a plurality of demand characteristic indexes of the bidding documents, calculating demand priorities of the demand characteristic indexes and constructing a demand knowledge graph according to the demand priorities; The determining module is used for constructing a real-time response knowledge graph according to the enterprise database and determining a real-time data set based on the demand knowledge graph and the real-time response knowledge graph; the construction module is used for constructing a large language model, inputting the real-time data set into the large model and obtaining an initial bidding document; And the verification module is used for carrying out quality detection and competition simulation on the initial bidding file, and judging whether to generate an optimization instruction according to the detection result and the simulation result.
2. The large model based bid document personalization system of claim 1, wherein calculating demand priorities for demand characteristic indicators comprises: analyzing the bid-inviting file and extracting key field information in the bid-inviting file; identifying and structuring key field information in the bidding document based on natural language processing technology to obtain a plurality of bidding requirement information; Generating a plurality of requirement characteristic indexes according to all bidding requirement information; Generating a plurality of historical demand characteristic indexes of the historical bidding documents, and carrying out similarity analysis on the historical demand characteristic indexes and the demand characteristic indexes of the current bidding documents to obtain similarity; extracting historical bidding documents with similarity larger than a preset similarity threshold, and collecting historical bidding data and historical bidding results of each extracted historical bidding document; Dividing the extracted historical bidding documents according to the historical bidding results, and constructing a first bidding data set and a second bidding data set according to the dividing results and the corresponding historical bidding data; the first bidding data set comprises a plurality of first bidding data subsets, the second bidding data set comprises a plurality of second bidding data subsets, and each bidding data subset is mapped with a corresponding weight coefficient; And comparing and analyzing the first bidding data set and the second bidding data set, and determining the demand priority of each bidding demand information according to the analysis result.
3. The large model based bid document personalization system of claim 2, wherein determining a demand priority for each bid requirement information based on the analysis results comprises: performing association analysis on historical bid data in each first bid data subset and each second bid data subset and the demand characteristic index to obtain association coefficients; if the association coefficient of the historical bidding data and the demand characteristic index in the first bidding data subset is larger than a preset association coefficient threshold value, setting the corresponding historical bidding data as a standard data set corresponding to the demand characteristic index; If the association coefficient of the historical bidding data and the demand characteristic index in the second bidding data subset is larger than a preset association coefficient threshold value, setting the corresponding historical bidding data as an abnormal data set corresponding to the demand characteristic index; Sequentially generating a plurality of standard data sets and a plurality of abnormal data sets of each demand characteristic index; Carrying out commonality analysis on all standard data sets of the same demand characteristic index to obtain a plurality of commonality characteristics, and setting an initial weight coefficient of each commonality characteristic by combining the occurrence frequency; Carrying out difference analysis on each abnormal data set of the same demand characteristic index and the corresponding common characteristic to obtain a plurality of sub-difference coefficients of the same common characteristic, and carrying out weight processing by combining the weight coefficient corresponding to each abnormal data set to obtain the difference coefficient of the same common characteristic; Generating a comprehensive difference coefficient of the corresponding demand characteristic index according to the difference coefficient of each common characteristic of the same demand characteristic index and the corresponding initial weight coefficient; and sequencing all the demand characteristic indexes according to the comprehensive difference coefficient, and setting the demand priority of the demand characteristic indexes according to the sequencing result.
4. The large model based bid document personalization system of claim 3, wherein constructing a demand knowledge graph according to demand priority comprises: Determining the position relation of each demand characteristic index in a demand knowledge graph based on the demand priority, and generating the demand knowledge graph comprising nodes and edges by combining a preset graph construction rule; Each node represents a demand characteristic index, the edges represent association characteristics and association weights among different demand characteristic indexes, and bidding demand information corresponding to the demand characteristic indexes and a plurality of corresponding standard data sets are mapped at each node.
5. The large model based bid document personalization system of claim 4, wherein constructing a real-time response knowledge graph from an enterprise database comprises: searching the enterprise database based on each node in the demand knowledge graph to obtain real-time response information corresponding to the bidding demand information at each node, real-time data sets corresponding to the standard data sets and real-time associated features corresponding to the associated features of each side; Converting real-time response information at each node in the demand knowledge graph into a response node, and converting real-time association features into edges connected with the corresponding response node; And generating a real-time response knowledge graph according to the plurality of response nodes and the corresponding edges, wherein each response node is mapped with corresponding real-time response information and a real-time data set.
6. The large model based bid document personalization system of claim 5, wherein determining a real-time dataset based on a demand knowledge-graph and a real-time response knowledge-graph comprises: Generating a plurality of node corresponding relations and edge corresponding relations according to the corresponding relations between the demand knowledge graph and the real-time response knowledge graph; carrying out information matching analysis on the nodes in the corresponding relation of each node to obtain information matching degree; Carrying out logic matching analysis on the edges in the corresponding relation of each edge to obtain logic matching degree; Presetting an information matching degree threshold and a logic matching degree threshold; Screening out node corresponding relations with information matching degree larger than an information matching degree threshold value and edge corresponding relations with logic matching degree threshold value larger than a logic matching degree threshold value; matching corresponding labels with real-time response nodes and edges in the real-time response knowledge graph according to the screened node corresponding relation and the edge corresponding relation, wherein the labels comprise trusted labels and unknown labels; generating a trusted coefficient of each real-time response node according to the label result; The method comprises the steps of adjusting a real-time response node with a trusted coefficient smaller than a preset trusted coefficient threshold until the trusted coefficient is larger than the preset trusted coefficient threshold, taking real-time response information at the real-time response node and a corresponding real-time data set as a main data set, and taking real-time association characteristics corresponding to edges connected with the real-time response node as an auxiliary data set; The primary data set and the secondary data set of all the real-time response nodes are formed into a real-time data set.
7. The large model based bid document personalization system of claim 6, wherein generating the confidence coefficient for each real-time response node based on the label results comprises: Determining a first class edge and a second class edge according to the label result of the edge connected with each real-time response node; The calculation formula of the trusted coefficient is as follows: ; Wherein K is a trusted coefficient, c1 is a first weight coefficient, c2 is a second weight coefficient, m1 is the number of first class edges, m2 is the number of second class edges, d1 is a first conversion coefficient, d2 is a second conversion coefficient, d3 is a third conversion coefficient, To select the coefficient when In the time-course of which the first and second contact surfaces, When =1 In the time-course of which the first and second contact surfaces, =0, H is the information matching degree of the real-time response node, h0 is the information matching degree threshold, ls is the logic matching degree of the s first class edge, l0 is the logic matching degree threshold, lv is the logic matching degree of the v second class edge, qs is the weight coefficient of the s first class edge, and qv is the weight coefficient of the v second class edge.
8. The large model based bid document personalization system of claim 7, wherein constructing a large language model comprises: The method comprises the steps of obtaining bid-inviting demand information, a historical main data set and a historical auxiliary data set at each node in a demand knowledge graph, taking the bid-inviting demand information, the historical main data set and the historical auxiliary data set at each node as training input data, and taking a plurality of standard data sets at each node as training output data; Training the initial large language model according to training input data and training output data to obtain a language large model; And inputting the real-time data set into the language big model, generating bidding contents of each demand characteristic index, and assembling an initial bidding file.
9. The large model based bid document personalization system of claim 8, wherein performing quality testing and competition simulation on the initial bid document comprises: constructing a quality detection model based on a preset bidding document quality evaluation index system; detecting an initial bidding document based on a quality detection model to obtain a detection result, wherein the detection result comprises content integrity, content accuracy, content standardization, content pertinence and content innovativeness; Generating a detection evaluation value according to the detection result; Constructing a competition simulation model; Generating a plurality of predicted bid files based on the competition simulation model and the plurality of bid requirement information; Comparing and analyzing the predicted bid files with the initial bid files to obtain a simulation result, wherein the simulation result comprises a plurality of advantage points of the initial bid files for the predicted bid files; and generating a simulation evaluation value according to the simulation result.
10. The large model based bid document personalization system of claim 9, wherein determining whether to generate the optimization instruction based on the detection result and the simulation result comprises: presetting a detection evaluation value threshold and an analog evaluation value threshold; When the detection evaluation value is larger than the detection evaluation value threshold value and the simulation evaluation value is larger than the simulation evaluation value threshold value, generating no optimization instruction; And generating an optimization instruction when the detection evaluation value is not greater than the detection evaluation value threshold or the simulation evaluation value is not greater than the simulation evaluation value threshold.

Description

Bid file personalized compiling system based on large model Technical Field The application relates to the technical field of bidding document compiling, in particular to a large-model-based bidding document personalized compiling system. Background In the bidding field, preparing bid documents with high quality and high efficiency is the key for enterprises to acquire projects. However, the existing bidding document compiling method mainly depends on manual operation, but has insufficient analysis depth for bidding documents, so that the enterprise resources are low in matching efficiency and matching precision with bidding requirements, the bidding documents are uneven in content, a checking means is lacked, and the bidding probability is reduced. Disclosure of Invention In order to solve the technical problems, the application provides a bid document personalized compiling system based on a large model, which constructs a demand knowledge graph according to the demand priority of demand characteristic indexes of bid documents, constructs a real-time response knowledge graph by combining an enterprise database, determines a real-time data set and inputs the real-time data set into the constructed large language model to obtain an initial bid document, performs quality detection and competition simulation, deeply mines the bid document and determines key demand characteristics, improves the matching efficiency and the matching precision of enterprise resources and bid demands, timely discovers and optimizes the defects in the initial bid document, and increases the bid probability. In some embodiments of the present application, there is provided a big model-based bid document personalization system comprising: the analysis module is used for determining a plurality of demand characteristic indexes of the bidding documents, calculating demand priorities of the demand characteristic indexes and constructing a demand knowledge graph according to the demand priorities; The determining module is used for constructing a real-time response knowledge graph according to the enterprise database and determining a real-time data set based on the demand knowledge graph and the real-time response knowledge graph; the construction module is used for constructing a large language model, inputting the real-time data set into the large model and obtaining an initial bidding document; And the verification module is used for carrying out quality detection and competition simulation on the initial bidding file, and judging whether to generate an optimization instruction according to the detection result and the simulation result. In some embodiments of the present application, calculating demand priorities for demand characteristic indicators includes: analyzing the bid-inviting file and extracting key field information in the bid-inviting file; identifying and structuring key field information in the bidding document based on natural language processing technology to obtain a plurality of bidding requirement information; Generating a plurality of requirement characteristic indexes according to all bidding requirement information; Generating a plurality of historical demand characteristic indexes of the historical bidding documents, and carrying out similarity analysis on the historical demand characteristic indexes and the demand characteristic indexes of the current bidding documents to obtain similarity; extracting historical bidding documents with similarity larger than a preset similarity threshold, and collecting historical bidding data and historical bidding results of each extracted historical bidding document; Dividing the extracted historical bidding documents according to the historical bidding results, and constructing a first bidding data set and a second bidding data set according to the dividing results and the corresponding historical bidding data; the first bidding data set comprises a plurality of first bidding data subsets, the second bidding data set comprises a plurality of second bidding data subsets, and each bidding data subset is mapped with a corresponding weight coefficient; And comparing and analyzing the first bidding data set and the second bidding data set, and determining the demand priority of each bidding demand information according to the analysis result. In some embodiments of the present application, determining the demand priority of each bidding demand information based on the analysis results includes: performing association analysis on historical bid data in each first bid data subset and each second bid data subset and the demand characteristic index to obtain association coefficients; if the association coefficient of the historical bidding data and the demand characteristic index in the first bidding data subset is larger than a preset association coefficient threshold value, setting the corresponding historical bidding data as a standard data set corresponding to the demand charac