Intelligent auxiliary bid evaluation method and system based on machine learning and cloud platform

Abstract

The invention relates to the technical field of auxiliary bid evaluation, and discloses an intelligent auxiliary bid evaluation method and system based on machine learning and a cloud platform, and the method comprises the steps: automatically analyzing project information and a bid invitation file through a natural language processing module, constructing a structured evaluation task definition, such as a compliance list and a comparison matrix header, and carrying out the automatic analysis of the project information and the bid invitation file; and converting the unstructured bid evaluation requirement into a machine executable task. Meanwhile, the bidding file is parsed in parallel and structured into a unified data format; and based on a preset review rule, performing efficient extraction and compliance verification on mass bidding data to form a multi-dimensional data cube. And then, cleaning, alignment and difference analysis are carried out on the data cubes, a visual compliance report and a KPI comparison matrix are automatically generated, and bid evaluation experts are assisted to make decisions. Therefore, by introducing the machine learning and cloud platform technology, automation, intelligence and standardization of the bid evaluation process are realized, the bid evaluation quality and efficiency are remarkably improved, and powerful technical support is provided for bidding and tendering.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of auxiliary bid evaluation, and more specifically, to an intelligent auxiliary bid evaluation method and system based on machine learning and a cloud platform. BACKGROUND

[0002] The traditional bid evaluation process is highly dependent on manual work and faces many challenges. First, as the scale and complexity of projects increase, the number of bid documents is large and the content is complex, involving technology, business, law and other aspects. Manual review is time-consuming and inefficient. Second, manual bid evaluation is easily influenced by subjective factors, leading to a lack of consistency and objectivity in the evaluation results, and even possible disputes. Third, strict compliance requirements require bid evaluators to carefully check each clause, and any oversight may result in compliance risks. In addition, quickly and accurately extracting key information from massive amounts of unstructured text and making horizontal comparisons is a great test of the professional ability and energy of bid evaluators, and it is difficult to achieve efficient and accurate decision assistance. These problems not only prolong the bid evaluation period and increase labor costs, but also may affect the accuracy and credibility of the bid evaluation results.

[0003] To address the above challenges, the industry has attempted to introduce information technology to assist in bid evaluation. However, existing auxiliary bid evaluation systems mostly stop at electronic document management, process automation or simple keyword search, and fail to fully utilize artificial intelligence technology for deep understanding and intelligent analysis of bid documents. For example, some systems can implement online submission and storage of bid documents, or provide basic text search functions, but they generally lack the ability to automatically parse unstructured text, extract key information, intelligently verify compliance clauses, and conduct multi-dimensional data comparison analysis. In particular, for complex rules and scoring standards in the tender document, and massive amounts of free text in the bid document, existing technologies are difficult to achieve automated and intelligent structured processing and semantic understanding, resulting in bid evaluators still needing to invest a lot of effort in manual reading, comparison and judgment. This limits the efficiency and accuracy of the bid evaluation process and makes it difficult to meet the growing demand for complex project bid evaluation. Therefore, in order to overcome the shortcomings of existing technologies, the present application provides an intelligent auxiliary bid evaluation scheme based on machine learning and a cloud platform. SUMMARY

[0004] To solve the problems of low efficiency, strong subjectivity and high compliance risk of traditional manual bid evaluation, and the lack of intelligence in existing auxiliary systems, the present application is proposed. Embodiments of the present application provide an intelligent auxiliary bid evaluation method and system based on machine learning and a cloud platform.

[0005] According to an aspect of the present application, an intelligent auxiliary bid evaluation method based on machine learning and cloud platform is provided, comprising: obtaining project basic information and a bid document; processing the bid document and the project basic information based on a natural language processing module to obtain a structured evaluation task definition, the structured evaluation task definition comprising a compliance check list and a table header structure of a key information comparison matrix; obtaining a set of bid documents of all bidders, and performing document parsing and content structuring on each bid document in the set of bid documents to obtain a set of bid document structured document data; based on the structured evaluation task definition, performing parallel information extraction and compliance verification on the set of bid document structured document data to obtain extracted bid document data cubes; performing data alignment, clarification and differential analysis on the extracted bid document data cubes to obtain evaluation archive data, the evaluation archive data comprising a final compliance report and a final KPI comparison matrix; and displaying the evaluation archive data.

[0006] In the above-mentioned intelligent auxiliary bid evaluation method based on machine learning and cloud platform, the natural language processing module is used to process the bid document and the project basic information to obtain a structured evaluation task definition, the structured evaluation task definition comprising a compliance check list and a table header structure of a key information comparison matrix, which comprises: performing document parsing on the bid document to obtain bid document text data; performing rule mining on the bid document text data to obtain a candidate compliance check list; performing scoring point mining on the bid document text data to obtain a candidate KPI matrix table header list; and performing expert interactive review and confirmation on the candidate compliance check list and the candidate KPI matrix table header list to obtain the compliance check list and the table header structure of the key information comparison matrix.

[0007] In the above-mentioned intelligent auxiliary bid evaluation method based on machine learning and cloud platform, the rule mining on the bid document text data to obtain a candidate compliance check list comprises: performing rule-based pattern matching on the bid document text data to obtain the candidate compliance check list.

[0008] In the above-mentioned intelligent auxiliary bid evaluation method based on machine learning and cloud platform, the scoring point mining on the bid document text data to obtain a candidate KPI matrix table header list comprises: locating a scoring table section in the bid document text data by keyword matching; performing structured information extraction on the scoring table section to obtain a scoring item set; and performing named entity recognition on each scoring item in the scoring item set to obtain the candidate KPI matrix table header list.

[0009] In the intelligent auxiliary bid evaluation method based on machine learning and cloud platform, based on the structured evaluation task definition, the information extraction and compliance verification of the structured document data set of the bid file are performed in parallel to obtain the extracted bid file data cube, including: extracting a first compliance check from the compliance check list; extracting the first bid file structured document data from the bid file structured document data set; extracting the check keyword from the first compliance check, and performing keyword search on the first bid file structured document data with the check keyword as the query condition to obtain the search result; in response to the search result being a hit, writing the original text segment and source location in the first bid file structured document data into the extracted bid file data cube.

[0010] In the intelligent auxiliary bid evaluation method based on machine learning and cloud platform, based on the structured evaluation task definition, the information extraction and compliance verification of the structured document data set of the bid file are performed in parallel to obtain the extracted bid file data cube, including: extracting a first score point from the table header structure of the key information comparison matrix; based on the first score point, generating a first score point generation question; inputting the first score point generation question and the first bid file structured document data into a QA model to obtain an answer text segment; extracting the value, original text segment, source location and confidence from the answer text segment, and writing them into the extracted bid file data cube.

[0011] In the intelligent auxiliary bid evaluation method based on machine learning and cloud platform, based on the structured evaluation task definition, the information extraction and compliance verification of the structured document data set of the bid file are performed in parallel to obtain the extracted bid file data cube, including: setting the compliance check list and the table header structure of the key information comparison matrix included in the structured evaluation task definition as a lock-free access external queue, and defining a set of bid file structured document data as a plurality of local queues corresponding to a plurality of threads, each local queue including one or more bid file structured document data, wherein the bid file structured document data in the local queue is set to be loaded into the lock-free access external queue when performing information extraction and compliance verification, and the structured evaluation task definition in the lock-free access external queue is extracted by the thread corresponding to the local queue for information extraction and compliance verification.

[0012] In the intelligent auxiliary bid evaluation method based on machine learning and cloud platform, based on the structured review task definition, the structured document data set of the bid file is subjected to parallel information extraction and compliance verification to obtain the extracted bid file data cube, and further comprising: in response to the out-of-column condition of the thread corresponding to the local queue being failed, triggering the thread to obtain bid file structured document data from the external queue; in response to the out-of-column condition of the thread corresponding to the local queue being successful, locking the thread, wherein the scheduling metadata set by the external queue monitors the locking state of the thread.

[0013] According to another aspect of the present application, an intelligent auxiliary bid evaluation system based on machine learning and cloud platform is also provided for performing the above-mentioned intelligent auxiliary bid evaluation system based on machine learning and cloud platform, comprising: a project information and tender file acquisition module for acquiring project basic information and tender files; a review task structured definition module for processing the tender files and project basic information based on a natural language processing module to obtain a structured review task definition, the structured review task definition including a compliance check list and a table header structure of a key information comparison matrix; a bid document intelligent analysis module for acquiring a bid file set of all bidders, and performing document analysis and content structuring on each bid file in the bid file set to obtain a bid file structured document data set; an intelligent review analysis module for performing parallel information extraction and compliance verification on the bid file structured document data set based on the structured review task definition to obtain an extracted bid file data cube; a review result generation module for performing data alignment, clarity and differentiation analysis on the extracted bid file data cube to obtain review archive data, the review archive data including a final compliance report and a final KPI comparison matrix; and a review result display module for displaying the review archive data.

[0014] Compared with the prior art, the intelligent auxiliary bid evaluation method and system based on machine learning and cloud platform provided by the present application automatically analyzes project information and tender files through a natural language processing module, constructs a structured review task definition such as a compliance list and a comparison matrix table header, and converts unstructured bid evaluation requirements into machine executable tasks. At the same time, the bid files are parsed in parallel and structured into a unified data format. Based on the preset review rules, the massive bid data is efficiently extracted and verified for compliance to form a multi-dimensional data cube. Subsequently, the data cube is cleaned, aligned and analyzed for differences to automatically generate a visual compliance report and a KPI comparison matrix to assist bid evaluation experts in decision-making. In this way, by introducing machine learning and cloud platform technology, the bid evaluation process is automated, intelligentized and standardized, significantly improving the bid evaluation quality and efficiency, and providing strong technical support for bidding and tendering. BRIEF DESCRIPTION OF DRAWINGS

[0015] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description thereof taken in conjunction with the accompanying drawings, in which:

[0016] Figure 1 FIG. 1 illustrates a schematic flow chart of an intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to an embodiment of the present application.

[0017] Figure 2 FIG. 2 illustrates a schematic flow chart of S2 in the intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to an embodiment of the present application.

[0018] Figure 3 FIG. 3 illustrates a schematic flow chart of S23 in the intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to an embodiment of the present application.

[0019] Figure 4 FIG. 4 illustrates a schematic flow chart of S4 in the intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to an embodiment of the present application.

[0020] Figure 5 FIG. 5 illustrates a schematic flow chart of another embodiment of S4 in the intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to an embodiment of the present application.

[0021] Figure 6 FIG. 6 illustrates a schematic block diagram of an intelligent auxiliary bid evaluation system based on machine learning and cloud platform according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only a part but not all of the embodiments of the present application. It should be understood that the present application is not limited to the described embodiments.

[0023] Based on this, the present application provides an intelligent auxiliary bid evaluation method based on machine learning and cloud platform, Figure 1 FIG. 1 illustrates a schematic flow chart of an intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to an embodiment of the present application. As shown in FIG. 1, the intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to an embodiment of the present application includes the following steps. Figure 1As shown, the intelligent auxiliary bid evaluation method based on machine learning and cloud platform comprises: S1, obtaining project basic information and a tender document; S2, processing the tender document and the project basic information based on a natural language processing module to obtain a structured evaluation task definition, wherein the structured evaluation task definition comprises a compliance check list and a table header structure of a key information comparison matrix; S3, obtaining a set of tender documents of all bidders, and performing document parsing and content structuring on each tender document in the set of tender documents to obtain a set of tender file structured document data; S4, based on the structured evaluation task definition, performing parallel information extraction and compliance verification on the set of tender file structured document data to obtain extracted tender file data cubes; S5, performing data alignment, clarification and differential analysis on the extracted tender file data cubes to obtain evaluation archive data, wherein the evaluation archive data comprises a final compliance report and a final KPI comparison matrix; and S6, displaying the evaluation archive data.

[0024] Exemplarily, in step S1, the project basic information and the tender document are obtained. It should be understood that the project basic information provides necessary background and management parameters for the bid evaluation activity, ensuring that the bid evaluation work can accurately focus on the core elements of a specific project; and the tender document, as the core normative document of the bidding activity, defines the technical requirements, commercial terms, compliance standards, and specific bid evaluation rules and scoring details of the project in detail, and is the only authoritative basis for subsequent structured evaluation task definition, information extraction and compliance verification. If these key information cannot be accurately and completely obtained, the subsequent intelligent processing will be impossible, and even the deviation or invalidity of the bid evaluation result may be caused, thereby affecting the fairness and efficiency of the bid evaluation.

[0025] In one embodiment, to ensure that the project basic information and the tender document can be obtained comprehensively and accurately, the execution step includes but is not limited to the following implementation manners: for obtaining the project basic information, it can be manually input through a user interactive interface, or the relevant data can be imported from the existing project management system of the enterprise through an interface. The user interactive interface is configured as a graphical operation platform, allowing the user to input the basic information related to the project, including but not limited to the project name, project number, tendering unit, bid opening time, bid evaluation expert group, and other key fields. In addition, to improve the data collection efficiency and accuracy, integration with external project management systems is also supported, and the automatic import of project basic information is realized through standardized data interface protocols (such as API interface, database synchronization, etc.).

[0026] Further, for the process of obtaining the bidding document, the specific implementation is to extract the bidding document content from the electronic document provided by the bidding unit. The electronic document can come from multiple channels, such as electronic bidding platform, email attachment, local storage device or cloud shared file, and the document format usually includes common office document formats such as PDF, DOCX, XLSX, etc. Through the built-in document parsing module, the above files are identified and structured processed, so as to extract the key information for subsequent process.

[0027] Exemplarily, in step S2, the bidding document and the project basic information are processed based on a natural language processing module to obtain a structured review task definition, and the structured review task definition includes a compliance check list and a table header structure of a key information comparison matrix. It should be understood that the bidding document is usually a long free text containing a large number of rules, standards and scoring details. It is time-consuming and laborious to understand and extract these information manually, and it is easy to make mistakes. Through natural language processing technology, the core elements of bid evaluation, i.e. the compliance clauses to be checked and the KPIs to be compared, can be automatically identified and refined, so as to convert manual experience into systematic review logic, greatly improving the efficiency, accuracy and objectivity of bid evaluation.

[0028] In one embodiment, as shown in Figure 2 the bidding document and the project basic information are processed based on a natural language processing module to obtain a structured review task definition, and the structured review task definition includes a compliance check list and a table header structure of a key information comparison matrix, including: S21, document parsing of the bidding document to obtain bidding document text data; S22, rule mining of the bidding document text data to obtain a candidate compliance check list; S23, score point mining of the bidding document text data to obtain a candidate KPI matrix table header list; S24, expert human interaction review and confirmation of the candidate compliance check list and the candidate KPI matrix table header list to obtain the compliance check list and the table header structure of the key information comparison matrix.

[0029] Specifically, first, the bidding document is parsed to obtain bidding document text data. This sub-step is the basis for all subsequent natural language processing, which uses OCR (Optical Character Recognition) technology and PDF parsing library to convert the bidding document into pure text data and remove non-text information such as pictures and tables, to obtain bidding document text data for subsequent processing. Through this processing, it is ensured that the content of the bidding document can exist in a unified and text form that can be directly analyzed by the natural language processing module.

[0030] Secondly, rule mining is performed on the tender document text data to obtain a candidate compliance check list. This sub-step aims to automatically identify potential compliance requirements from the tender document text. In one embodiment, rule mining is performed on the tender document text data to obtain a candidate compliance check list, including: performing rule-based pattern matching on the tender document text data to obtain the candidate compliance check list. In a specific embodiment, a series of rules are pre-set or manually configured, for example: matching the “bidder qualification requirements” chapter, extracting keywords and sentences such as “registered capital not less than 50 million yuan” and “related industry Class A qualification”, and generating candidate compliance check items “whether the registered capital of the bidder meets the requirements” and “whether the bidder has related industry Class A qualification”. Matching the “tender document format requirements” chapter, extracting keywords and sentences such as “tender documents should be sealed with a seal” and “tender documents should be bound into a book”, and generating candidate compliance check items “whether the tender documents are sealed with a seal” and “whether the tender documents are bound into a book”. Through these rules, a preliminary compliance check list is automatically identified and extracted from the tender document text data.

[0031] Thirdly, score point mining is performed on the tender document text data to obtain a candidate KPI matrix table header list. This sub-step aims to identify key indicators for scoring from the tender document text. In one embodiment, as shown in FIG. 8, score point mining is performed on the tender document text data to obtain a candidate KPI matrix table header list, including: S211, locating a scoring table chapter in the tender document text data through keyword matching; S212, performing structured information extraction on the scoring table chapter to obtain a scoring item set; S213, performing named entity recognition on each scoring item in the scoring item set to obtain the candidate KPI matrix table header list. Figure 3

[0032] ​Specifically, first, the scoring table chapter in the tender document text data is located by keyword matching, for example, the keywords such as "scoring standard", "scoring method", "technical score", "commercial score" in the tender document are identified to locate the chapter containing the scoring rules, for example, "Chapter 3 Scoring Method and Standard" in a certain tender document. Then, the scoring table chapter is subjected to structured information extraction to obtain a scoring item set, which uses "table recognition and text parsing technology to extract the scoring item and its corresponding score, description and other information from the located scoring table chapter to form a scoring item set. For example, "technical scheme advancement (20 points)", "project team experience (15 points)", "service response time (10 points)" and the like are extracted. Finally, each scoring item in the scoring item set is subjected to named entity recognition to obtain the candidate KPI matrix table header list, that is, the extracted scoring item set is subjected to named entity recognition (NER), and the key entity in the scoring item is identified as the table header of the KPI. For example, "technical scheme advancement" is identified as the KPI table header from "technical scheme advancement"; "project team experience" is identified as the KPI table header from "project team experience".

[0033] Finally, the candidate compliance check list and the candidate KPI matrix table header list are subjected to expert human interaction review and confirmation to obtain the compliance check list and the table header structure of the key information comparison matrix. Specifically, the automatically generated candidate compliance check list and candidate KPI matrix table header list are displayed to the bid evaluation expert. The bid evaluation expert reviews, modifies, adds or deletes the list, for example, deletes unnecessary check items, combines similar KPIs, or adds new check items or KPIs according to actual bid evaluation needs. After expert confirmation, the final compliance check list and the table header structure of the key information comparison matrix for this bid evaluation are formed. This human-computer interaction link ensures the accuracy of the automatic extraction result and meets the actual bid evaluation needs.

[0034] Exemplarily, in step S3, a set of bid documents of all bidders is obtained, and each bid document in the set of bid documents is subjected to document parsing and content structuring to obtain a set of bid document structured document data. It can be understood that the original form of the bid document is usually a document such as PDF, Word, etc., which contains a large amount of text, table, picture and other information, and lacks a unified logical structure identifier. If the document parsing and content structuring are not performed, the subsequent machine learning model and natural language processing module will not be able to efficiently and accurately extract key information, perform compliance checking or perform semantic understanding.

[0035] In one embodiment, the bid document processing step includes two main execution links, aiming to comprehensively obtain and efficiently process all original bid documents submitted by bidders, thereby providing structured data support for subsequent intelligent analysis and evaluation processes. First, the operation of obtaining a set of bid documents is performed, and its specific implementation includes but is not limited to automatically downloading bid documents from an electronic bidding platform or importing bid documents through manual uploading. Among them, the electronic bidding platform and the system interact through a standardized interface, which can batch obtain all bid documents submitted by bidders participating in the project bidding; at the same time, in the case of non-platform submission or special circumstances, the system supports users to manually upload bid documents through a graphical interface, ensuring the flexibility and integrity of data acquisition. Bid documents usually exist in multiple document formats, such as Microsoft Word (.docx), Portable Document Format (PDF), etc., submitted by different bidding subjects such as Company A, Company B, Company C, etc.

[0036] Secondly, the obtained all bid documents are executed document parsing and content structuring processing. This process uses a multi-format adaptive document parsing tool to identify content and extract structure for different types of bid documents. Specifically, for PDF format files, OCR (Optical Character Recognition) technology is used in combination with PDF structure parsing algorithm to extract text content, table information, image position and page number information; for Word format files, the internal Document Object Model (DOM) is parsed to extract paragraphs, titles, lists, charts and other content and their hierarchical relationships. Subsequently, the above parsing results are uniformly converted into a standard structured data format such as JSON or XML, which includes but is not limited to text content, chapter title, page number, picture coordinate, table data, etc. metadata, and retains the logical structure information of the original document (such as chapter attribution, paragraph hierarchy). For example, the bid document of Company A is parsed and structured into A Company bid document structured document data.

[0037] Exemplarily, in step S4, based on the structured review task definition, parallel information extraction and compliance verification are performed on the set of bid document structured document data to obtain an extracted bid document data cube. It should be understood that traditional manual bid review faces huge workload and challenges at this link, and needs to read, compare and judge word by word, which is extremely easy to miss, make mistakes or have subjective bias. By combining the structured data of the bid document which has uniform format but still needs in-depth analysis, with the defined and machine understandable review task (i.e. the table header structure of the compliance check list and the key information comparison matrix), the key information required for bid review can be automatically and efficiently extracted from the massive text, and each content of the bid document is strictly checked for compliance, thereby significantly improving the efficiency, accuracy and objectivity of bid review, and providing high-quality and structured raw data for subsequent data alignment, cleaning and differentiated analysis.

[0038] In one embodiment, as shown in Figure 4 Based on the structured review task definition, the parallel information extraction and compliance verification are performed on the set of bid document structured document data to obtain an extracted bid document data cube, including: S41, extracting a first compliance check from the compliance check list; S42, extracting a first bid document structured document data from the set of bid document structured document data; S43, extracting a check keyword from the first compliance check, and performing a keyword search in the first bid document structured document data with the check keyword as a query condition to obtain a search result; S44, in response to the search result being a hit, writing an original text segment and a source location in the first bid document structured document data into the extracted bid document data cube.

[0039] Specifically, first, the corresponding compliance check items are extracted from the preset compliance check list, and each compliance requirement is automatically checked in combination with the structured bid document data. For example, the compliance check item such as “whether the registered capital of the bidder meets the requirements” can be extracted, and the related keywords such as “registered capital” and “registered funds” can be identified, and then these keywords are used as query conditions to efficiently search in the structured document data of the corresponding bidder. If a matching text segment (for example, “A company has registered capital of 60 million yuan”) is found in the document, the original text and its source location (for example, the chapter of “Company Basic Situation” on page 5) will be recorded to the unified extracted bid document data cube, thereby realizing automatic extraction and structured storage of compliance information.

[0040] In another embodiment, as shown in Figure 5As shown, based on the structured review task definition, parallel information extraction and compliance verification are performed on the set of bid document structured document data to obtain an extracted bid document data cube, including: S45, extracting first scoring points from the table header structure of the key information comparison matrix; S46, generating first scoring point generation questions based on the first scoring points; S47, inputting the first scoring point generation questions and the first bid document structured document data into a QA model to obtain an answer text segment; S48, extracting an extracted value, an original text segment, a source location, and a confidence from the answer text segment, and writing them into the extracted bid document data cube.

[0041] Specifically, first, scoring points are extracted from a pre-defined key information comparison matrix, such as "technical scheme advancement", and corresponding guiding questions are generated based on the scoring points, such as "Please describe in which aspects of the technical scheme of the bidding party does it reflect advancement?" or "What innovative technologies does the technical scheme of the bidding party adopt?". Subsequently, these questions and the corresponding structured document data of the bidder are input into a pre-trained QA model, such as a model based on BERT or GPT, which can find answers in the document and return relevant answer text segments through its semantic understanding ability. For example, the model may return "Company A adopts AI-based intelligent traffic prediction system and blockchain data security technology in the technical scheme, significantly improving data processing efficiency and security." Further, structured information values, original text content, source locations in the document, and model confidence scores for the answers are extracted from the answers and written into the "extracted bid document data cube".

[0042] Here, it should be understood that the QA model (Question Answering Model) is, as the name implies, an artificial intelligence model that can understand natural language questions and find or generate answers from given text. In this application, the QA model is used to accurately extract key information related to specific scoring points from the massive text of the bid document. This aims to transform the time-consuming and error-prone process of traditional manual reading, understanding, and extracting information into efficient, automated, and accurate machine intelligence processing, thereby significantly improving the efficiency and accuracy of key information acquisition in the bid evaluation process, and providing high-quality structured data for subsequent data comparison analysis and bid evaluation expert decision-making.

[0043] In particular, the QA model employed in the present application is an extractive QA model, which aims to directly locate and extract the precise segment of the answer from the provided text passage, rather than generating a brand new text. This model type is particularly suitable for the present application scenario, as it can ensure the originality and traceability of the extracted information, i.e. the answer is directly derived from the original text of the tender document, facilitating verification by the bid evaluation experts. In a specific embodiment, the input of the QA model consists of two parts: the question and the context. These two texts are first processed by a tokenizer, and converted into a series of discrete tokens. In order to enable the model to distinguish between the question and the context, and establish semantic connections between them, a special separator is usually inserted between the question and the context, and a classifier is added at the beginning of the entire sequence. Subsequently, each token is assigned three types of embedding vectors, which are stacked as the input of the model: first, word embeddings, which capture the semantic information of the token itself; second, passage embeddings, which identify whether the token is derived from the question or the context, thereby helping the model to understand the semantic roles of different parts; and finally, position embeddings, which encode the relative position information of the token in the sequence, to make up for the lack of sequence order perception ability of the Transformer architecture itself. These stacked embedding vectors are then fed into the encoder part of the Transformer. The encoder is stacked by multiple identical Transformer blocks, and the core components of each Transformer block are multi-head self-attention mechanism and feed-forward neural network. The self-attention mechanism allows the model to simultaneously focus on all other tokens in the sequence when processing each token in the sequence, thereby capturing complex dependency relationships and contextual semantics between tokens; the multi-head mechanism enables the model to understand the input sequence from different “representation subspaces” or “attention points”, thereby extracting richer semantic features. The feed-forward neural network performs a non-linear transformation on the output of the attention mechanism, further extracting high-level features. In addition, each Transformer block also contains residual connections and layer normalization, which help to train deeper networks and accelerate model convergence. After the Transformer encoder processes the input sequence, it generates a high-dimensional context vector representation for each token in the sequence. For the extractive QA task, the model usually adds a simple linear layer and a Softmax activation function on top of the encoder output. This linear layer predicts the probability of each token in the input sequence as the start position and end position of the answer. The goal of the model is to find a continuous sequence of tokens, such that the product of the predicted probabilities of the start token and the end token is maximized, and the start position is not later than the end position.

[0044] The data processing process of the QA model mainly includes a training phase and an inference phase. In the training phase, the QA model is trained through supervised learning, which requires a large amount of labeled data set. Each training sample usually contains a context paragraph, a question, and the correct answer segment (i.e. the start and end positions of the answer) of the question in the context. The data preparation process includes concatenating the context and the question, tokenizing, adding special markers, and generating various embeddings, while determining the true start and end token indices of the annotated answer segment in the concatenated sequence. Then, the prepared input is sent to the Transformer model for forward propagation, and the model outputs each token as a prediction score for the start and end positions of the answer. The cross-entropy loss function is used to measure the difference between the model's predicted start / end position probability distribution and the true labeled position, and the gradient is calculated using the backpropagation algorithm. Finally, the model parameters are updated by the optimizer to minimize the loss and improve the model performance. In the inference phase, when a new question and bid document structured document data are received, similar input preparation as in the training phase is performed, including tokenization, concatenation, and embedding of the question and bid document text. Then, the prepared input is sent to the trained QA model for forward propagation, and the model outputs a prediction score for each token in the input sequence as the start and end positions of the answer. All possible answer segments (i.e. all legal start-end token pairs) are traversed, their joint probability (usually the product of the start probability and the end probability) is calculated, and the segment with the highest joint probability is selected as the final answer. At the same time, the model can also output a confidence score representing the degree of confidence in the extracted answer, which helps to evaluate the reliability of the answer. Finally, the predicted token segment is mapped back to the corresponding character segment in the original bid document text, and its source location in the original document, such as page number and chapter, is determined to facilitate the verification and tracing of the evaluation experts, thereby ensuring the accuracy and verifiability of information extraction.

[0045] Here, in the process of parallel information extraction and compliance verification of the structured review task definition based on the structured review task definition, the structured review task definition is seen as a single queue, and parallel information extraction and compliance verification based on the structured review task definition for multiple queues of the structured review task definition are required. However, the number of parallel threads is usually less than the number of documents in the structured review task definition, that is, multiple documents need to be processed by each thread, and the processing time of a single document is not balanced. Therefore, a balanced closed-loop driving mechanism with conditional branch load fallback is designed to realize low-latency and high-concurrency processing.

[0046] In one preferred embodiment, based on the structured review task definition, the parallel information extraction and compliance verification is performed on the set of bid document structured document data to obtain the extracted bid document data cube, including: setting the table header structure of the compliance check list and the key information comparison matrix included in the structured review task definition as a lock-free access external queue, and defining the set of bid document structured document data as a plurality of local queues corresponding to a plurality of threads, each local queue including one or more bid document structured document data, wherein the bid document structured document data in the local queue is set to be loaded into the lock-free access external queue when performing information extraction and compliance verification, and the structured review task definition in the lock-free access external queue is extracted by the thread corresponding to the local queue to perform information extraction and compliance verification.

[0047] In this embodiment, based on the structured review task definition, the parallel information extraction and compliance verification is performed on the set of bid document structured document data to obtain the extracted bid document data cube, further including: in response to the out-of-condition of the thread corresponding to the local queue being failed, triggering the thread to obtain bid document structured document data from the external queue; and in response to the out-of-condition of the thread corresponding to the local queue being successful, locking the thread, wherein the scheduling metadata set by the external queue monitors the locking state of the thread.

[0048] Specifically, first, for the structured review task definition, i.e. the table header structure of the compliance check list and the key information comparison matrix included therein, it is defined as a lock-free access external queue, and each set of bid document structured document data is defined as a plurality of local queues corresponding to a plurality of threads, whereby the local queue corresponding to each thread constitutes a first-in-first-out (FIFO) queue by including a plurality of document units, and an out-of-condition needs to be set for each thread as a conditional branch. Moreover, each document in the local queue is set to be loaded into the external queue first when performing the task, and the structured review task definition in the external queue is extracted by the thread corresponding to the local queue to perform information extraction and compliance verification.

[0049] Then, when the out-of-condition of a certain thread is failed, indicating that the current document unit of the thread has been completely processed, it can be triggered to obtain documents from other threads for processing, and based on the lock-free access feature of the external queue and the shared feature of the local queue, the thread obtains documents from other threads through the external queue. That is, as described above, since each document is loaded into the external queue first when performing the task, when an idle thread needs to obtain a document, it can obtain the document originally processed by other threads from the external queue, and then enter the local queue to be processed by the thread, thereby triggering a load backoff mechanism, so as to achieve load balancing without affecting other threads.

[0050] Then, in the case of triggering the load fallback mechanism, since the document in the external queue is taken away by other threads at this time, the thread to which the document originally belongs also needs to be notified. For the external queue, it needs to set the scheduling metadata to monitor each thread. Correspondingly, the local queue guarantees concurrent safety through thread locking when the thread is executing a task, that is, the dequeue condition is successful. Therefore, the scheduling metadata can monitor the thread idle state through the thread locking state. Moreover, since the processing of all documents is executed in the corresponding thread through the local queue, parallel control can be achieved only through the locking state of the thread corresponding to the local queue, that is, the dequeue condition-locking state constitutes an efficient closed loop of thread scheduling, thereby realizing efficient and balanced closed loop driving between threads.

[0051] Exemplarily, in step S5, data alignment, clarification and differential analysis are performed on the extracted bid document data cube to obtain evaluation archive data, which includes a final compliance report and a final KPI comparison matrix. It should be understood that although information extraction has been completed, these original extraction results may still need to be further standardized and compared for truly assisting experts in making fair and accurate judgments. Therefore, further data alignment, clarification and differential analysis are performed on the extracted bid document data cube to eliminate data noise, unify data formats, and present the advantages and disadvantages and compliance status of each bidder in an intuitive and contrasting manner, and finally form a comprehensive and traceable evaluation archive, greatly improving the efficiency, accuracy and transparency of bid evaluation decisions.

[0052] In one embodiment, the post-processing and analysis step of the bid document is a multi-stage process aimed at converting the data extracted from the original document into an evaluation archive directly usable by the evaluation experts. This process mainly includes three key links: data alignment, data cleaning and differential analysis.

[0053] Firstly, data alignment is performed, which aims to standardize and unify the extraction results of different bidders under the same evaluation dimension for horizontal comparison. All extraction results of bidders corresponding to the same compliance check item or the same KPI in the data cube are identified and logically or physically arranged together to form a structure convenient for quick comparison. For example, in the evaluation of registered capital and technical scheme advancement, the relevant information of A, B and C companies can be displayed side by side, so that the evaluation experts can intuitively see the performance differences of each company under the same standard.

[0054] Next is data cleaning, which aims to improve the quality and consistency of the extracted data, eliminating potential errors or redundant information. A series of cleaning operations will be performed on the extracted data, including but not limited to removing duplicate information, correcting recognition errors, and standardizing data formats. For example, various forms of representing amounts (such as "5000 million" and "five thousand million") will be unified into the standard format "5000 million". In addition, inaccurate information caused by OCR recognition errors or natural language processing model understanding bias will be corrected to ensure the consistency and standardization of all data, thereby enhancing the comparability of the data.

[0055] Finally, a differentiated analysis is conducted, and the final review archive data, i.e., the compliance report and KPI comparison matrix, is automatically generated based on the aligned and cleaned data. For the generation of the compliance report, a detailed report is automatically compiled based on the compliance check list and the results of each bidder in each compliance check. This report not only lists whether each bidder meets all the compliance requirements, but also specifically marks the non-compliance items and their corresponding original text fragments and source locations in the bid documents. For example, if A company passes all compliance items while B company's "qualification certificate" does not meet the requirements, the report will clearly indicate the specific details and sources of these problems, making it easy for review experts to verify.

[0056] At the same time, the final KPI comparison matrix is also generated, which is filled in the table cells according to the pre-defined key information comparison matrix table header structure and the extracted values of each bidder at the corresponding scoring points. This allows the bid evaluation experts to easily compare the specific performance of each bidder in key indicators such as "technical scheme advancement" through intuitive table forms. For example, in the "technical scheme advancement" column, A company's "AI plus blockchain" solution, B company's "big data analysis" solution, and C company's "traditional solution" are displayed, allowing the review experts to quickly grasp the technical characteristics and advantages of each bidder and assist in making more scientific and reasonable evaluation decisions.

[0057] Exemplarily, in step S6, the review archive data is displayed. It should be understood that although a large amount of data extraction, compliance verification, and comparison analysis work has been completed, the final bid evaluation decision still needs to be made by bid evaluation experts with professional knowledge and experience. Therefore, this step is a bridge connecting machine intelligence and human wisdom, which ensures that experts can efficiently review, verify, and utilize these intelligently generated auxiliary information to make fair, accurate, and traceable bid evaluation judgments. Without a clear and effective display mechanism, even if the previous processing is accurate, the value of its auxiliary decision-making is difficult to fully realize, and it may even increase the understanding burden of experts due to improper information presentation.

[0058] In one embodiment, the step of displaying the review archive data aims to achieve comprehensive and efficient presentation of the information required by the bid evaluation experts through user-friendly interface design and diversified visualization means. This step mainly covers two aspects of information display: one is the display of the final compliance report, and the other is the display of the final KPI comparison matrix. First, the final compliance report is presented to the bid evaluation experts in a clear and intuitive manner through a graphical user interface. Specifically, a special display area is provided on the operation interface of the bid evaluation system for displaying the judgment results of each bidder in each compliance check, including the status identification of “pass” or “fail”, and the results can be classified and sorted through lists, tables, highlights, etc. to enable the experts to quickly grasp the overall compliance of each bidder. To further enhance the transparency and traceability of the review process, the bid evaluation experts can jump directly or pop up the display of the original text fragments and their specific source locations (such as page X, chapter name, etc.) in the bid documents by clicking on the relevant items when viewing a compliance judgment result, thereby facilitating the experts to verify and review the judgment basis on site, and improving the accuracy and credibility of the review.

[0059] At the same time, the finally generated KPI comparison matrix is also displayed in a structured and visualized form to assist the bid evaluation experts in systematic comparison and analysis of the key indicators of each bidder, such as technology and business. Specifically, according to the preset key information comparison matrix table header structure, the extracted values of all bidders on each KPI are uniformly organized into a comparison table, which is displayed on the interface. To enhance the readability and analysis efficiency of the data, various chart forms are also supported for auxiliary display, such as column chart, radar chart, line chart, etc., so that the performance differences of different bidders in multiple dimensions such as technical scheme advancement, project team experience, and service response capability can be more intuitively identified and compared. This multi-dimensional comparison mechanism not only improves the visualization of the review work, but also provides strong data support for the experts in the scoring, grading, and final decision-making process.

[0060] In summary, the intelligent assisted bid evaluation method based on machine learning and cloud platform provided in the present application automatically parses project information and tender documents through a natural language processing module, constructs structured review task definitions such as compliance checklist and comparison matrix table header, and converts unstructured bid evaluation requirements into machine executable tasks. At the same time, the bid documents are parsed in parallel and structured into a unified data format. Based on the preset review rules, the massive bid data is efficiently extracted and compliance checked to form a multi-dimensional data cube. Then, the data cube is cleaned, aligned, and analyzed for differences to automatically generate visual compliance reports and KPI comparison matrices to assist bid evaluation experts in decision-making. In this way, by introducing machine learning and cloud platform technology, the bid evaluation process is automated, intelligentized, and standardized, significantly improving the bid evaluation quality and efficiency, and providing strong technical support for bidding.

[0061] The application also provides an intelligent auxiliary bid evaluation system based on machine learning and a cloud platform, which is used to execute the intelligent auxiliary bid evaluation system based on machine learning and a cloud platform as described above. Figure 6 As shown in the figure, the intelligent auxiliary bid evaluation system 600 based on machine learning and a cloud platform includes: a project information and tender document acquisition module 610, which is used to acquire project basic information and a tender document; an evaluation task structured definition module 620, which is used to process the tender document and the project basic information based on a natural language processing module to obtain a structured evaluation task definition, the structured evaluation task definition including a compliance check list and a table header structure of a key information comparison matrix; an intelligent bid document analysis module 630, which is used to acquire a set of bid documents of all bidders, and perform document analysis and content structuring on each bid document in the set of bid documents to obtain a set of bid document structured data; an intelligent evaluation analysis module 640, which is used to perform parallel information extraction and compliance verification on the set of bid document structured data based on the structured evaluation task definition to obtain extracted bid document data cubes; an evaluation result generation module 650, which is used to perform data alignment, clear and differentiated analysis on the extracted bid document data cubes to obtain evaluation archive data, the evaluation archive data including a final compliance report and a final KPI comparison matrix; and an evaluation result display module 660, which is used to display the evaluation archive data.

[0062] The basic principles of the application are described above in combination with specific embodiments, but it should be noted that the advantages, advantages, effects and the like mentioned in the application are only examples and are not limited, and these advantages, advantages, effects and the like cannot be considered as the must-have of each embodiment of the application. In addition, the above-mentioned specific details are only for the purpose of example and for the purpose of understanding, and are not limited to the above-mentioned specific details, which are not limited to the above-mentioned specific details.

[0063] The computer readable storage medium provided by the embodiments of the application stores computer program codes, and when the computer program codes are run on a computer, the computer executes the related method steps to realize the intelligent auxiliary bid evaluation method based on machine learning and a cloud platform provided by the above-mentioned embodiments.

[0064] The computer program product provided by the embodiments of the application, when the computer program product is run on a computer, makes the computer execute the related steps to realize the intelligent auxiliary bid evaluation method based on machine learning and a cloud platform provided by the above-mentioned embodiments.

[0065] The system, the computer readable storage medium or the computer program product provided by the embodiments of the present application are used to execute the corresponding method provided above, thus the beneficial effects achieved thereby can refer to the beneficial effects of the corresponding method provided above, which will not be described here again.

[0066] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments.

[0067] The processes depicted in the figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some implementations, multitasking and parallel processing can be advantageous. The various embodiments described herein are described with progressive specificity. Like reference numerals can refer to like elements throughout the description.

Claims

1. An intelligent auxiliary bid evaluation method based on machine learning and a cloud platform, characterized in that, The method comprises: obtaining project basic information and a tender document; processing the tender document and the project basic information based on a natural language processing module to obtain a structured review task definition, the structured review task definition comprising a compliance check list and a table header structure of a key information comparison matrix; obtaining a set of bid documents of all bidders, and performing document parsing and content structuring on each bid document in the set of bid documents to obtain a set of bid document structured document data; based on the structured review task definition, performing parallel information extraction and compliance verification on the set of bid document structured document data to obtain an extracted bid document data cube; performing data alignment, clarification and differential analysis on the extracted bid document data cube to obtain review archive data, the review archive data comprising a final compliance report and a final KPI comparison matrix; displaying the review archive data. 2.The intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to claim 1, wherein, The method comprises: performing document parsing on the tender document to obtain tender document text data; performing rule mining on the tender document text data to obtain a candidate compliance check list; performing scoring point mining on the tender document text data to obtain a candidate KPI matrix table header list; performing expert human interaction review and confirmation on the candidate compliance check list and the candidate KPI matrix table header list to obtain the compliance check list and the table header structure of the key information comparison matrix. 3.The intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to claim 2, characterized in that, The method comprises: 4.The intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to claim 2, characterized in that, performing rule-based pattern matching on the tender document text data to obtain the candidate compliance check list. The method comprises: locating a scoring table section in the tender document text data through keyword matching; performing structured information extraction on the scoring table section to obtain a set of scoring items; 5.The intelligent evaluation method based on machine learning and cloud platform according to claim 1, wherein, performing named entity recognition on each scoring item in the set of scoring items to obtain the candidate KPI matrix table header list. The method comprises: extracting a first compliance check from the compliance check list; extracting first bid document structured document data from the set of bid document structured document data; extracting a check keyword from the first compliance check, and performing keyword search on the first bid document structured document data with the check keyword as a query condition to obtain a search result; in response to the search result being a hit, writing an original text segment and a source location in the first bid document structured document data into the extracted bid document data cube. 6.The intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to claim 5, characterized in that, based on the structured review task definition, performing parallel information extraction and compliance verification on the set of bid document structured document data to obtain an extracted bid document data cube, comprising: extracting first score points from the table header structure of the key information comparison matrix; generating first score point generation questions based on the first score points; inputting the first score point generation questions and the first bid document structured document data into a QA model to obtain answer text segments; extracting extracted values, original text segments, source locations and confidence from the answer text segments and writing them into the extracted bid document data cube. 7.The intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to claim 5, wherein, based on the structured review task definition, performing parallel information extraction and compliance verification on the set of bid document structured document data to obtain an extracted bid document data cube, comprising: The compliance check list and the table header structure of the key information comparison matrix included in the structured review task definition are set as a lock-free access external queue, and a set of bid document structured document data is defined as a plurality of local queues corresponding to a plurality of threads, each local queue including one or more bid document structured document data, wherein the bid document structured document data in the local queue is set to be loaded into the lock-free access external queue when performing information extraction and compliance verification, and the structured review task definition in the lock-free access external queue is extracted and verified by the thread corresponding to the local queue. 8.The intelligent auxiliary bid evaluation method based on machine learning and cloud platform according to claim 7, characterized in that, based on the structured review task definition, performing parallel information extraction and compliance verification on the set of bid document structured document data to obtain an extracted bid document data cube, further comprising: in response to the de-queue condition of the thread corresponding to the local queue being failed, triggering the thread to obtain bid document structured document data from the external queue; in response to the de-queue condition of the thread corresponding to the local queue being successful, locking the thread, wherein the scheduling metadata set by the external queue monitors the locking state of the thread.

9. An intelligent auxiliary bid evaluation system based on machine learning and cloud platform, characterized in that, comprising: a project information and tender document acquisition module for acquiring project basic information and tender documents; a review task structured definition module for processing the tender documents and project basic information based on a natural language processing module to obtain a structured review task definition, the structured review task definition including a compliance check list and a table header structure of a key information comparison matrix; a bid document intelligent parsing module for obtaining a set of bid documents of all bidders, and performing document parsing and content structuring on each bid document in the set of bid documents to obtain a set of bid document structured document data; an intelligent review analysis module for performing parallel information extraction and compliance verification on the set of bid document structured document data based on the structured review task definition to obtain an extracted bid document data cube; a review result generation module for performing data alignment, clarity and differential analysis on the extracted bid document data cube to obtain a review archive data, the review archive data including a final compliance report and a final KPI comparison matrix; a review result display module configured to display the review archive data.

Images