Bidding file auxiliary generation method and system based on AI robot

Abstract

The invention provides a bidding file auxiliary generation method and system based on an AI robot, and relates to the technical field of artificial intelligence. According to the method, the natural language processing algorithm is used for intelligently analyzing the bid invitation file, key information such as qualification requirements and technical parameters is extracted, manual omission or misunderstanding is avoided, and the accuracy and efficiency of information processing are improved. Then, based on a semantic matching algorithm and a bidding file knowledge base, a bidding file first draft conforming to the bidding requirement is automatically retrieved and filled, the time cost for manually looking up data and organizing content is reduced, and the compilation efficiency is improved. And finally, through systematic compliance, consistency and integrity review, targeted optimization is carried out on the bidding document based on a review result, and the risk of waste bidding caused by human negligence is effectively avoided. According to the method, AI driving of bidding document compiling is realized, the problems of low efficiency, high error rate and high bidding waste risk in manual bidding document compiling are solved, and the bidding document compiling efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for assisting in the generation of tender documents based on an AI robot. Background Technology

[0002] In the bidding and procurement process of the power industry, the preparation of bid documents is a highly complex, specialized, and time-sensitive task. Traditionally, bid document writing relies on manual completion, but key clauses, technical parameters, and scoring criteria in the bidding documents are scattered, and manual reading can easily lead to the omission or misunderstanding of important information.

[0003] Furthermore, bidders need to repeatedly review historical data, industry standards, and company qualifications, manually organizing the content, which is time-consuming and prone to inconsistencies in format, repetition, or omissions, resulting in low efficiency in preparing bid documents. Moreover, if the bid documents do not fully comply with the mandatory clauses of the tender documents, or contain inconsistencies in data, such as discrepancies between itemized prices and the total price, there is a risk of bid rejection. Summary of the Invention

[0004] This invention provides a method and system for generating bid documents based on AI robots, which solves the problems of low efficiency, error-proneness, and high risk of rejection in manual bid document preparation, and improves the efficiency of bid document preparation.

[0005] In a first aspect, the present invention provides a method for assisting in the generation of tender documents based on an AI robot. The method includes: receiving tender documents uploaded by a user; using a natural language processing algorithm of the AI ​​robot to perform structured parsing of the tender documents to obtain key information, including qualification requirements, technical parameters, scoring criteria, and mandatory clauses; based on the key information, using a semantic matching algorithm of the AI ​​robot and a tender document knowledge base to perform retrieval, matching, and automatic filling to obtain a draft tender document corresponding to the key information; the tender document knowledge base stores historical winning bids, standard technical templates, and enterprise qualification data; reviewing the draft tender document for compliance, consistency, and completeness to obtain review results; and based on the review results, optimizing the draft tender document and outputting the optimized tender document.

[0006] Secondly, embodiments of the present invention provide an AI-based robot-assisted tender document generation device, comprising: a communication module and a processing module; the communication module receiving tender documents uploaded by a user; and the processing module using the AI ​​robot's natural language processing algorithm to perform structured parsing of the tender documents to obtain key information, including qualification requirements, technical parameters, scoring criteria, and mandatory clauses; based on the key information, using the AI ​​robot's semantic matching algorithm and a tender document knowledge base to perform retrieval, matching, and automatic filling to obtain a draft tender document corresponding to the key information; the tender document knowledge base storing historical winning bids, standard technical templates, and enterprise qualification data; reviewing the draft tender document for compliance, consistency, and completeness to obtain review results; and based on the review results, optimizing the draft tender document and outputting the optimized tender document.

[0007] Thirdly, embodiments of the present invention provide an electronic device including a memory and a processor. The memory stores a computer program, and the processor is configured to call and run the computer program stored in the memory to perform the steps of the method as described in the first aspect and any possible implementation thereof.

[0008] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps of the method as described in the first aspect and any possible implementation thereof.

[0009] This invention provides an AI-driven method and system for generating tender documents. It utilizes natural language processing algorithms to intelligently parse tender documents, extracting key information such as qualification requirements and technical parameters, avoiding human omissions or misunderstandings and improving the accuracy and efficiency of information processing. Subsequently, based on semantic matching algorithms and a tender document knowledge base, it automatically retrieves and populates draft tender documents that match the tender requirements, reducing the time cost of manual document review and content organization, and improving compilation efficiency. Finally, through a systematic review of compliance, consistency, and completeness, and based on the review results, the tender documents are optimized in a targeted manner, effectively mitigating the risk of rejection due to human negligence. This invention achieves AI-driven tender document compilation, solving the problems of low efficiency, error-proneness, and high risk of rejection associated with manual tender document compilation, thus improving the efficiency of tender document preparation. Attached Figure Description

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

[0011] Figure 1 This is a flowchart illustrating a method for assisting in the generation of tender documents based on an AI robot, as provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a tender document generation device based on an AI robot provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0012] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0013] In the description of this invention, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" and "more than one" refer to two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0014] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner to facilitate understanding.

[0015] Furthermore, the terms "comprising" and "having," and any variations thereof, used in the description of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not limited to the steps or modules listed, but may optionally include other steps or modules not listed, or may optionally include other steps or modules inherent to such process, method, product, or device.

[0016] To make the objectives, technical solutions, and advantages of the present invention clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0017] like Figure 1 As shown, this embodiment of the invention provides a method for assisting in the generation of tender documents based on an AI robot. The method includes steps S101-S105.

[0018] S101. Receive the tender documents uploaded by the user.

[0019] For example, users can upload tender documents via the web or mobile device, in formats such as PDF, DOCX, and TXT.

[0020] In some embodiments, the tender documents are a collection of normative and legal documents prepared by the tendering party or its entrusted tendering agent in accordance with relevant laws and regulations and project requirements, and are intended to clearly inform potential bidders of the procurement subject matter, bidding procedures, technical requirements, evaluation criteria and contract terms.

[0021] S102. Use the natural language processing algorithm of AI robot to perform structured parsing of the tender documents to obtain key information.

[0022] In this application embodiment, key information includes qualification requirements, technical parameters, scoring criteria, and mandatory clauses.

[0023] For example, key information refers to the set of core elements extracted from the tender documents that are crucial to the preparation of the bid. These include: Qualification requirements: The qualifications that bidders must possess, such as registered capital, industry certifications, and past performance. Technical parameters: The specific technical specifications, performance indicators, and functional requirements for the product or service specified by the tendering party. Evaluation criteria: The detailed rules used to quantitatively evaluate the quality of bids during the evaluation process, including the weight of each indicator and the basis for scoring. Mandatory clauses: A rigid requirement that bids must unconditionally respond to the requirements; otherwise, they will be considered invalid.

[0024] In some embodiments, the AI ​​robot is equipped with algorithms such as natural language processing, machine learning, and knowledge graphs to simulate the cognitive and decision-making processes of human experts, thereby automating and intelligentizing the processing of tender documents.

[0025] In some embodiments, natural language processing algorithms refer to algorithmic models that enable computers to understand, interpret, and manipulate human language. This includes methods for parsing unstructured tender document text, such as named entity recognition, dependency parsing, semantic role labeling, text classification, and relation extraction. It transforms unstructured text into structured data that machines can understand and process.

[0026] In some embodiments, structured parsing refers to the process of identifying predefined, meaningful components from unstructured tender document text and extracting and summarizing them into a unified and standardized format.

[0027] As one possible implementation, step S102 can be specifically implemented as steps S1021-S1025.

[0028] S1021. A named entity recognition model based on AI robots is used to traverse the tender documents and identify the key entities that represent the tender requirements.

[0029] In some embodiments, key entities include qualification categories, technical parameter indicators, scoring items, and mandatory clause identifiers.

[0030] For example, key entities are specific objects identified from the tender document text using a named entity recognition model. Examples include: qualification categories such as ISO9001 quality system certification and safety production license; technical parameters such as voltage level and response time; scoring items such as technical score and price score; and mandatory clause identifiers such as "must be prohibited."

[0031] For example, embodiments of the present invention can use a pre-trained multi-task NER model (such as the BERT-CRF joint model) to identify key entity types in tender documents.

[0032] S1022. Based on the tender documents, use dependency parsing and relation extraction techniques to construct semantic relationships between key entities.

[0033] In some embodiments, dependency parsing is used to analyze the grammatical dependencies (such as subject-verb, verb-object, and attributive-head) between words in a sentence to understand the sentence structure. Relation extraction techniques, based on syntactic analysis, identify and extract semantic relationships (such as possess, equal to, and require) between entities. The combination of these two techniques is used to establish semantic associations between key entities; for example, to determine the specific numerical unit of a technical parameter or the specific assessment content corresponding to a scoring item.

[0034] In some embodiments, semantic association refers to the logical semantic connection between key entities. For example, there is a value relationship between the technical parameter indicator entity voltage level and the numerical entity 220kV; there is a weight relationship between the scoring item entity project experience and the weight entity 20 points.

[0035] For example, embodiments of the present invention can perform dependency parsing to extract subject-verb-object structures and modification relations; use a relation extraction model to identify relationships between entities, such as technical parameters-numerical units, scoring items-weights; construct an entity relationship graph, and use GraphSAGE for graph embedding representation.

[0036] S1023. Based on the semantic relationships between key entities and the preset structured template, integrate the key entities to construct a set of key information.

[0037] In some embodiments, a preset structured template is used to organize and store key information, specifying the arrangement and storage of different categories of key information and their relationships, such as a JSON template containing fields such as qualification requirements, technical parameters, scoring criteria, and mandatory clauses.

[0038] S1024. Based on the key information set and the word vector model, the technical terms of each key entity are expanded with synonyms and near-synonyms to obtain the expanded key information set.

[0039] In some embodiments, the word vector model is a natural language processing technique that maps words or phrases to real-valued vectors. Words with similar semantics are also located close to each other in the vector space. Synonyms and near-synonyms are used to expand technical terms to improve the recall rate of subsequent semantic matching and avoid matching failures due to different terminology.

[0040] For example, embodiments of the present invention may use a word vector model to semantically expand technical terms; for example, voltage level may be expanded to rated voltage operating voltage; the expanded term set is used to enhance the recall rate of retrieval matching.

[0041] S1025. Based on the expanded set of key information, a contextual semantic understanding model is used to verify the semantic integrity and logical consistency of the set of key information to obtain the key information.

[0042] In some embodiments, the context semantic understanding model is a deep learning model capable of understanding the overall meaning of a text fragment in a specific context. It is used to validate the integrated set of key information to determine whether its semantics are complete (e.g., whether key constraints are missing) and whether its logic is consistent (e.g., whether there are contradictory requirements).

[0043] For example, embodiments of the present invention may employ a pre-trained semantic understanding model to perform contextual encoding on the key information set; use a logical consistency detection module to determine whether there are contradictory descriptions; and output the verified key information set to ensure its semantic integrity and logical consistency.

[0044] Thus, this invention achieves high-precision parsing of bidding documents and intelligent generation of bid documents, significantly improving the quality of bid documents and the probability of winning the bid. Simultaneously, through automated review and optimization, it reduces manual intervention and improves the efficiency and fairness of bidding and procurement.

[0045] S103. Based on key information, the semantic matching algorithm of the AI ​​robot and the knowledge base of the tender document are used to perform retrieval, matching and automatic filling to obtain the initial draft of the tender document corresponding to the key information.

[0046] In this embodiment of the application, the tender document knowledge base stores historical winning bids, standard technical templates, and enterprise qualification data.

[0047] For example, a tender document knowledge base is a structured or semi-structured database built to assist in the generation of tender documents. The tender document knowledge base stores reusable, tender-related knowledge assets. These include: Historical winning bids: Excellent technical and service solutions from previously successful tender documents; Standard technical templates: Pre-written, standardized document fragments or templates for common technical requirements; and Enterprise qualification data: Fixed information such as the enterprise's business license, qualification certificates, patent certificates, and typical cases.

[0048] In some embodiments, semantic matching algorithms are used to calculate the semantic similarity between two text segments. Semantic matching algorithms are capable of understanding the deeper meanings of words, phrases, and sentences. Similarity calculations are performed in vector space between key information and content in the tender document knowledge base to find the most semantically relevant material.

[0049] In some embodiments, the initial draft of the tender document refers to the preliminary version of the tender document generated after an AI robot retrieves, matches, and automatically fills in the knowledge base content.

[0050] As one possible implementation, step S103 can be specifically implemented as steps S1031-S1035.

[0051] S1031. Generate a query vector based on key information.

[0052] For example, embodiments of the present invention can extract the content of each field in a key information set and concatenate them into one or more coherent query text segments. For instance, all technical parameters and their numerical units can be concatenated into one text segment, and all qualification requirements into another. Text cleaning is then performed, such as removing stop words and standardizing capitalization. A pre-trained semantic text vectorization model is used to convert the above query text segments into dense vectors of fixed dimensions (i.e., query vectors). For example, a Sentence-BERT model based on the Transformer architecture can be used. Optimization is performed on sentence-level and paragraph-level semantic representations, mapping semantically similar texts to positions close to each other in the vector space.

[0053] S1032. Vectorize the historical winning bids, standard technical templates, and enterprise qualification data in the bid document knowledge base to obtain a vector database.

[0054] For example, embodiments of the present invention can perform text extraction and normalization processing on various materials stored in the knowledge base (historical winning bid documents, standard technical template texts, enterprise qualification data descriptions, etc.). For instance, from a historical winning bid, its technical solution summary and qualification response description can be extracted as independent text units. Using the Sentence-BERT model, all the preprocessed text units are converted into vectors. A mapping relationship is established between the generated vectors and their corresponding original material content (such as text fragments, template IDs, file paths).

[0055] S1033. Based on the query vector, a semantic matching algorithm is used to calculate the semantic similarity between the query vector and the material vectors of various types in the vector database.

[0056] For example, embodiments of the present invention can calculate the similarity score between the query vector and all candidate vectors in the knowledge base. For instance, cosine similarity can be used as a metric to effectively measure the difference in direction between vectors. The value of cosine similarity ranges from [-1, 1], and the closer the value is to 1, the more semantically similar the two vectors are.

[0057] S1034. Based on semantic similarity, sort and filter the vectors of various types of materials in the vector database to obtain a set of recommended materials that match the key information.

[0058] For example, in embodiments of the present invention, all candidate materials can be sorted from high to low based on the calculated cosine similarity score. A similarity threshold (e.g., 0.75) is set. Materials with similarity scores higher than this threshold are selected, and a recommended material set is formed based on their similarity scores. For content of the same type (such as a technical description of a certain technical parameter), the top N most relevant materials can be retained for selection.

[0059] S1035. Based on preset mapping rules, automatically fill the materials in the recommended material set into the bid document template to generate the initial draft of the bid document.

[0060] For example, preset mapping rules define the mapping relationship between key information types and the sections of the tender document template. For example: Rule 1: IF (Information type == "Technical parameters") THEN (fill to the "Technical solution" section); Rule 2: IF (Information type == "Qualification requirements") THEN (fill to the "Business credit" section); Rule 3: IF (Information type == "Service requirements" in "Evaluation criteria") THEN (fill to the "Service commitment" section).

[0061] For example, embodiments of the present invention can determine the target location by invoking corresponding mapping rules based on the type of key information. The content ranked first or best in the recommended material set is automatically inserted into the target location. For cases requiring the integration of multiple materials, they can be spliced ​​and merged according to preset logic to ensure smooth writing. Finally, a draft tender document with rich content and a complete structure is output.

[0062] Thus, embodiments of the present invention can achieve accurate, efficient, and automated mapping from tender document requirements to bid document content through query vector generation → vector database retrieval → semantic similarity matching → rule mapping and filling. This not only overcomes the limitations of traditional keyword matching and understands the deep semantics of user needs, but also ensures retrieval speed and accuracy in a large-scale knowledge base through vector database technology, significantly improving the quality and efficiency of bid document preparation.

[0063] S104. Conduct a compliance, consistency and completeness review of the initial draft of the tender documents and obtain the review results.

[0064] In some embodiments, compliance, consistency, and completeness review refers to the automated quality inspection process performed by an AI robot on the initial draft of the tender document. Compliance: Checking whether the initial draft fully responds to all mandatory clauses in the tender document. Consistency: Checking whether there are logical inconsistencies or data conflicts within the initial draft, such as whether the description of the technical solution conforms to the technical parameters, or whether the sum of the itemized prices equals the total price. Completeness: Checking whether the initial draft contains all the required sections, appendices, and content elements required by the tender document.

[0065] As one possible implementation, step S104 can be specifically implemented as steps S1041-S1046.

[0066] S1041. Based on the initial draft of the tender document and the mandatory clauses of key information, perform responsiveness calculation to obtain the response results of the initial draft of the tender document to each mandatory clause.

[0067] In some embodiments, the response result includes a response and no response.

[0068] For example, for each mandatory clause, determine whether the requirements of the mandatory clause are semantically implied (i.e., satisfied) in the full text of the initial draft tender document. Output an implied probability score (e.g., between 0 and 1). Set a threshold (e.g., 0.9). If the probability score is ≥ the threshold, it is considered a response. If the probability score is < the threshold, or the model does not find any relevant text, it is considered a no-response. Output a list of response results, containing each mandatory clause and its corresponding response result.

[0069] S1042. Based on the response results of the initial draft of the tender documents to each mandatory clause, determine the compliance review results of the tender documents.

[0070] For example, embodiments of the present invention can iterate through the response result list. If any mandatory clause receives a non-response, the overall compliance review result is marked as failed, and all non-response clauses are recorded. If all mandatory clauses receive a response, the compliance review result is marked as passed.

[0071] S1043. Based on the initial draft of the tender document, conduct logical reasoning and data correlation analysis to determine the consistency review results.

[0072] In some embodiments, the consistency review results include the consistency between the technical solution description and technical parameters, the consistency between the itemized quotations and the total price, and the consistency of the description of the same matter in different chapters.

[0073] For example, ensuring consistency between the technical solution description and technical parameters involves using a rule engine and information extraction. First, descriptive performance commitments (e.g., high system reliability) are extracted from the initial draft's technical solution section, while specific quantitative parameters (e.g., mean time between failures: >10,000 hours) are extracted from the technical parameter table section. The two are then checked for consistency; for instance, a rule might stipulate that if high reliability is mentioned, a corresponding mean time between failures parameter must be provided, and its value must be greater than a certain standard value.

[0074] For example, to ensure consistency between itemized quotes and the total price: data extraction and calculation verification are used. Using OCR or parsing a DOCX table, all sub-item amounts in the initial draft's itemized quote table are extracted and summed. The calculated sum is compared with the value entered in the bid total price field. If the absolute error exceeds the allowable range (e.g., ±1 yuan), it is considered inconsistent.

[0075] For example, consistency in the description of the same matter across different chapters can be achieved through coreference resolution and contradiction detection. For instance, an NLP model can be used to identify that the total number of employees (100) mentioned in the company introduction chapter and the 200 engineers mentioned in the project implementation team chapter may refer to the same entity.

[0076] In some embodiments, the conformity review results also include competitive conformity.

[0077] Accordingly, step S1043 can be specifically implemented as steps A1-A5.

[0078] A1. Extract the technical configuration parameters, service commitment indicators, and bid price from the initial draft of the bid documents.

[0079] For example, technical configuration parameter extraction: Using named entity recognition and table parsing technology, specific parameter values ​​are extracted from the technical solution and technical parameter table sections. For example: {"Processor frequency": "2.5GHz", "Memory capacity": "64GB", "Hard disk type": "SSD", "Concurrent users": ">1000"}. Service commitment indicator extraction: Quantitative indicators are extracted from the service solution and service commitment sections. For example: {"Response time": "≤4 hours", "On-site arrival time": "≤8 hours", "Fault resolution rate": "≥99%", "Warranty period": "5 years"}. Bid price extraction: The total bid price or the sum of itemized bid prices is accurately extracted through OCR or document parsing.

[0080] A2. Based on technical configuration parameters, service commitment indicators, and bid prices, construct a solution-price feature vector.

[0081] For example, embodiments of the present invention can convert parameters and indicators described in text into numerical values. For instance, ≥99% is converted to 0.99, and 5 years is converted to 5. Features with different dimensions are normalized to bring them to the same scale. For example, Min-Max normalization is used to scale features such as price, time, and percentage to the [0, 1] range. The processed technical parameters, service indicators, and bid prices (after normalization) are then concatenated into a unified, fixed-length scheme-bid feature vector.

[0082] A3. Input the scheme-quotation feature vector into the pre-trained competitiveness evaluation model to obtain the probability of winning the bid.

[0083] In some embodiments, the competitiveness assessment model is trained and generated based on historical winning bid datasets to quantify the probability of winning a bid for a scheme-bid combination.

[0084] An example of the construction and training of a competitiveness assessment model: Training Data: The historical bidding dataset forms the basis for model training. This dataset contains a large amount of bidding data from historical tender projects. Each data point includes: Feature Vector: The feature vector of each bidder's proposal-price offer at that time. Label: Whether the bidder won the bid (1 indicates winning, 0 indicates not winning). Model Selection and Training: This invention preferably uses a gradient boosting decision tree model, such as XGBoost or LightGBM. This type of model can effectively handle structured features and capture the complex nonlinear relationship between features and bidding results. The training process involves learning patterns in historical data through algorithms to establish a mapping function from the proposal-price offer feature vector to the probability of winning the bid.

[0085] For example, in this embodiment of the invention, the scheme-price feature vector of the current bid document can be input into the competitiveness evaluation model. The output is a value between 0 and 1, which is the predicted probability of winning the bid. This probability value quantifies the likelihood of this scheme-price combination winning under the current historical data pattern.

[0086] A4. When the probability of winning the bid is lower than the preset competitiveness threshold, it is determined that there is a problem of inconsistent competitiveness.

[0087] For example, embodiments of the present invention can set a preset competitiveness threshold (e.g., 0.3 or 0.4, which can be adjusted based on historical data and company strategy). This threshold represents the minimum acceptable level of competition. The predicted probability of winning the bid is compared with this threshold. If the probability of winning the bid is less than the preset competitiveness threshold, it is determined that there is a competitiveness inconsistency problem in the bid documents. This indicates that although there may be no logical errors within the documents, the combination of its proposal and price lacks competitiveness in the market and is inconsistent with the ultimate goal of winning the bid.

[0088] A5. If there is an inconsistency in competitiveness, then based on the output of the competitiveness assessment model, identify the characteristic dimensions that lead to insufficient competitiveness and generate optimization suggestions.

[0089] In some embodiments, optimization suggestions include technical solution adjustments, service commitment optimizations, and / or price revisions.

[0090] For example, embodiments of the present invention can utilize the feature importance analysis function built into a competitiveness assessment model (such as XGBoost) to identify the negative feature dimensions that contribute most to the low probability of winning the bid. For instance, the model might output that excessively high price is the primary factor, followed by an excessively short warranty period.

[0091] For example, technical solution adjustments: If technical parameters are a major negative factor, a possible suggestion is to upgrade the hard drive type from HDD to SSD. This change is expected to increase the probability of winning the bid by approximately 15%. Service commitment optimization: If service commitment is a major negative factor, a possible suggestion is to increase the current '3-year warranty period' to '5 years', which is below the market average. This change is expected to increase the probability of winning the bid by approximately 10%. Price correction: If price is a major negative factor, a possible suggestion is to lower the current price by 5%-8% to enter the most competitive price range, as the current price is higher than the 90th percentile of historical winning bid prices.

[0092] Thus, embodiments of the present invention can elevate the consistency review of tender documents from the level of internal logical consistency to the level of external market competitiveness consistency. This involves quantitative evaluation within the context of historical market data. Before submitting tender documents, the present invention pre-assesses the probability of winning the bid, achieving proactive early warning. It identifies the specific reasons for insufficient competitiveness, achieving precise diagnosis. It provides data-driven, quantifiable optimization directions to assist decision-makers in making precise adjustments, thereby significantly improving the success rate through intelligent recommendations.

[0093] S1044. Based on the pre-set list of essential elements of the tender documents, review the chapters, attachments, and key contents of the initial draft of the tender documents to obtain the results of the completeness review.

[0094] For example, embodiments of the present invention can analyze the table of contents and attachment list of the initial draft of the tender document, compare it with the list, and check for any omissions. Keyword extraction and text positioning techniques are used to search for essential key content in the relevant chapters. For example, keywords such as "total tender price" and "RMB" can be searched in the tender letter chapter. The completeness review results are output, listing all missing chapters, attachments, or key content.

[0095] S1045. Based on the results of the compliance review, consistency review, and integrity review, summarize the issues to obtain a list of issues.

[0096] In some embodiments, the issue list includes issue type and issue location.

[0097] For example, embodiments of the present invention can aggregate the three review results—compliance review, consistency review, and integrity review—to generate a structured list of issues, typically in list form. Each issue entry should include at least: Issue type: e.g., Compliance—failure to respond to mandatory clauses, Consistency—discrepancy between itemized pricing and total price, Integrity—missing business license attachment. Issue location: pointing to the specific chapter, page number, or paragraph in the initial draft of the tender document. Issue description: a detailed explanation of the issue, e.g., failure to respond to mandatory clauses 'must provide a 3-year warranty'. Severity level: e.g., fatal, error, warning.

[0098] S1046. Based on the compliance review results, consistency review results, and integrity review results, as well as the issue list, generate the review results.

[0099] For example, embodiments of the present invention can integrate all outputs into a comprehensive review result. The review result includes: an overall conclusion (e.g., review fails due to fatal errors); itemized conclusions (including pass / fail status for compliance, consistency, and completeness); a detailed list of issues; and a review summary (a statistical overview of the issues).

[0100] Thus, this invention achieves in-depth quality inspection of draft bid documents through a multi-dimensional, automated review process. It simulates the three core elements that human experts focus on during bid evaluation: compliance, consistency, and completeness, resulting in higher efficiency, accuracy, and objectivity. It effectively identifies logical contradictions and data errors that are easily overlooked during manual review, significantly improving the quality of bid documents from the source and reducing the risk of bid rejection.

[0101] S105. Based on the review results, optimize the initial draft of the tender document and output the optimized tender document.

[0102] As one possible implementation, step S105 can be specifically implemented as steps S1051-S1055.

[0103] S1051. Based on the compliance review results and mandatory clauses, retrieve and match compliant content from the tender document knowledge base using a semantic matching algorithm.

[0104] For example, when the compliance review results in failure and the issue list includes unresponsive mandatory clauses, each unresponsive mandatory clause is treated as a separate query text. Using semantic matching algorithms and a vector database, the content fragment that best matches the semantics of that mandatory clause is retrieved from the tender document knowledge base. For instance, if the unresponsive clause requires 24 / 7 technical support, the detailed service description paragraphs regarding 24 / 7 technical support are retrieved from the knowledge base.

[0105] S1052. Based on the re-search and matching of compliant content, replace the non-compliant content in the initial draft of the tender document.

[0106] For example, embodiments of the present invention can accurately locate missing or improperly responded paragraphs in the initial draft based on the problem locations recorded in the problem list. If content is missing at a problem location, the optimal compliant content is directly inserted into that location. If existing but unqualified content exists at a problem location, the system will use a differential update algorithm to replace the unqualified part with compliant content, while retaining other valid contextual information.

[0107] S1053. Based on the results of the consistency review, the baseline data in the initial draft of the tender document is determined through a logical reasoning model.

[0108] For example, embodiments of the present invention may include a built-in priority rule base to determine benchmark data through a logical reasoning model. For instance, for technical parameters, the quantitative values ​​in the technical parameter table take precedence over the descriptive text in the technical solution. Therefore, when there is a conflict, the data in the parameter table takes precedence. For price data, the summation result of the itemized quotation table serves as the benchmark for the total price. Therefore, when the summation result differs from the entered total price, the summation result takes precedence. For service commitments, the indicators in the service commitment overview table take precedence over the descriptions in the main text of the chapter.

[0109] S1054. Based on the benchmark data in the initial draft of the tender document, correct any inconsistencies in the initial draft of the tender document.

[0110] For example, technical parameter correction: Based on the benchmark data in the technical parameter table, automatically correct descriptive text in the technical solution section that contradicts it. For instance, automatically correct the description of 16GB of memory in the solution to 64GB of memory, consistent with the parameter table. Price correction: Automatically use the calculated sum of itemized prices to cover the initially entered inconsistent bid price. Text contradiction correction: For contradictions between sections, the system will highlight the descriptions with lower priority based on the benchmark data and prompt the user to manually confirm the best correction method, or automatically replace them according to a predefined template.

[0111] S1055. Based on the results of the integrity review and the tender document knowledge base, determine the supplementary content.

[0112] For example, embodiments of the present invention can retrieve corresponding content from the enterprise qualification data and standard technical template area of ​​the tender document knowledge base based on the missing items indicated in the integrity review results (such as missing business license attachments or missing project team composition tables). For missing qualification documents, the electronic version of the document already stored in the knowledge base is directly linked. For missing text sections or tables, the corresponding standard paragraph or table template is retrieved from the standard template library.

[0113] S1056. Based on the supplementary content, the missing items in the initial draft of the tender document are completed to obtain the optimized tender document.

[0114] For example, embodiments of the present invention can precisely insert the determined supplementary content into the missing positions according to the template structure of the tender document. For instance, a copy of the business license can be inserted into the business attachments section, and a project team table can be inserted into the human resources section of the technical solution. After all replacement, correction, and supplementation operations are completed, the system calls the document rendering engine to regenerate the optimized structured data into a final document with a standardized format and complete content—the optimized tender document. This document also includes an optimization report detailing all automatically corrected issues.

[0115] Thus, this invention, through an automated optimization chain of retrieval-replacement-benchmarking-correction-supplementation, enables the self-correction capability of the initial draft of the tender document. It proactively and accurately resolves problems, generating a final draft of the tender document that meets high standards in compliance, consistency, and completeness. This significantly reduces the workload of manual revisions and subjective errors, thereby comprehensively improving the efficiency and quality of the bidding process.

[0116] This invention provides an AI-assisted method for generating tender documents. It utilizes natural language processing algorithms to intelligently parse tender documents, extracting key information such as qualification requirements and technical parameters, avoiding human omissions or misunderstandings and improving the accuracy and efficiency of information processing. Subsequently, based on semantic matching algorithms and a tender document knowledge base, it automatically retrieves and populates draft tender documents that match the tender requirements, reducing the time cost of manual document review and content organization, and improving compilation efficiency. Finally, through a systematic review of compliance, consistency, and completeness, and based on the review results, the tender documents are optimized in a targeted manner, effectively mitigating the risk of rejection due to human negligence. This invention achieves AI-driven tender document compilation, solving the problems of low efficiency, error-proneness, and high risk of rejection associated with manual tender document compilation, and improving the efficiency of tender document preparation.

[0117] Optionally, the AI ​​robot-based bid document generation method provided in this embodiment of the invention further includes steps S201-S204 in step S105.

[0118] S201. Receive the final version of the winning bid document uploaded by the user.

[0119] In some embodiments, the final version of the winning bid document includes the winning bid proposal and the enterprise qualification data corresponding to the winning bid proposal.

[0120] For example, the final version of the bid award document typically refers to the official bid award announcement issued by the tendering party, the public announcement document of the winning bidder, or the finally signed transaction contract. It includes the most valuable bid proposal and its corresponding qualification data of the winning company.

[0121] S202. Use the natural language processing algorithm of AI robot to perform structured parsing of the final bid-winning document to obtain the winning technical solution, the winning technical template and the qualification data of the winning enterprise.

[0122] For example, the winning technical solution: From the technical bid section of the winning bid document, the core ideas, technical routes, innovative points, and key parameters of the successful solution are extracted. Winning bid technical template: The overall structure, chapter layout, expression style, and language habits of the winning bid document are analyzed and abstracted into a reusable document template or paragraph template. Winning company qualification data: From the commercial bid section of the winning company, key qualification information such as performance cases, patent lists, and team certifications are extracted. This embodiment of the invention can employ techniques such as named entity recognition, relation extraction, and dependency parsing to ensure that the parsed structured information is consistent with the format of the system's existing knowledge base.

[0123] S203. Compare the winning technical solution, winning technical template and winning enterprise qualification data with the existing materials in the tender document knowledge base for similarity and validity verification, and obtain the verification results.

[0124] For example, embodiments of the present invention may use a semantic matching algorithm. The newly parsed winning bid material is converted into a vector, and its cosine similarity to all existing material vectors in the knowledge base is calculated. If the similarity of a new material to existing materials in the database exceeds a high threshold (e.g., 0.95), it is considered highly duplicated and does not need to be added back to the database. The system checks whether the winning bid scheme contains new technologies, methods, or better parameter configurations not yet included in the knowledge base. It also checks whether the content structure complies with laws, regulations, and industry standards to avoid including non-standard documents in the knowledge base. The system calculates a quality score for the new material based on preset rules (e.g., content completeness, logical clarity, and parameter quantification). A structured list is output, indicating whether each new material has passed verification, its quality score, and its deduplication status.

[0125] S204. Based on the verification results, the winning technical solutions, winning technical templates, and winning enterprise qualification data that meet the preset quality standards are used as new materials to update the bid document knowledge base.

[0126] For example, the preset quality standards are: similarity below the deduplication threshold (e.g., <0.95); quality score above the minimum requirement (e.g., >80); and passing novelty and prescriptive tests.

[0127] For example, embodiments of the present invention can vectorize new materials that conform to the standards and add them to the FAISS vector database, updating its index to ensure that subsequent semantic searches can immediately find this new content. In the knowledge graph, entity nodes are created for the new materials, and relationships are established with relevant bidding requirements, technical tags, enterprise entities, etc., enriching the graph's network structure. Rich metadata tags are added to the new materials, such as the winning bid time, project type, bidding unit, and winning bid price range.

[0128] Thus, this invention achieves the evolution from a static knowledge base to a dynamically growing knowledge ecosystem. By learning from the winning bid documents, this invention makes its recommended content increasingly competitive.

[0129] Optionally, the AI ​​robot-based bid document generation method provided in this embodiment of the invention further includes steps S301-S304 in step S105.

[0130] S301. Based on the knowledge base of the tender documents, construct the user's corporate profile.

[0131] In some embodiments, the corporate profile includes the company’s weaknesses in qualifications, areas of technological strength, and historical bidding strategy preferences.

[0132] For example, embodiments of the present invention can identify a company's missing or weak core qualifications by comparing them with common qualifications of industry benchmark companies, such as the lack of CCRC information security service qualifications. By analyzing the technical solutions of historically won bids, and using text classification and keyword frequency statistics, the core technological advantages of the company can be extracted, such as a significant advantage in big data platform architecture. By analyzing historical quotations, the average price discount rate of the company can be calculated; by analyzing technical solutions, strategy tags such as a tendency to use domestically produced components can be summarized.

[0133] S302. Based on the enterprise profile and key information, conduct multi-dimensional matching analysis to identify potential risk items in the optimized bid documents.

[0134] In some embodiments, potential risks are those that do not align with the user company's strengths or fail to address the company's weaknesses.

[0135] For example, embodiments of the present invention can compare the qualification requirements of the tender documents with the company's own qualification weaknesses. If a requirement falls within the weakness area, it is identified as a potential risk item. For example, if the tender requires a CCRC Level 2 qualification, which the company does not possess, this item is marked as high risk. The technical parameters / scoring criteria of the tender documents are semantically matched with the company's areas of technical strength. If a key scoring item is not within the company's areas of strength, or if the company's strengths are not fully demonstrated in this tender, it is identified as a mismatch risk. The optimized tender documents are checked to see if they successfully circumvent the company's known weaknesses and maximize the highlighting of core strengths. A list of potential risk items is output, detailing the risk type, risk location (section in the tender documents), and risk description.

[0136] S303. For potential risk items, the user's strengths and risk avoidance solutions are retrieved from the tender document knowledge base using a semantic matching algorithm.

[0137] For example, regarding the risk of qualification gaps: The knowledge base is searched for existing qualifications, patents, or achievements that are most relevant to the weak qualification or can form a substitute advantage, which are then used as strengths. For instance, even without CCRC qualification, the knowledge base can be used to mitigate the impact of the weakness by retrieving information such as multiple data security-related patents and involvement in national-level security projects. Regarding the risk of mismatched strengths: The knowledge base is proactively searched for the company's strongest supporting materials in that area of ​​strength, such as details of typical cases, technical white papers, and performance test reports, as enhanced strengths. For all risks, standard risk mitigation solution templates are also retrieved from the knowledge base. For example, for a missing qualification, a standard solution might include a commitment to obtain the qualification within X months of winning the bid, or a proposed consortium with Company Y, which possesses the qualification.

[0138] S304. Based on the user's strengths and risk avoidance solutions, enhance the content and optimize the strategies of the optimized bid documents to generate personalized bid documents.

[0139] For example, embodiments of the present invention can strategically insert the most compelling evidence of the company's strengths into the sections on corporate advantages or technical solutions in the tender documents. The technical solutions can be fine-tuned to better reflect the company's familiar and advantageous technological paths. Risk mitigation strategies can be elegantly used in the business deviation table or response description to address the company's qualification weaknesses. For example, a statement could be added: "Given our company's deep experience and numerous patents in the field of…, although we have not yet obtained CCRC qualification, we fully possess and exceed the technical and service capabilities required for that qualification, and we promise…" This adjusts the narrative logic to guide the evaluation experts' attention to the company's core strengths. A personalized tender document is then generated. This document is not only compliant, consistent, and complete, but also a deeply customized, competitive document that leverages strengths and mitigates weaknesses, fully showcasing the company's unique value and problem-solving capabilities.

[0140] Thus, embodiments of the present invention can achieve a strategic upgrade in bid document generation. Going beyond simple information filling and error correction, by introducing enterprise profiling and strategy analysis, the AI ​​robot can act as a seasoned bidding strategy consultant. This invention can provide early warnings of deep-seated risks related to the enterprise's own conditions within the bid documents. Based on the enterprise's unique knowledge assets, it injects the most competitive personalized content into the bid documents. This achieves a shift in bidding strategy, significantly increasing the probability of winning bids in complex, highly competitive projects.

[0141] Optionally, the AI ​​robot-based bid document generation method provided in this embodiment of the invention further includes steps S401-S405 in step S105.

[0142] S401. Obtain information on multiple ongoing tender announcements.

[0143] For example, embodiments of the present invention can automatically and in real-time collect bidding announcement information from multiple sources, including the China Public Service Platform for Bidding and Tendering, public resource trading centers at all levels, procurement platforms of large enterprises (such as e-commerce platforms), and industry-specific bidding websites. The collected metadata includes the announcement title, bidding party, publication time, deadline, project overview, budget amount, and a link to detailed bidding documents.

[0144] S402. Use the natural language processing algorithm of AI robot to perform structured parsing of multiple bidding announcement information to obtain the core requirements, technical parameters and qualification conditions of each bidding project.

[0145] For example, the core requirements are: extract key information such as the project's core objectives, scale, and schedule from the announcement text; identify and extract key technical indicators and performance requirements mentioned in the announcement; and extract mandatory qualification requirements that bidders must possess, such as corporate qualifications, performance requirements, and financial requirements. A structured project profile is generated for each bidding project.

[0146] S403. Based on the core requirements, technical parameters, and qualification conditions of each bidding project, as well as the user's corporate profile, multi-dimensional matching calculations are performed to obtain the matching degree calculation results.

[0147] In some embodiments, the matching degree calculation results include qualification compliance, technical advantage matching degree, and historical winning probability.

[0148] For example, qualification compliance calculation employs rule matching and weighted scoring. The qualification requirements in the tender documents are compared item by item with the enterprise qualification data in the enterprise profile. Each compliant qualification receives 1 point, and each non-compliant qualification receives 0 points. A weighted sum is calculated based on the importance of the qualifications (e.g., mandatory qualifications have higher weight), and finally, a normalized qualification compliance score between 0 and 1 is obtained.

[0149] For example, the calculation of the matching degree of technical advantages is carried out using semantic matching. The technical parameters and core requirements of the bidding project are converted into query vectors. The cosine similarity between this query vector and the technical advantage domain vector in the enterprise profile (composed of a set of keywords representing the enterprise's advantageous technologies) is calculated. The similarity score is the matching degree of technical advantages; the higher the score, the better the project can leverage the enterprise's technical strengths.

[0150] For example, historical bid-winning probability calculation: A competitiveness assessment model is reused. Based on the core requirements, technical parameters, and budget of the bidding project (simulating a competitive bidding range), a virtual scheme-price feature vector is constructed and input into the competitiveness assessment model to predict the possible historical bid-winning probability for the project. For each bidding project, a tuple of matching degree calculation results is output, including qualification compliance, technical advantage matching degree, and historical bid-winning probability.

[0151] S404. Based on the matching degree calculation results, prioritize each bidding project and select high-quality bidding projects with a matching degree higher than the preset threshold.

[0152] For example, this invention uses a comprehensive ranking algorithm (such as a weighted total score method) to calculate the final priority score for each project. Final score = W1 * Qualification compliance + W2 * Technical advantage matching + W3 * Historical winning probability; where W1, W2, and W3 are configurable weights that can be adjusted according to company strategy (e.g., W3 has a higher weight if a high winning rate is desired). A total preset threshold is set. All bidding projects are sorted in descending order of their final scores, and projects with scores higher than the threshold are selected and marked as high-quality bidding projects.

[0153] S405. Generate a personalized list of recommended bidding projects based on high-quality bidding projects with a matching degree higher than a preset threshold.

[0154] In some embodiments, the recommendation list includes project matching analysis and bidding strategy suggestions.

[0155] For example, embodiments of the present invention can generate a structured, personalized list of recommended bidding projects. This list includes basic project information and in-depth analytical data: Project Matching Analysis: Visually displaying the scores of each recommended project across dimensions such as qualifications, technology, and probability of winning, using charts or text descriptions, clearly indicating the project's strengths and potential risks. Bidding Strategy Suggestions: Based on the matching analysis results, the system generates preliminary bidding strategy suggestions. For example: If qualifications are fully met and the technological advantage matching degree is as high as 90%, it is recommended as a key bidding project, employing a strategy of technological leadership and moderate pricing. If qualifications are met, but the technological matching degree is average, the probability of winning is moderate. It is recommended as a follow-up project, focusing on researching competitors and optimizing the details of the technical solution. If one qualification is not met (XX certificate), the cost and time of obtaining that qualification should be immediately assessed, or a consortium with a company possessing that qualification should be considered for bidding.

[0156] Thus, embodiments of the present invention can upgrade the capabilities of AI robots from passive document generation tools to proactive business opportunity analysis and recommendation engines. This automates the screening of massive amounts of bidding information, improving efficiency. Through multi-dimensional quantitative analysis, it helps companies concentrate their limited bidding resources on projects with the highest success rate and those best suited to their strengths, avoiding blind bidding and achieving precise focus.

[0157] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0158] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0159] Figure 2A schematic diagram of a tender document generation device based on an AI robot, according to an embodiment of the present invention, is shown. The generation device 500 includes a communication module 501 and a processing module 502.

[0160] The communication module 501 is used to receive the tender documents uploaded by the user.

[0161] Processing module 502 is used to perform structured parsing of the tender documents using the natural language processing algorithm of an AI robot to obtain key information, including qualification requirements, technical parameters, scoring criteria, and mandatory clauses. Based on the key information, the module uses the semantic matching algorithm of the AI ​​robot and the tender document knowledge base to perform retrieval, matching, and automatic filling to obtain a draft tender document corresponding to the key information. The tender document knowledge base stores historical winning bids, standard technical templates, and enterprise qualification data. The module performs compliance, consistency, and completeness reviews on the draft tender document to obtain the review results. Based on the review results, the module optimizes the draft tender document and outputs the optimized tender document.

[0162] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. The electronic device 600 includes: a processor 601, a memory 602, and a computer program 603 stored in the memory 602 and executable on the processor 601. When the processor 601 executes the computer program 603, it implements the steps in the above-described method embodiments. Alternatively, when the processor 601 executes the computer program 603, it implements the functions of each module / unit in the above-described device embodiments.

[0163] For example, the computer program 603 may be divided into one or more modules / units, which are stored in the memory 602 and executed by the processor 601 to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program 603 in the electronic device 600.

[0164] The processor 601 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0165] The memory 602 can be an internal storage unit of the electronic device 600, such as a hard disk or memory of the electronic device 600. The memory 602 can also be an external storage device of the electronic device 600, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD) card, flash card, etc., equipped on the electronic device 600. Furthermore, the memory 602 can include both internal and external storage units of the electronic device 600. The memory 602 is used to store the computer program and other programs and data required by the terminal. The memory 602 can also be used to temporarily store data that has been output or will be output.

[0166] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for assisting in the generation of tender documents based on an AI robot, characterized in that, include: Receive tender documents uploaded by users; The tender document is structured and parsed using a natural language processing algorithm of an AI robot to obtain key information, including qualification requirements, technical parameters, scoring criteria, and mandatory clauses. Based on the aforementioned key information, the semantic matching algorithm of the AI ​​robot and the knowledge base of the tender document are used to perform retrieval, matching and automatic filling to obtain the initial draft of the tender document corresponding to the key information; The tender document knowledge base stores historical winning bids, standard technical templates, and enterprise qualification data. The initial draft of the tender documents was reviewed for compliance, consistency, and completeness, and the review results were obtained. Based on the review results, the initial draft of the tender document is optimized, and the optimized tender document is output.

2. The method for assisting in the generation of tender documents based on an AI robot according to claim 1, characterized in that, The natural language processing algorithm using AI robots performs structured parsing of the tender documents to obtain key information, including: The named entity recognition model based on AI robots traverses the tender documents to identify key entities that represent the tender requirements; the key entities include qualification categories, technical parameter indicators, scoring items, and mandatory clause identifiers. Based on the tender documents, dependency parsing and relation extraction techniques are used to construct semantic relationships between key entities. Based on the semantic relationships between the key entities and the preset structured template, the key entities are integrated to construct a set of key information. Based on the aforementioned key information set and word vector model, the technical terms of each key entity are expanded using synonyms and near-synonyms to obtain an expanded key information set. Based on the expanded set of key information, a contextual semantic understanding model is used to verify the semantic integrity and logical consistency of the key information set, thereby obtaining the key information.

3. The method for assisting in the generation of tender documents based on an AI robot according to claim 1, characterized in that, Based on the key information, the AI ​​robot's semantic matching algorithm and the tender document knowledge base are used to perform retrieval, matching, and automatic filling to obtain a draft tender document corresponding to the key information, including: Based on the aforementioned key information, a query vector is generated; The historical winning bids, standard technical templates, and enterprise qualification data in the bid document knowledge base are vectorized to obtain a vector database. Based on the query vector, a semantic matching algorithm is used to calculate the semantic similarity between the query vector and the material vectors of each type in the vector database; Based on the semantic similarity, the vectors of each type of material in the vector database are sorted and filtered to obtain a set of recommended materials that match the key information; Based on preset mapping rules, the materials in the recommended material set are automatically filled into the bid document template to generate the initial draft of the bid document.

4. The method for assisting in the generation of tender documents based on an AI robot according to claim 1, characterized in that, The review of the initial draft of the tender documents for compliance, consistency, and completeness, and the resulting review results, include: Based on the initial draft of the tender document and the mandatory clauses of the key information, a responsiveness calculation is performed to obtain the response results of the initial draft of the tender document to each mandatory clause, and the response results include response and non-response; Based on the response results of the initial draft of the tender documents to each mandatory clause, the compliance review results of the tender documents are determined. Based on the initial draft of the tender document, logical reasoning and data correlation analysis are performed to determine the consistency review results. The consistency review results include the consistency between the technical solution description and technical parameters, the consistency between the itemized price and the total price, and the consistency of the description of the same matter in different chapters. Based on the pre-set list of essential elements for the tender documents, the chapters, attachments, and key contents of the initial draft of the tender documents are reviewed to obtain the results of the completeness review. Based on the compliance review results, consistency review results, and integrity review results, issues are summarized to obtain an issue list; the issue list includes issue type and issue location; The review results are generated based on the compliance review results, consistency review results, and integrity review results, as well as the issue list.

5. The method for assisting in the generation of tender documents based on an AI robot according to claim 4, characterized in that, The consistency review results also include competitiveness consistency; Accordingly, the process of determining the consistency review results based on the initial draft of the tender document through logical reasoning and data correlation analysis includes: Extract the technical configuration parameters, service commitment indicators, and bid price from the initial draft of the tender document; Based on the aforementioned technical configuration parameters, service commitment indicators, and bid price, a solution-bid feature vector is constructed. The scheme-quote feature vector is input into a pre-trained competitiveness evaluation model to obtain the winning probability; the competitiveness evaluation model is trained and generated based on historical winning bid datasets and is used to quantify the winning probability of scheme-quote combinations. When the probability of winning the bid is lower than a preset competitiveness threshold, it is determined that there is a problem of inconsistent competitiveness. If there is an inconsistency in competitiveness, the characteristic dimensions that lead to insufficient competitiveness will be identified based on the output of the competitiveness assessment model, and optimization suggestions will be generated. The optimization suggestions include technical solution adjustment items, service commitment optimization items, and / or quotation modification items.

6. The method for assisting in the generation of tender documents based on an AI robot according to claim 1, characterized in that, Based on the review results, the process of optimizing the initial draft of the tender document and outputting the optimized tender document includes: Based on the compliance review results and the mandatory clauses, the semantic matching algorithm is used to re-retrieve and match compliance content from the tender document knowledge base; Based on the re-searched and matched compliant content, the non-compliant content in the initial draft of the tender document is replaced; Based on the consistency review results, the baseline data in the initial draft of the tender document is determined through a logical reasoning model; Based on the baseline data in the initial draft of the tender document, inconsistencies in the initial draft of the tender document are corrected; Based on the results of the integrity review and the tender document knowledge base, supplementary content is determined; Based on the supplementary information, the missing items in the initial draft of the tender document were completed, resulting in an optimized tender document.

7. The method for assisting in the generation of tender documents based on an AI robot according to claim 1, characterized in that, After optimizing the initial draft of the tender document based on the review results and outputting the optimized tender document, the process further includes: Receive the final version of the winning bid document uploaded by the user; the final version of the winning bid document includes the winning bid proposal and the enterprise qualification data corresponding to the winning bid proposal; The AI ​​robot's natural language processing algorithm is used to perform structured parsing on the final version of the winning bid document to obtain the winning technical solution, the winning technical template, and the winning company's qualification data; The winning technical solution, winning technical template, and winning enterprise qualification data are compared with the existing materials in the bid document knowledge base for similarity and validity verification, and the verification results are obtained. Based on the verification results, the winning technical solutions, winning technical templates, and winning enterprise qualification data that meet the preset quality standards are used as new materials to update the bid document knowledge base.

8. The method for assisting in the generation of tender documents based on an AI robot according to claim 1, characterized in that, After optimizing the initial draft of the tender document based on the review results and outputting the optimized tender document, the process further includes: Based on the aforementioned tender document knowledge base, a user's corporate profile is constructed; the corporate profile includes the company's qualification weaknesses, areas of technological strength, and historical bidding strategy preferences. Based on the enterprise profile and the key information, a multi-dimensional matching analysis is performed to identify potential risk items in the optimized bid document; the potential risk items are those that do not match the user enterprise's strengths or fail to avoid the enterprise's weaknesses. For the aforementioned potential risk items, a semantic matching algorithm is used to retrieve the user's strengths and risk avoidance solutions from the tender document knowledge base; Based on the user's strengths in content display and risk avoidance solutions, the optimized bid documents are enhanced and the strategies are adjusted to generate personalized bid documents.

9. The method for assisting in the generation of tender documents based on an AI robot according to claim 1, characterized in that, After optimizing the initial draft of the tender document based on the review results and outputting the optimized tender document, the process further includes: Obtain information on multiple ongoing tender announcements; The AI ​​robot's natural language processing algorithm is used to perform structured parsing of the multiple bidding announcements to obtain the core requirements, technical parameters, and qualification conditions of each bidding project; Based on the core requirements, technical parameters, and qualification conditions of each bidding project, as well as the user's corporate profile, a multi-dimensional matching calculation is performed to obtain the matching degree calculation result; the matching degree calculation result includes qualification compliance, technical advantage matching degree, and historical winning probability; Based on the matching degree calculation results, each bidding project is prioritized and selected as a high-quality bidding project with a matching degree higher than a preset threshold. Based on high-quality bidding projects with a matching degree higher than a preset threshold, a personalized bidding project recommendation list is generated. The recommendation list includes project matching degree analysis and bidding strategy suggestions.

10. A tender document generation system based on an AI robot, characterized in that, The system includes an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor is used to invoke and run the computer program stored in the memory to perform the method as described in any one of claims 1 to 9.

Images