Bid document personalized compilation system based on large model

Abstract

The invention relates to the technical field of bidding document compilation, and discloses a bidding document personalized compilation system based on a large model, and the system comprises an analysis module which is used for determining a plurality of demand feature indexes of a bidding document, calculating the demand priority of the demand feature indexes, and constructing a demand knowledge graph according to the demand priority; the determination 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 and inputting the real-time data set into the large model to obtain an initial bidding file; and the verification module is used for carrying out quality detection and competition simulation on the initial bidding file, and judging whether an optimization instruction is generated or not according to a detection result and a simulation result, so that the matching efficiency and the matching precision of enterprise resources and bidding demands are improved, the defects in the initial bidding file are found and optimized in time, and the bidding winning probability is increased.

Description

Technical Field

[0001] This application relates to the field of bid document preparation technology, and in particular to a personalized bid document preparation system based on a large model. Background Technology

[0002] In the bidding field, preparing high-quality and efficient bid documents is crucial for companies to win projects. However, existing bid document preparation methods mainly rely on manual operation, which lacks sufficient depth in analyzing the bidding documents. This results in low efficiency and accuracy in matching company resources with bidding needs, and the content of bid documents varies greatly, lacking verification methods, thus reducing the probability of winning the bid. Summary of the Invention

[0003] To address the aforementioned technical challenges, this application provides a personalized bid document preparation system based on a large model. This system constructs a requirement knowledge graph according to the requirement priority of the tender document's characteristic indicators, and combines this with a real-time response knowledge graph built from an enterprise database. The system determines a real-time dataset and inputs it into the constructed large language model to obtain an initial bid document. This initial bid document undergoes quality testing and competitive simulation, deeply mining the tender document to identify key requirement features. This improves the matching efficiency and accuracy between enterprise resources and tender requirements, promptly identifies and optimizes deficiencies in the initial bid document, and increases the probability of winning the bid.

[0004] In some embodiments of this application, a personalized tender document preparation system based on a large model is provided, including:

[0005] The parsing module is used to determine several requirement characteristic indicators in the tender documents, calculate the requirement priority of the requirement characteristic indicators, and construct a requirement knowledge graph according to the requirement priority.

[0006] The determination module is used to construct a real-time response knowledge graph based on the enterprise database, and to determine the real-time dataset based on the demand knowledge graph and the real-time response knowledge graph.

[0007] The building module is used to construct a large language model, inputting real-time datasets into the large model to obtain the initial tender documents;

[0008] The verification module is used to perform quality checks on the initial bid documents and conduct competitive simulations. Based on the test results and simulation results, it determines whether to generate optimization instructions.

[0009] In some embodiments of this application, the calculation of demand priority based on demand characteristic indicators includes:

[0010] Parse the tender documents and extract key field information from them;

[0011] Based on natural language processing technology, key field information in the bidding documents is identified and structured to obtain several bidding requirements.

[0012] Several demand characteristic indicators are generated based on all bidding demand information.

[0013] Generate several historical demand characteristic indicators from historical bidding documents and perform similarity analysis with the demand characteristic indicators of the current bidding documents to obtain the similarity score.

[0014] Extract historical tender documents with a similarity greater than a preset similarity threshold, and collect historical bidding data and historical bidding results for each extracted historical tender document;

[0015] The extracted historical bidding documents are divided according to the historical bidding results, and the first bidding dataset and the second bidding dataset are constructed based on the division results and the corresponding historical bidding data.

[0016] The first bidding dataset includes several first bidding data subsets, and the second bidding dataset includes several second bidding data subsets, with each bidding data subset mapped to a corresponding weight coefficient.

[0017] A comparative analysis was conducted on the first and second bid datasets, and the priority of each bidding requirement was determined based on the analysis results.

[0018] In some embodiments of this application, the demand priority of each bidding requirement is determined based on the analysis results, including:

[0019] The historical bidding data in each first and second subset of bidding data are correlated with demand characteristic indicators to obtain the correlation coefficient.

[0020] If the correlation coefficient between the historical bidding data and the demand characteristic indicators in the first bidding data subset is greater than the preset correlation coefficient threshold, the corresponding historical bidding data will be set as a standard dataset for the corresponding demand characteristic indicators.

[0021] If the correlation coefficient between the historical bidding data and the demand characteristic indicator in the second bidding data subset is greater than the preset correlation coefficient threshold, the corresponding historical bidding data will be set as an abnormal dataset for the corresponding demand characteristic indicator.

[0022] Several standard datasets and several abnormal datasets are generated sequentially for each demand characteristic indicator;

[0023] A commonality analysis is performed on all standard datasets of the same demand characteristic index to obtain several common characteristics, and the initial weight coefficient of each common characteristic is set based on the frequency of occurrence.

[0024] The difference analysis is performed between each abnormal dataset and its corresponding common feature for the same demand characteristic indicator to obtain several sub-difference coefficients for the same common feature. These are then combined with the weight coefficients corresponding to each abnormal dataset to perform weight processing, resulting in the difference coefficients for the same common feature.

[0025] The comprehensive difference coefficient of the corresponding demand characteristic index is generated based on the difference coefficient of each common feature of the same demand characteristic index and the corresponding initial weight coefficient.

[0026] All demand characteristic indicators are sorted according to the comprehensive difference coefficient, and the demand priority of demand characteristic indicators is set according to the sorting results.

[0027] In some embodiments of this application, a requirement knowledge graph is constructed according to requirement priority, including:

[0028] Based on the priority of requirements, the positional relationship of each requirement feature indicator in the requirement knowledge graph is determined, and combined with the preset graph construction rules, a requirement knowledge graph containing nodes and edges is generated.

[0029] Each node represents a demand feature indicator, and the edges represent the correlation features and correlation weights between different demand feature indicators. Each node is mapped with the bidding demand information of the corresponding demand feature indicator and several corresponding standard datasets.

[0030] In some embodiments of this application, a real-time response knowledge graph is constructed based on an enterprise database, including:

[0031] Based on each node in the demand knowledge graph, the enterprise database is retrieved to obtain the real-time response information corresponding to the bidding demand information at each node, the real-time dataset corresponding to several standard datasets, and the real-time association features corresponding to the association features of each edge.

[0032] The real-time response information at each node in the demand knowledge graph is converted into a response node, and the real-time association features are converted into edges connecting the corresponding response nodes.

[0033] A real-time response knowledge graph is generated based on several response nodes and their corresponding edges. Each response node is mapped to corresponding real-time response information and a real-time dataset.

[0034] In some embodiments of this application, a real-time dataset is determined based on a demand knowledge graph and a real-time response knowledge graph, including:

[0035] Based on the correspondence between the demand knowledge graph and the real-time response knowledge graph, generate several node correspondences and edge correspondences.

[0036] For each node in the corresponding relationship, perform information matching analysis to obtain the information matching degree;

[0037] Perform logical matching analysis on the edges in each edge correspondence to obtain the logical matching degree;

[0038] Pre-set information matching thresholds and logical matching thresholds;

[0039] Filter out node correspondences where the information matching degree is greater than the information matching degree threshold, and edge correspondences where the logical matching degree threshold is greater than the logical matching degree threshold;

[0040] Based on the selected node and edge correspondences, the real-time response nodes and edges in the real-time response knowledge graph are matched with corresponding labels, including trusted labels and unknown labels.

[0041] Generate a credibility coefficient for each real-time response node based on the tag results;

[0042] Adjust the real-time response nodes whose credibility coefficient is less than the preset credibility coefficient threshold until the credibility coefficient is greater than the preset credibility coefficient threshold. Use the real-time response information at the real-time response node and the corresponding real-time dataset as the main dataset, and use the real-time association features corresponding to the edges connected to the real-time response node as the auxiliary dataset.

[0043] The main dataset and auxiliary datasets of all real-time response nodes are combined to form the real-time dataset.

[0044] In some embodiments of this application, the reliability coefficient of each real-time response node is generated based on the tagging results, including:

[0045] The first and second types of edges are determined based on the label results of the edges connected to each real-time response node;

[0046] The formula for calculating the credibility coefficient is as follows:

[0047] ;

[0048] Where K is the confidence coefficient, c1 is the first weight coefficient, c2 is the second weight coefficient, m1 is the number of edges of the first type, m2 is the number of edges of the second type, d1 is the first transformation coefficient, d2 is the second transformation coefficient, and d3 is the third transformation coefficient. For selection coefficients, when hour, =1, when hour, =0, h is the information matching degree of the real-time response node, h0 is the information matching degree threshold, ls is the logical matching degree of the s-th first-type edge, l0 is the logical matching degree threshold, lv is the logical matching degree of the v-th second-type edge, qs is the weight coefficient of the s-th first-type edge, and qv is the weight coefficient of the v-th second-type edge.

[0049] In some embodiments of this application, constructing a large language model includes:

[0050] Obtain the bidding demand information, historical main dataset, and historical auxiliary dataset at each node in the demand knowledge graph, and use them as training input data. Use several standard datasets at each node as training output data.

[0051] The initial large language model is trained based on the training input data and training output data to obtain the large language model;

[0052] The real-time dataset is input into the language big model to generate bid content for each requirement feature indicator and assemble it into an initial bid document.

[0053] In some embodiments of this application, quality checks and competition simulations are performed on the initial tender documents, including:

[0054] A quality inspection model is constructed based on a pre-set bid document quality assessment index system;

[0055] The initial tender documents are inspected based on a quality inspection model to obtain inspection results, which include content completeness, content accuracy, content standardization, content relevance, and content innovation.

[0056] Generate a test evaluation value based on the test results;

[0057] Construct a competition simulation model;

[0058] Several predicted bid documents were generated based on a competition simulation model and some bidding requirements information.

[0059] Several predicted bid documents were compared and analyzed with the initial bid documents to obtain simulation results. The simulation results included several advantages of the initial bid documents over the predicted bid documents.

[0060] Simulated evaluation values ​​are generated based on the simulation results.

[0061] In some embodiments of this application, determining whether to generate optimization instructions based on detection results and simulation results includes:

[0062] Pre-set the detection evaluation value threshold and the simulated evaluation value threshold;

[0063] When the detected evaluation value is greater than the detected evaluation value threshold and the simulated evaluation value is greater than the simulated evaluation value threshold, no optimization instruction is generated;

[0064] An optimization instruction is generated when the detected evaluation value is not greater than the detection evaluation value threshold or the simulated evaluation value is not greater than the simulated evaluation value threshold.

[0065] The personalized bid document preparation system based on large models in this application has the following advantages compared with the prior art:

[0066] By constructing a requirement knowledge graph based on the requirement feature indicators of the bidding documents and prioritizing them, and combining it with the enterprise database to construct a real-time response knowledge graph, the real-time dataset is determined and input into the constructed large language model to obtain the initial bid document, and conduct quality inspection and competition simulation. The bidding documents are deeply mined and key requirement features are determined, which improves the matching efficiency and accuracy between enterprise resources and bidding requirements, timely identifies and optimizes the deficiencies in the initial bid document, and increases the probability of winning the bid. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of a personalized bid document preparation system based on a large model, as described in this application embodiment. Detailed Implementation

[0068] The specific embodiments of this application will be described in further detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate this application, but are not intended to limit the scope of this application.

[0069] In the description of this application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0070] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0071] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0072] like Figure 1As shown in the embodiment of this application, the personalized bid document preparation system based on a large model includes:

[0073] The parsing module is used to determine several requirement characteristic indicators in the tender documents, calculate the requirement priority of the requirement characteristic indicators, and construct a requirement knowledge graph according to the requirement priority.

[0074] The determination module is used to construct a real-time response knowledge graph based on the enterprise database, and to determine the real-time dataset based on the demand knowledge graph and the real-time response knowledge graph.

[0075] The building module is used to construct a large language model, inputting real-time datasets into the large model to obtain the initial tender documents;

[0076] The verification module is used to perform quality checks on the initial bid documents and conduct competitive simulations. Based on the test results and simulation results, it determines whether to generate optimization instructions.

[0077] In some embodiments of this application, the calculation of demand priority based on demand characteristic indicators includes:

[0078] Parse the tender documents and extract key field information from them;

[0079] Based on natural language processing technology, key field information in the bidding documents is identified and structured to obtain several bidding requirements.

[0080] Several demand characteristic indicators are generated based on all bidding demand information.

[0081] Generate several historical demand characteristic indicators from historical bidding documents and perform similarity analysis with the demand characteristic indicators of the current bidding documents to obtain the similarity score.

[0082] Extract historical tender documents with a similarity greater than a preset similarity threshold, and collect historical bidding data and historical bidding results for each extracted historical tender document;

[0083] The extracted historical bidding documents are divided according to the historical bidding results, and the first bidding dataset and the second bidding dataset are constructed based on the division results and the corresponding historical bidding data.

[0084] The first bidding dataset includes several first bidding data subsets, and the second bidding dataset includes several second bidding data subsets, with each bidding data subset mapped to a corresponding weight coefficient.

[0085] A comparative analysis was conducted on the first and second bid datasets, and the priority of each bidding requirement was determined based on the analysis results.

[0086] In this embodiment, the demand characteristic indicators include project background, technical requirements, business requirements, qualification requirements, and format specifications. The bidding demand information includes project name, budget, construction period, enterprise qualification level, personnel qualifications, past performance threshold, core technical indicators, product parameters, service standards, quotation method, and document structure. By deeply analyzing these demand characteristic indicators, we can more accurately grasp the core demands of the bidding party and lay a solid foundation for subsequent demand priority calculation.

[0087] In this embodiment, the first bidding dataset refers to the historical bidding data of historical bidding documents with successful bids, and the second bidding dataset refers to the historical bidding data of historical bidding documents with unsuccessful bids.

[0088] In this embodiment, the first subset of bidding data is constructed from the historical bidding data of the same historical bidding document that won the bid. The corresponding weight coefficient is calculated based on the similarity and the number of historical bidding documents extracted. The second subset of bidding data is constructed from the historical bidding data of the same historical bidding document that did not win the bid. The corresponding weight coefficient is similar and will not be described again here.

[0089] In this embodiment, the similarity is obtained by quantitatively evaluating and weighting the similarity between the demand feature indicators and the historical demand feature indicators of the same historical tender document. The similarity threshold is set according to historical data, and is specifically 0.8 in this application.

[0090] In some embodiments of this application, the demand priority of each bidding requirement is determined based on the analysis results, including:

[0091] The historical bidding data in each first and second subset of bidding data are correlated with demand characteristic indicators to obtain the correlation coefficient.

[0092] If the correlation coefficient between the historical bidding data and the demand characteristic indicators in the first bidding data subset is greater than the preset correlation coefficient threshold, the corresponding historical bidding data will be set as a standard dataset for the corresponding demand characteristic indicators.

[0093] If the correlation coefficient between the historical bidding data and the demand characteristic indicator in the second bidding data subset is greater than the preset correlation coefficient threshold, the corresponding historical bidding data will be set as an abnormal dataset for the corresponding demand characteristic indicator.

[0094] Several standard datasets and several abnormal datasets are generated sequentially for each demand characteristic indicator;

[0095] A commonality analysis is performed on all standard datasets of the same demand characteristic index to obtain several common characteristics, and the initial weight coefficient of each common characteristic is set based on the frequency of occurrence.

[0096] The difference analysis is performed between each abnormal dataset and its corresponding common feature for the same demand characteristic indicator to obtain several sub-difference coefficients for the same common feature. These are then combined with the weight coefficients corresponding to each abnormal dataset to perform weight processing, resulting in the difference coefficients for the same common feature.

[0097] The comprehensive difference coefficient of the corresponding demand characteristic index is generated based on the difference coefficient of each common feature of the same demand characteristic index and the corresponding initial weight coefficient.

[0098] All demand characteristic indicators are sorted according to the comprehensive difference coefficient, and the demand priority of demand characteristic indicators is set according to the sorting results.

[0099] In this embodiment, common features refer to the common manifestations of each requirement feature indicator in the historical bid documents of successful bidders. For example, in the requirement feature indicator of technical requirements, common features may be explicit requirements for specific technical parameters, standardized descriptions of technical implementation processes, etc. These common features reflect the aspects that the bidding parties generally pay attention to and value in historical successful bid cases.

[0100] In this embodiment, setting the first weight coefficient for each common feature based on its frequency of occurrence is to more accurately measure the importance of each common feature in the overall requirements. The higher the frequency of occurrence, the more likely the common feature is reflected in multiple winning bid cases, and the higher its weight coefficient will be.

[0101] In this embodiment, by performing a difference analysis between each abnormal dataset and its corresponding common features for the same demand characteristic indicator, it is possible to identify deviations from common success stories. These differences may represent certain special circumstances or risk points, which are of great significance for a comprehensive understanding of bidding requirements.

[0102] In this embodiment, all demand characteristic indicators are sorted according to the comprehensive difference coefficient. The sorting results can clearly show which demand characteristic indicators have a more significant impact on the bidding results, thereby setting the demand priority of demand characteristic indicators and providing targeted guidance for the subsequent preparation of bid documents, ensuring that those demand characteristic indicators that have a greater impact on the bidding results are met first during the preparation process.

[0103] In this embodiment, several sub-difference coefficients of the same common feature are obtained, including:

[0104] A common feature of the demand characteristic indicators is randomly selected as the target common feature;

[0105] Pre-defined evaluation indicators for common characteristics of the targets;

[0106] Based on several difference evaluation indicators of common target characteristics, we evaluate and analyze several abnormal datasets with the same demand characteristic indicators to obtain the sub-difference coefficient between the common target characteristics and each abnormal dataset.

[0107] The formula for calculating the sub-difference coefficient is as follows:

[0108] ;

[0109] Where Y1 is the sub-difference coefficient, z is the difference transformation coefficient, n is the number of difference evaluation indicators of the target common features, p0i is the reference value of the i-th difference evaluation indicator generated by the target common features, pi is the actual value of the i-th difference evaluation indicator generated by the abnormal dataset, and ai is the weight coefficient of the i-th difference evaluation indicator.

[0110] The sub-difference coefficients between each common feature and several abnormal datasets are generated sequentially.

[0111] In this embodiment, the evaluation indicators for differences between different common features are different and are set according to the specific requirements involved in the common features. For example, for the common feature of technical requirements, the evaluation indicators may include the accuracy of technical parameters, the completeness of the technical implementation process, and the degree of technological innovation; for the common feature of business requirements, the evaluation indicators may include the reasonableness of the quotation, the flexibility of the payment method, and the guarantee of after-sales service.

[0112] In this embodiment, the reference value refers to the ideal standard value of the difference evaluation index generated from the common features of the target, while the actual value is the specific value of the corresponding difference evaluation index extracted from the abnormal dataset. The difference conversion coefficient refers to converting the absolute value of the difference into a value with the same dimension as the sub-difference coefficient. When the absolute value of the difference is larger, the corresponding sub-difference coefficient is larger, indicating that the deviation between each abnormal dataset and the common features of the target is greater, and vice versa, thus providing strong support for subsequent difference analysis and demand priority setting.

[0113] In some embodiments of this application, a requirement knowledge graph is constructed according to requirement priority, including:

[0114] Based on the priority of requirements, the positional relationship of each requirement feature indicator in the requirement knowledge graph is determined, and combined with the preset graph construction rules, a requirement knowledge graph containing nodes and edges is generated.

[0115] Each node represents a demand feature indicator, and the edges represent the correlation features and correlation weights between different demand feature indicators. Each node is mapped with the bidding demand information of the corresponding demand feature indicator and several corresponding standard datasets.

[0116] In this embodiment, the preset map construction rules include, but are not limited to, the hierarchical, parallel, and dependent relationships of demand feature indicators. Prioritizing demand requirements and supplemented by other preset rules, these rules clearly present the complex relationships between various demand feature indicators. For example, in the hierarchical relationship, core demand feature indicators are at the top, with closely related indicators of slightly lower priority below them. The parallel relationship demonstrates the equal importance of demand feature indicators, providing an intuitive and comprehensive demand framework for subsequent tender document preparation, enabling a quick grasp of the overall picture and key points of the bidding requirements.

[0117] In this embodiment, the edges of the demand knowledge graph represent the correlation features between different demand characteristic indicators, which are derived based on actual business logic and historical data mining. These correlation features specifically include the mutual influence and constraint relationships between different demand characteristic indicators in the bidding process. For example, specific technical parameters in the technical requirements may be related to the price in the business requirements; higher technical parameter requirements may correspond to higher prices. The correlation weight is obtained by quantitatively evaluating the tightness and importance of these correlation features. The tighter the correlation feature and the higher its importance, the greater the corresponding correlation weight. When the correlation weight is greater, the edge is shorter, and vice versa.

[0118] In this embodiment, by constructing a requirement knowledge graph according to requirement priority, the relationship and importance between various requirement characteristic indicators can be displayed more intuitively, providing strong support for the subsequent preparation of tender documents.

[0119] In some embodiments of this application, a real-time response knowledge graph is constructed based on an enterprise database, including:

[0120] Based on each node in the demand knowledge graph, the enterprise database is retrieved to obtain the real-time response information corresponding to the bidding demand information at each node, the real-time dataset corresponding to several standard datasets, and the real-time association features corresponding to the association features of each edge.

[0121] The real-time response information at each node in the demand knowledge graph is converted into a response node, and the real-time association features are converted into edges connecting the corresponding response nodes.

[0122] A real-time response knowledge graph is generated based on several response nodes and their corresponding edges. Each response node is mapped to corresponding real-time response information and a real-time dataset.

[0123] In this embodiment, the enterprise database covers various aspects of data information, including the enterprise's past bidding data, project experience, enterprise qualifications, personnel information, and technical capabilities.

[0124] In this embodiment, real-time response information refers to information that matches the bidding requirements information. For example, when a node in the requirements knowledge graph represents a specific technical parameter in the technical requirements, real-time response information such as the company's current actual capability data in terms of that technical parameter and relevant application data in past projects can be obtained by searching the company database. The real-time dataset reflects the company's actual capability performance in response to real-time response information under the current market environment and technological development level.

[0125] In this embodiment, real-time correlation features refer to the actual situation under the correlation features of different demand feature indicators that influence and restrict each other in the current enterprise database. For example, in the current enterprise database, the real-time correlation between specific technical parameters in the technical requirements and the quotation in the business requirements may be manifested as the enterprise increasing its quotation accordingly when the technical parameters are improved, based on its own technical capabilities. The real-time correlation weight is calculated based on the real-time correlation features, and the length of the edge is set.

[0126] In this embodiment, the real-time response knowledge graph can clearly present the enterprise's real-time response capability to bidding requirements and the real-time correlation between various requirement characteristic indicators, providing more practical and timely support for the subsequent personalized preparation of bid documents.

[0127] In some embodiments of this application, a real-time dataset is determined based on a demand knowledge graph and a real-time response knowledge graph, including:

[0128] Based on the correspondence between the demand knowledge graph and the real-time response knowledge graph, generate several node correspondences and edge correspondences.

[0129] For each node in the corresponding relationship, perform information matching analysis to obtain the information matching degree;

[0130] Perform logical matching analysis on the edges in each edge correspondence to obtain the logical matching degree;

[0131] Pre-set information matching thresholds and logical matching thresholds;

[0132] Filter out node correspondences where the information matching degree is greater than the information matching degree threshold, and edge correspondences where the logical matching degree threshold is greater than the logical matching degree threshold;

[0133] Based on the selected node and edge correspondences, the real-time response nodes and edges in the real-time response knowledge graph are matched with corresponding labels, including trusted labels and unknown labels.

[0134] Generate a credibility coefficient for each real-time response node based on the tag results;

[0135] Adjust the real-time response nodes whose credibility coefficient is less than the preset credibility coefficient threshold until the credibility coefficient is greater than the preset credibility coefficient threshold. Use the real-time response information at the real-time response node and the corresponding real-time dataset as the main dataset, and use the real-time association features corresponding to the edges connected to the real-time response node as the auxiliary dataset.

[0136] The main dataset and auxiliary datasets of all real-time response nodes are combined to form the real-time dataset.

[0137] In this embodiment, real-time response nodes correspond to nodes in the demand knowledge graph. Simultaneously, real-time association features are converted into edges connecting the corresponding response nodes, enabling the real-time response knowledge graph to intuitively display the enterprise's real-time response to various demand feature indicators. For example, if a response node represents an enterprise's technical solution for a specific technical requirement, then the edges connecting that node may reflect the association between that technical solution and other demand feature indicators (such as business requirements, qualification requirements, etc.).

[0138] In this embodiment, each node correspondence includes a node in the demand knowledge graph and a real-time response node, and each edge correspondence includes an edge in the demand knowledge graph and an edge connecting the corresponding real-time response node.

[0139] In this embodiment, information matching analysis of node correspondence refers to the consistency evaluation of the bidding demand information in the node with the real-time response information in the real-time response node, as well as the consistency between several standard datasets and the real-time dataset. The higher the consistency, the greater the information matching degree, and vice versa.

[0140] In this embodiment, logical matching analysis of edge correspondences refers to evaluating the logical consistency between the association features and weights represented by edges in the requirement knowledge graph and the real-time association features and weights represented by edges in the real-time response knowledge graph. Higher consistency indicates a greater logical match, and vice versa. For example, analyzing whether the logical association between technical requirements and business requirements in the requirement knowledge graph matches the logical association between these two in the enterprise's real-time response.

[0141] In this embodiment, the information matching degree threshold and the logical matching degree threshold refer to the minimum matching degree required to determine the validity of the node correspondence and edge correspondence, and are set based on historical data.

[0142] In this embodiment, real-time response nodes with a credibility coefficient less than a preset credibility coefficient threshold are adjusted. The adjustment method can be determined according to the specific situation. If it is due to insufficient information matching, the enterprise database can be further explored to mine more matching real-time response information. If it is due to insufficient logical matching, the reasons for the difference between the actual situation of the enterprise and the logic of the knowledge graph can be analyzed, and the real-time association features or association weights can be corrected. After adjustment, the credibility coefficient is recalculated until the credibility coefficient is greater than the preset credibility coefficient threshold.

[0143] In this embodiment, the main dataset directly targets the key requirement characteristics indicators of the bidding requirements, which can accurately reflect the actual situation and capabilities of enterprises in these aspects. The auxiliary dataset can supplement and optimize the key requirement characteristics indicators and reflect the real-time correlation between different requirement characteristics indicators. This helps to fully consider the mutual influence between various requirement characteristics indicators when preparing the bid documents, making the bid documents more complete and reasonable.

[0144] In this embodiment, the real-time input dataset covers real-time and accurate information on various demand feature indicators of enterprises for bidding documents, and combines the priority of demand feature indicators to provide a solid data foundation for the subsequent personalized preparation of bid documents based on a large model.

[0145] In some embodiments of this application, the reliability coefficient of each real-time response node is generated based on the tagging results, including:

[0146] The first and second types of edges are determined based on the label results of the edges connected to each real-time response node;

[0147] The formula for calculating the credibility coefficient is as follows:

[0148] ;

[0149] Where K is the confidence coefficient, c1 is the first weight coefficient, c2 is the second weight coefficient, m1 is the number of edges of the first type, m2 is the number of edges of the second type, d1 is the first transformation coefficient, d2 is the second transformation coefficient, and d3 is the third transformation coefficient. For selection coefficients, when hour, =1, when hour, =0, h is the information matching degree of the real-time response node, h0 is the information matching degree threshold, ls is the logical matching degree of the s-th first-type edge, l0 is the logical matching degree threshold, lv is the logical matching degree of the v-th second-type edge, qs is the weight coefficient of the s-th first-type edge, and qv is the weight coefficient of the v-th second-type edge.

[0150] In this embodiment, the first type of edge refers to the edge with a trusted label, and the second type of edge refers to the edge with an unknown label.

[0151] In this embodiment, the first conversion coefficient refers to converting the information matching difference into a value with the same dimension as the credibility coefficient. When the information matching difference is greater than 0 and larger, the credibility coefficient is larger, and vice versa. The value range is (-1, 1). The second conversion coefficient refers to converting the sum of logical matching differences (all greater than 0) into a value with the same dimension as the confidence coefficient. When the logical matching difference is larger, the confidence coefficient is larger, and vice versa. The value range is (0,1). The third conversion coefficient refers to converting the sum of logical matching differences (all less than 0) into a value with the same dimension as the confidence coefficient. The closer the logical matching difference is to 0, the larger the confidence coefficient, and vice versa. The value range is (-1, 0).

[0152] In this embodiment, c1+c2=1, and in this application, c1 is 0.6 and c2 is 0.4.

[0153] In this embodiment, the credibility of each real-time response node is assessed by calculating a credibility coefficient, which provides more reliable data for subsequent bid preparation. A high credibility coefficient for a real-time response node indicates a high degree of matching between the real-time response information and the bidding requirements, and good logical consistency of its associated real-time features. Conversely, a low credibility coefficient requires further analysis of the reasons, such as insufficient information matching or inadequate logical matching, and then corresponding adjustment measures should be taken to ensure that the data entering the bid preparation stage has undergone rigorous screening and evaluation, thereby improving the quality and competitiveness of the bid.

[0154] In some embodiments of this application, constructing a large language model includes:

[0155] Obtain the bidding demand information, historical main dataset, and historical auxiliary dataset at each node in the demand knowledge graph, and use them as training input data. Use several standard datasets at each node as training output data.

[0156] The initial large language model is trained based on the training input data and training output data to obtain the large language model;

[0157] The real-time dataset is input into the language big model to generate bid content for each requirement feature indicator and assemble it into an initial bid document.

[0158] In this embodiment, the initial large language model refers to a pre-built model framework with basic language understanding and generation capabilities, which provides the basic architecture and initial parameter settings for subsequent training based on specific bidding requirements information and related datasets.

[0159] In this embodiment, the initial large language model is typically pre-trained on a large amount of general text data, mastering certain language rules, grammatical knowledge, and semantic understanding capabilities. It is then further trained in a targeted manner by combining bidding requirement information, historical main datasets, historical auxiliary datasets, and several corresponding standard datasets from the requirement knowledge graph. This allows it to learn the complex relationships between different bidding requirement feature indicators, the company's past successful experience in bidding, and its actual performance capabilities, thereby improving the accuracy and professionalism of personalized bid document preparation and better meeting the needs of actual bidding business.

[0160] In this embodiment, several standard datasets at each node are used as training output data to clarify the standards and goals that the large model should achieve under specific demand characteristic indicators, thereby improving the accuracy and rationality of the generated tender documents.

[0161] In this embodiment, the trained large language model can comprehensively consider various information in the demand knowledge graph and combine it with the actual situation of the enterprise to generate a tender document that meets the bidding requirements and highlights the enterprise's advantages, which greatly improves the efficiency and quality of tender document preparation.

[0162] In some embodiments of this application, quality checks and competition simulations are performed on the initial tender documents, including:

[0163] A quality inspection model is constructed based on a pre-set bid document quality assessment index system;

[0164] The initial tender documents are inspected based on a quality inspection model to obtain inspection results, which include content completeness, content accuracy, content standardization, content relevance, and content innovation.

[0165] Generate a test evaluation value based on the test results;

[0166] Construct a competition simulation model;

[0167] Several predicted bid documents were generated based on a competition simulation model and some bidding requirements information.

[0168] Several predicted bid documents were compared and analyzed with the initial bid documents to obtain simulation results. The simulation results included several advantages of the initial bid documents over the predicted bid documents.

[0169] Simulated evaluation values ​​are generated based on the simulation results.

[0170] In this embodiment, the preset bid document quality evaluation index system was determined by a combination of factors, including expert discussion, industry standards, and practical bidding experience, and can comprehensively and objectively reflect the quality status of the bid documents.

[0171] In this embodiment, the quality inspection model is constructed using algorithms such as machine learning and deep learning, based on a preset bid document quality evaluation index system.

[0172] In this embodiment, the quality inspection results are uniformly quantified and calculated using a weighted average method. Different weights are assigned according to the importance of each quality assessment indicator. Then, the inspection results of each indicator are weighted and summed to obtain the final quality inspection evaluation value. The higher the evaluation value, the better the bid document performs in all aspects and the more it meets the standards of a high-quality bid document.

[0173] In this embodiment, the content completeness check determines whether the tender document covers all the key contents required by the tender document and whether any important clauses or technical parameters are missing; the content accuracy check determines whether the data and information in the tender document are accurate and consistent with the company's actual situation and the tender requirements; the content standardization check examines whether the format, layout, font, etc. of the tender document meet the format requirements stipulated in the tender document; the content relevance check assesses whether the tender document closely revolves around the tender requirements, highlights the company's own advantages and characteristics, and differentiates itself from competitors; and the content innovation check considers whether the tender document proposes novel technical solutions, service models, or solutions.

[0174] In this embodiment, the competition simulation model is constructed based on a variety of factors, including historical bidding data, market environment information, and bidding demand characteristics.

[0175] In this embodiment, the advantages refer to the significant strengths exhibited by the initial bid document compared to the predicted bid document. The simulated evaluation value is the result of a comprehensive quantitative assessment of these advantages. By meticulously comparing the predicted and initial bid documents across various key dimensions, such as the innovativeness of the technical solution, the superiority of the service model, and the reasonableness of the price, the advantages of the initial bid document are determined. These advantages are then quantified and scored, ultimately yielding the simulated evaluation value. A higher simulated evaluation value indicates a greater competitive advantage for the initial bid document compared to the predicted bid document in the market. Through this competitive simulation, companies can clearly understand the competitiveness of their bid documents in the market and identify potential shortcomings.

[0176] In some embodiments of this application, determining whether to generate optimization instructions based on detection results and simulation results includes:

[0177] Pre-set the detection evaluation value threshold and the simulated evaluation value threshold;

[0178] When the detected evaluation value is greater than the detected evaluation value threshold and the simulated evaluation value is greater than the simulated evaluation value threshold, no optimization instruction is generated;

[0179] An optimization instruction is generated when the detected evaluation value is not greater than the detection evaluation value threshold or the simulated evaluation value is not greater than the simulated evaluation value threshold.

[0180] In this embodiment, the detection evaluation value threshold and the simulated evaluation value threshold are set based on the industry average level and the company's past bidding experience. They are the minimum detection evaluation value and the minimum simulated evaluation value for judging whether the initial bid documents meet the quality requirements and competitive requirements.

[0181] In this embodiment, after generating optimization instructions, the system determines the direction and focus of optimization based on the specific problems found in the detection and simulation results. For example, if the detection results show insufficient completeness, the system will analyze which key clauses or technical parameters are missing and supplement them accordingly. If the simulation results show insufficient innovation in the technical solution, unclear superiority of the service model, or lack of reasonableness in the pricing, optimization suggestions may include adjusting the technical solution, improving the service model, and optimizing the pricing strategy, aiming to enhance the competitiveness of the initial tender documents in the market.

[0182] In this embodiment, after the optimization suggestions are made, the system will regenerate the optimized bid document and re-perform quality checks and competition simulations to ensure that the optimized bid document can meet the quality and competition requirements.

[0183] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and substitutions can be made without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application.

Claims

1. A personalized bid document preparation system based on a large model, characterized in that: include: The parsing module is used to determine several requirement characteristic indicators in the tender documents, calculate the requirement priority of the requirement characteristic indicators, and construct a requirement knowledge graph according to the requirement priority. The determination module is used to construct a real-time response knowledge graph based on the enterprise database, and to determine the real-time dataset based on the demand knowledge graph and the real-time response knowledge graph. The building module is used to construct a large language model, inputting real-time datasets into the large model to obtain the initial tender documents; The verification module is used to perform quality checks on the initial bid documents and conduct competitive simulations. Based on the test results and simulation results, it determines whether to generate optimization instructions.

2. The personalized bid document preparation system based on a large model as described in claim 1, characterized in that, Calculate the demand priority based on demand characteristic indicators, including: Parse the tender documents and extract key field information from them; Based on natural language processing technology, key field information in the bidding documents is identified and structured to obtain several bidding requirements. Several demand characteristic indicators are generated based on all bidding demand information. Generate several historical demand characteristic indicators from historical bidding documents and perform similarity analysis with the demand characteristic indicators of the current bidding documents to obtain the similarity score. Extract historical tender documents with a similarity greater than a preset similarity threshold, and collect historical bidding data and historical bidding results for each extracted historical tender document; The extracted historical bidding documents are divided according to the historical bidding results, and the first bidding dataset and the second bidding dataset are constructed based on the division results and the corresponding historical bidding data. The first bidding dataset includes several first bidding data subsets, and the second bidding dataset includes several second bidding data subsets, with each bidding data subset mapped to a corresponding weight coefficient. A comparative analysis was conducted on the first and second bid datasets, and the priority of each bidding requirement was determined based on the analysis results.

3. The personalized bid document preparation system based on a large model as described in claim 2, characterized in that, Based on the analysis results, the priority of each bidding requirement information is determined, including: The historical bidding data in each first and second subset of bidding data are correlated with demand characteristic indicators to obtain the correlation coefficient. If the correlation coefficient between the historical bidding data and the demand characteristic indicators in the first bidding data subset is greater than the preset correlation coefficient threshold, the corresponding historical bidding data will be set as a standard dataset for the corresponding demand characteristic indicators. If the correlation coefficient between the historical bidding data and the demand characteristic indicator in the second bidding data subset is greater than the preset correlation coefficient threshold, the corresponding historical bidding data will be set as an abnormal dataset for the corresponding demand characteristic indicator. Several standard datasets and several abnormal datasets are generated sequentially for each demand characteristic indicator; A commonality analysis is performed on all standard datasets of the same demand characteristic index to obtain several common characteristics, and the initial weight coefficient of each common characteristic is set based on the frequency of occurrence. The difference analysis is performed between each abnormal dataset and its corresponding common feature for the same demand characteristic indicator to obtain several sub-difference coefficients for the same common feature. These are then combined with the weight coefficients corresponding to each abnormal dataset to perform weight processing, resulting in the difference coefficients for the same common feature. The comprehensive difference coefficient of the corresponding demand characteristic index is generated based on the difference coefficient of each common feature of the same demand characteristic index and the corresponding initial weight coefficient. All demand characteristic indicators are sorted according to the comprehensive difference coefficient, and the demand priority of demand characteristic indicators is set according to the sorting results.

4. The personalized bid document preparation system based on a large model as described in claim 3, characterized in that, Construct a requirement knowledge graph according to requirement priority, including: Based on the priority of requirements, the positional relationship of each requirement feature indicator in the requirement knowledge graph is determined, and combined with the preset graph construction rules, a requirement knowledge graph containing nodes and edges is generated. Each node represents a demand feature indicator, and the edges represent the correlation features and correlation weights between different demand feature indicators. Each node is mapped with the bidding demand information of the corresponding demand feature indicator and several corresponding standard datasets.

5. The personalized bid document preparation system based on a large model as described in claim 4, characterized in that, A real-time response knowledge graph is constructed based on the enterprise database, including: Based on each node in the demand knowledge graph, the enterprise database is retrieved to obtain the real-time response information corresponding to the bidding demand information at each node, the real-time dataset corresponding to several standard datasets, and the real-time association features corresponding to the association features of each edge. The real-time response information at each node in the demand knowledge graph is converted into a response node, and the real-time association features are converted into edges connecting the corresponding response nodes. A real-time response knowledge graph is generated based on several response nodes and their corresponding edges. Each response node is mapped to corresponding real-time response information and a real-time dataset.

6. The personalized bid document preparation system based on a large model as described in claim 5, characterized in that, The real-time dataset is determined based on the demand knowledge graph and the real-time response knowledge graph, including: Based on the correspondence between the demand knowledge graph and the real-time response knowledge graph, generate several node correspondences and edge correspondences. For each node in the corresponding relationship, perform information matching analysis to obtain the information matching degree; Perform logical matching analysis on the edges in each edge correspondence to obtain the logical matching degree; Pre-set information matching thresholds and logical matching thresholds; Filter out node correspondences where the information matching degree is greater than the information matching degree threshold, and edge correspondences where the logical matching degree threshold is greater than the logical matching degree threshold; Based on the selected node and edge correspondences, the real-time response nodes and edges in the real-time response knowledge graph are matched with corresponding labels, including trusted labels and unknown labels. Generate a credibility coefficient for each real-time response node based on the tag results; Adjust the real-time response nodes whose credibility coefficient is less than the preset credibility coefficient threshold until the credibility coefficient is greater than the preset credibility coefficient threshold. Use the real-time response information at the real-time response node and the corresponding real-time dataset as the main dataset, and use the real-time association features corresponding to the edges connected to the real-time response node as the auxiliary dataset. The main dataset and auxiliary datasets of all real-time response nodes are combined to form the real-time dataset.

7. The personalized bid document preparation system based on a large model as described in claim 6, characterized in that, The credibility coefficient for each real-time response node is generated based on the tagging results, including: The first and second types of edges are determined based on the label results of the edges connected to each real-time response node; The formula for calculating the credibility coefficient is as follows: ； Where K is the confidence coefficient, c1 is the first weight coefficient, c2 is the second weight coefficient, m1 is the number of edges of the first type, m2 is the number of edges of the second type, d1 is the first transformation coefficient, d2 is the second transformation coefficient, and d3 is the third transformation coefficient. For selection coefficients, when hour, =1, when hour, =0, h is the information matching degree of the real-time response node, h0 is the information matching degree threshold, ls is the logical matching degree of the s-th first-type edge, l0 is the logical matching degree threshold, lv is the logical matching degree of the v-th second-type edge, qs is the weight coefficient of the s-th first-type edge, and qv is the weight coefficient of the v-th second-type edge.

8. The personalized bid document preparation system based on a large model as described in claim 7, characterized in that, Building a large language model includes: Obtain the bidding demand information, historical main dataset, and historical auxiliary dataset at each node in the demand knowledge graph, and use them as training input data. Use several standard datasets at each node as training output data. The initial large language model is trained based on the training input data and training output data to obtain the large language model; The real-time dataset is input into the language big model to generate bid content for each requirement feature indicator and assemble it into an initial bid document.

9. The personalized bid document preparation system based on a large model as described in claim 8, characterized in that, Quality checks and competitive simulations were conducted on the initial bid documents, including: A quality inspection model is constructed based on a pre-set bid document quality assessment index system; The initial tender documents are inspected based on a quality inspection model to obtain inspection results, which include content completeness, content accuracy, content standardization, content relevance, and content innovation. Generate a test evaluation value based on the test results; Construct a competition simulation model; Several predicted bid documents were generated based on a competition simulation model and some bidding requirements information. Several predicted bid documents were compared and analyzed with the initial bid documents to obtain simulation results. The simulation results included several advantages of the initial bid documents over the predicted bid documents. Simulated evaluation values ​​are generated based on the simulation results.

10. The personalized bid document preparation system based on a large model as described in claim 9, characterized in that, Whether to generate optimization instructions is determined based on the test results and simulation results, including: Pre-set the detection evaluation value threshold and the simulated evaluation value threshold; When the detected evaluation value is greater than the detected evaluation value threshold and the simulated evaluation value is greater than the simulated evaluation value threshold, no optimization instruction is generated; An optimization instruction is generated when the detected evaluation value is not greater than the detection evaluation value threshold or the simulated evaluation value is not greater than the simulated evaluation value threshold.

Images