CN-121997921-A - Bid file response analysis system and method based on large model

CN121997921ACN 121997921 ACN121997921 ACN 121997921ACN-121997921-A

Abstract

The invention relates to the technical field of bid analysis, and discloses a bid file response analysis system and method based on a large model, wherein the current text content of a bid file is obtained; the method comprises the steps of collecting a historical sample set of a bidding document, constructing a text field evaluation model based on the historical sample set, outputting text content evaluation degree values of each current text content based on the text field evaluation model, calculating response analysis degree values of the bidding document according to all the text content evaluation degree values, presetting a preset response analysis degree value, judging whether the bidding document meets compliance requirements according to the relation between the response analysis degree values and the preset response analysis degree values, combining large-model semantic understanding capability and merging historical bidding data, ensuring evaluation analysis precision of the text content of the bidding document, further realizing response analysis of the bidding document, accurately judging whether the bidding document is compliant, and improving accuracy, efficiency and automation level of response analysis of the bidding document.

Inventors

TIAN SHIDONG
HUANG CHAOBO
HUI TIANLI

Assignees

华能招采数字科技有限公司

Dates

Publication Date: 20260508
Application Date: 20251201

Claims (10)

1. A bid document response analysis method based on a large model, comprising: Receiving a response analysis instruction of a bidding document, and obtaining the current text content of the bidding document, wherein the text content is a structured text content; Collecting a historical sample set of the bidding document, and constructing a text field evaluation model based on the historical sample set; outputting a text content evaluation degree value of each current text content based on the text field evaluation model, and calculating a response analysis degree value of the bidding document according to all the text content evaluation degree values; And presetting a preset response analysis metric value, and judging whether the bidding document meets compliance requirements according to the relation between the response analysis metric value and the preset response analysis metric value.
2. The large model based bid document response analysis method of claim 1, further comprising, prior to receiving the bid document response analysis instruction: Acquiring a previous instruction and determining an instruction category of the previous instruction; Matching the instruction category of the bidding document response analysis instruction with that of the previous instruction, and judging whether the bidding document response analysis instruction is the same as the previous instruction; if the bidding document response analysis instruction is matched with the instruction category of the previous instruction, judging that the bidding document response analysis instruction is the same as the previous instruction, acquiring a first time node for transmitting the bidding document response analysis instruction, and acquiring a second time node for transmitting the previous instruction; Calculating a time node difference value of the first time node and the second time node, and judging whether the bidding document response analysis instruction can be received or not according to the relation between the time node difference value and a preset time node difference value; If the time node difference value is larger than or equal to the preset time node difference value, judging that the bidding document response analysis instruction can be received; If the time node difference value is smaller than the preset time node difference value, judging that the bidding document response analysis instruction cannot be received, generating a log reminder, and sending the log reminder; and if the instruction category of the response analysis instruction of the bidding document is not matched with that of the previous instruction, judging that the response analysis instruction of the bidding document is not identical with that of the previous instruction, and receiving the response analysis instruction of the bidding document.
3. The large model-based bid document response analysis method of claim 1, wherein when collecting a historical sample set of the bid document, constructing a text field evaluation model based on the historical sample set comprises: Acquiring the history sample set, wherein the history sample set comprises a plurality of bidding document samples and scoring labels corresponding to the bidding document samples; Performing feature extraction on the bidding document sample to obtain structural feature data, wherein the feature extraction comprises text feature extraction and numerical feature extraction; Constructing a scoring model, wherein the model comprises an input layer, at least one hiding layer and an output layer; Training the scoring model using the structured feature data, adjusting model parameters by optimizing a loss function; And introducing an attention mechanism in the training process, and weighting different characteristics of the bidding document to obtain the text field evaluation model.
4. The large model based bidder response analysis method of claim 1, wherein when calculating the response analysis metric value of the bidder according to all text content evaluation degree values, comprising: Determining a standard text content evaluation degree value corresponding to each text content evaluation degree value; And calculating a response analysis metric value of the bidding document according to the text content evaluation level value and the standard text content evaluation level value.
5. The large model based bidder response analysis method of claim 4, wherein when calculating the response analysis metric value of the bidder according to the text content evaluation degree value and the standard text content evaluation degree value, comprising: Determining a characteristic text content evaluation degree value based on the text content evaluation degree value, and determining a text content evaluation degree value variation according to the text content evaluation degree value and the characteristic text content evaluation degree value; Determining a characteristic standard text content evaluation degree value based on the standard text content evaluation degree value, and determining a standard text content evaluation degree value variation according to the standard text content evaluation degree value and the characteristic standard text content evaluation degree value; And determining the absolute value of the difference between the text content evaluation degree value variation and the standard text content evaluation degree value variation to obtain the response analysis degree value of the bidding document.
6. The big model based bidder response analysis method of claim 5, wherein when determining a characteristic text content evaluation degree value based on the text content evaluation degree value and determining a text content evaluation degree value variation according to the text content evaluation degree value and the characteristic text content evaluation degree value, comprising: Calculating the average value of all the text content evaluation degree values, and determining the median of all the text content evaluation degree values; calculating a first difference absolute value between the text content evaluation degree value and the mean value; calculating a second difference absolute value between the text content evaluation degree value and the median; Taking the product value of the first difference absolute value and the second difference absolute value as a characteristic text content evaluation degree value of the text content evaluation degree value; sorting all the characteristic text content evaluation degree values in a descending order, forming a first initial characteristic text content evaluation degree value sequence according to the maximum characteristic text content evaluation degree value and the adjacent characteristic text content evaluation degree value, and taking a third characteristic text content evaluation degree value as a characteristic text content evaluation degree value to be updated; Calculating a first sum of two feature text content evaluation degree values in the first initial feature text content evaluation degree value sequence, calculating a first difference value between the whole attribute factor to be updated and the first sum, judging whether the first difference value is larger than or equal to a preset difference value, if so, adding the feature text content evaluation degree value to be updated into the first initial feature text content evaluation degree value sequence, and updating the first initial feature text content evaluation degree value sequence; If not, taking the first initial characteristic text content evaluation degree value sequence as a characteristic text content evaluation degree value sequence, acquiring a fourth characteristic text content evaluation degree value, forming a second initial characteristic text content evaluation degree value sequence according to the third characteristic text content evaluation degree value and the fourth characteristic text content evaluation degree value, and taking a fifth characteristic text content evaluation degree value as a new characteristic text content evaluation degree value to be updated; Calculating a second sum of two feature text content evaluation degree values in the second initial feature text content evaluation degree value sequence, calculating a second difference between the new overall attribute factor to be updated and the second sum, judging whether the second difference is larger than or equal to the preset difference, if so, adding the new feature text content evaluation degree value to be updated into the second initial feature text content evaluation degree value sequence, and updating the second initial feature text content evaluation degree value sequence; If not, taking the second initial characteristic text content evaluation degree value sequence as a characteristic text content evaluation degree value sequence; performing updating iteration according to the characteristic text content evaluation degree values to obtain a plurality of characteristic text content evaluation degree value sequences; and determining the change amount of the text content evaluation degree value according to all the characteristic text content evaluation degree value sequences.
7. The large model based bidder response analysis method of claim 6, wherein when determining the text content evaluation degree value variation amount according to all the characteristic text content evaluation degree value sequences, comprising: Calculating a corresponding sum value of each characteristic text content evaluation degree value sequence as a sequence sum; Performing curve fitting on all the sequence sums to obtain a sequence and a curve, and extracting a representative mutation value; and determining standard deviation of all sequence sums, and taking the product value of the standard deviation and the representative mutation value as the text content evaluation degree value variation.
8. The large model-based bidder response analysis method of claim 1, wherein when judging whether the bidder meets a compliance requirement according to a relationship between the response analysis metric value and the preset response analysis metric value, comprising: When the response analysis metric value is smaller than the preset response analysis metric value, judging that the bidding document meets compliance requirements; and when the response analysis measurement value is larger than or equal to the preset response analysis measurement value, judging that the bidding document does not meet the compliance requirement.
9. A big model based bid document response analysis system for use in a big model based bid document response analysis method according to any of claims 1 to 8, comprising: the content acquisition module is used for receiving a bid file response analysis instruction and acquiring the current text content of the bid file, wherein the text content is a structured text content; the model construction module is used for collecting a history sample set of the bidding document and constructing a text field evaluation model based on the history sample set; The content calculation module is used for outputting a text content evaluation degree value of each current text content based on the text field evaluation model, and calculating a response analysis degree value of the bidding document according to all the text content evaluation degree values; And the content analysis module is used for presetting a preset response analysis measurement value and judging whether the bidding document meets compliance requirements according to the relation between the response analysis measurement value and the preset response analysis measurement value.
10. The large model based bid document response analysis system of claim 9, further comprising: an instruction processing module for: Acquiring a previous instruction and determining an instruction category of the previous instruction; Matching the instruction category of the bidding document response analysis instruction with that of the previous instruction, and judging whether the bidding document response analysis instruction is the same as the previous instruction; if the bidding document response analysis instruction is matched with the instruction category of the previous instruction, judging that the bidding document response analysis instruction is the same as the previous instruction, acquiring a first time node for transmitting the bidding document response analysis instruction, and acquiring a second time node for transmitting the previous instruction; Calculating a time node difference value of the first time node and the second time node, and judging whether the bidding document response analysis instruction can be received or not according to the relation between the time node difference value and a preset time node difference value; If the time node difference value is larger than or equal to the preset time node difference value, judging that the bidding document response analysis instruction can be received; If the time node difference value is smaller than the preset time node difference value, judging that the bidding document response analysis instruction cannot be received, generating a log reminder, and sending the log reminder; and if the instruction category of the response analysis instruction of the bidding document is not matched with that of the previous instruction, judging that the response analysis instruction of the bidding document is not identical with that of the previous instruction, and receiving the response analysis instruction of the bidding document.

Description

Bid file response analysis system and method based on large model Technical Field The invention relates to the technical field of bid analysis, in particular to a bid file response analysis system and method based on a large model. Background With the rapid development of information technology and the increasing competition of enterprises, bidding activities are increasingly important in business activities. The bidding documents are used as core carriers for enterprises to participate in project competition, and the content quality and compliance of the bidding documents directly influence bid winning results. Conventionally, writing and auditing of a bid document is mainly completed manually, and a reviewer needs to check the response condition of the bid document item by item according to the requirement of the bid document, and evaluate whether the bid document meets the requirements in terms of technical specifications, business terms, laws and regulations and the like. However, since the bidding documents are usually complicated in content and various in structure, and involve professional field knowledge, manual auditing is not only low in efficiency, but also has the problems of strong subjectivity, poor consistency, easiness in occurrence of omission and the like. The prior art applies models to the bidding field, such as assisting in the preliminary screening of bidding documents by keyword matching, template-to-peer approach. However, these methods are often limited to surface features, lack of deep understanding of text semantics, and are difficult to adapt to complex scenes with different industries and different bidding requirements, and especially have limited effect when processing unstructured or semi-structured text. In addition, most of the conventional bid document evaluation systems do not fully consider accumulation and utilization of historical data and lack adaptive learning ability for domain knowledge. Meanwhile, when the overall compliance of the bidding document is comprehensively evaluated, the conventional method mostly adopts simple weighted scoring or rule judgment, cannot effectively quantify the deviation degree of text content, and cannot dynamically adapt to the personalized requirements of different bidding projects. Disclosure of Invention The embodiment of the invention provides a bid file response analysis system and a bid file response analysis method based on a large model, which combine semantic understanding capability of the large model and merge historical bid data to ensure evaluation analysis precision of text content of the bid file, further realize bid file response analysis, accurately judge whether the bid file is compliant, and improve accuracy, efficiency and automation level of bid file response analysis. In order to achieve the above object, the present invention provides a bid document response analysis method based on a large model, comprising: Receiving a response analysis instruction of a bidding document, and obtaining the current text content of the bidding document, wherein the text content is a structured text content; Collecting a historical sample set of the bidding document, and constructing a text field evaluation model based on the historical sample set; outputting a text content evaluation degree value of each current text content based on the text field evaluation model, and calculating a response analysis degree value of the bidding document according to all the text content evaluation degree values; And presetting a preset response analysis metric value, and judging whether the bidding document meets compliance requirements according to the relation between the response analysis metric value and the preset response analysis metric value. Further, before receiving the bid file response analysis instruction, the method further comprises: Acquiring a previous instruction and determining an instruction category of the previous instruction; Matching the instruction category of the bidding document response analysis instruction with that of the previous instruction, and judging whether the bidding document response analysis instruction is the same as the previous instruction; if the bidding document response analysis instruction is matched with the instruction category of the previous instruction, judging that the bidding document response analysis instruction is the same as the previous instruction, acquiring a first time node for transmitting the bidding document response analysis instruction, and acquiring a second time node for transmitting the previous instruction; Calculating a time node difference value of the first time node and the second time node, and judging whether the bidding document response analysis instruction can be received or not according to the relation between the time node difference value and a preset time node difference value; If the time node difference value is larger than or equal to the preset time node diffe