Artificial intelligence-based collection decision-making method and system, and storage medium
By using an AI-based procurement decision-making method, which utilizes AI models to generate and correct tender document templates, qualification requirements, and evaluation methods, the accuracy and efficiency issues of existing tendering and procurement decisions have been resolved, resulting in an efficient and accurate tendering and procurement process.
Patent Information
- Application Number
- CN202511538061.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-27
- Publication Date
- 2026-01-27
AI Technical Summary
Existing AI tools lack specificity in bidding and procurement decisions, resulting in decisions that do not match actual needs, missing or redundant key information, low accuracy, and an inability to meet the needs of rapidly advancing projects.
By employing an AI-based procurement decision-making approach, we utilize AI models to acquire project initiation information, generate reference document templates, and determine target document templates, qualification requirements, evaluation methods, and tender documents through multi-level review and correction processes. We combine historical information and rule bases for precise matching and adjustment to ensure compliance and accuracy.
It improves the efficiency and accuracy of bidding and procurement decisions, reduces manual intervention, enables the rapid generation of bidding documents that meet actual needs, and reduces legal risks and labor costs.
Smart Images

Figure CN121413591A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of bidding and procurement technology, and in particular to a bidding and procurement decision-making method, system and storage medium based on artificial intelligence. Background Technology
[0002] Tendering and procurement decisions refer to a series of information released by the procuring entity or tendering party when conducting procurement, engineering, or service projects, providing potential bidders with details, requirements, and conditions related to the project. Tendering and procurement has become a common, open, fair, and competitive method for governments, enterprises, and public institutions to select suppliers in engineering construction, goods procurement, and service purchase processes.
[0003] Traditional procurement decisions rely heavily on manual processes, including project initiation, setting qualification requirements, formulating evaluation methods, and drafting and reviewing tender documents. While assigning experienced personnel can ensure accuracy to some extent, the entire process is time-consuming and labor-intensive, resulting in low efficiency and failing to meet the demands of today's rapid project advancement.
[0004] With the continuous development of artificial intelligence technology, the method of determining bidding and procurement decisions is gradually shifting from manual operation to direct determination by AI tools. While this approach significantly improves work efficiency and can quickly generate bidding and procurement-related documents and decisions, most existing AI tools are designed for general use and lack deep adaptation to specific scenarios and professional needs in the bidding and procurement field. When facing bidding projects of different industries, scales, and types, the bidding and procurement decisions generated by AI tools often suffer from problems such as inconsistency with actual needs, missing or redundant key information, and insufficient compliance, resulting in low accuracy of the determined bidding and procurement decisions and failing to effectively guarantee the smooth progress of bidding and procurement projects. Therefore, there is an urgent need for a method that can balance efficiency and accuracy, and can accurately generate bidding and procurement decisions based on the actual situation of the bidding project. Summary of the Invention
[0005] In order to efficiently and accurately determine bidding and procurement decisions, this application provides an artificial intelligence-based bidding and procurement decision-making method, system, and storage medium.
[0006] Firstly, an artificial intelligence-based procurement decision-making method is provided, the method comprising: Obtain project initiation information, send the project initiation information to the AI model to obtain a reference document template, and determine the target document template based on the reference document template; The target file template is sent to the AI model to obtain reference qualification requirements, and the target qualification requirements are determined based on the reference qualification requirements. The target qualification requirements are sent to the AI model to obtain a reference evaluation method, and the target evaluation method is determined based on the reference evaluation method. The target evaluation method is sent to the AI model to obtain reference relevant elements, and the target relevant elements are determined based on the reference relevant elements. The target-related elements are sent to the AI model to obtain a reference tender document, and the target tender document is determined based on the reference tender document. The target tender document is sent to the AI model to obtain the review result of the target tender document. If the review result is that the review is passed, the target tender document is encrypted to obtain the procurement decision. If the audit result is that the audit fails, the target tender document shall be adjusted to obtain a new target tender document.
[0007] In some embodiments, the project initiation information includes project attributes, and determining the target file template based on the reference file template includes: Extract reference guide words from the reference document template, wherein the reference guide words include several reference guide sub-words; Determine whether all referenced guiding subkeys belong to the project attribute; if so, determine the referenced file template as the target file template. If not, obtain abnormal reference guide words that do not belong to the project attributes, and retrieve the correction reference guide words corresponding to the abnormal reference guide words from the preset historical file template library; The corrected reference guide words, the abnormal reference guide words, and the reference file template are sent to the AI model to obtain the target file template.
[0008] In some embodiments, the reference qualification requirements include several reference qualification sub-requirements, and determining the target qualification requirements based on the reference qualification requirements includes: Determine whether a supplementary qualification requirement has been received. If so, perform semantic fusion on the supplementary qualification requirement and add it to the reference qualification requirement to obtain a new reference qualification requirement. If not, the aforementioned reference qualification requirements shall be used as the new reference qualification requirements; Based on historical bidding information, obtain the reference compliant bidding information corresponding to each reference qualification sub-requirement, and determine the number of intersections of all reference compliant bidding information; Determine whether there are no conflicts among all the reference qualification sub-requirements and whether the number of intersections is not less than the preset number of bids. If so, determine the reference qualification requirement as the target qualification requirement. If not, obtain the adjusted reference qualification sub-requirements corresponding to the number of conflicting reference qualification sub-requirements and / or the number of intersections, and send the adjusted reference qualification sub-requirements and the compliant reference qualification sub-requirements that do not belong to the adjusted reference qualification sub-requirements to the AI model to obtain the target qualification requirements.
[0009] In some embodiments, the reference evaluation method includes several reference evaluation sub-methods, and each reference evaluation sub-method corresponds to an evaluation weight. Determining the target evaluation method based on the reference evaluation method includes: The reference bid evaluation sub-methods are sorted in descending order of their bid evaluation weights to obtain a sequence of reference bid evaluation sub-methods. Determine whether the reference evaluation method sequence conforms to the preset sequence; if it does, determine the reference evaluation method as the target evaluation method. If it does not meet the requirements, an adjusted evaluation method is generated based on the preset sequence, and the adjusted evaluation method is sent to the AI model to obtain the target evaluation method.
[0010] In some embodiments, determining the target relevant elements based on the reference relevant elements includes: Based on the supplementary qualification requirements, the evaluation criteria are determined, and it is determined whether the reference relevant elements meet the evaluation criteria. If they do, the reference relevant elements are determined as target relevant elements. If it does not meet the criteria, the adjustment element information that needs to be adjusted for the reference related elements is determined based on the evaluation element criteria, and the adjustment element information is sent to the AI model to obtain the target related elements.
[0011] In some embodiments, determining the target tender document based on the reference tender document includes: The reference tender document is structured and parsed to obtain non-fixed clause fields and key parameter fields; The target qualification requirements, target evaluation methods, and target-related elements are matched with the non-fixed clause fields using multimodal matching to generate clause-filling content; Based on a pre-defined knowledge graph of tender document compliance, the key parameter fields are validated for compliance. If the verification finds that the reference tender document is compliant, the reference tender document will be determined as the target tender document; If the verification finds that the project does not meet the requirements, the replacement parameters that match the project attributes are retrieved from the historical bidding documents. The replacement parameters are used to update the key parameter field to obtain the replacement key field. The replacement key field and the clause fill content are sent to the AI model to obtain the target bidding document.
[0012] In some embodiments, the method further includes: Obtain the revised document template corresponding to the target tender document, and send the revised document template to the AI model for storage.
[0013] In some embodiments, the method further includes: Obtain offline bidding documents, send the offline bidding documents to the AI model to obtain the review result of the offline bidding documents, and if the review result is that the review is passed, encrypt the target bidding documents to obtain the procurement decision; If the audit result is that the audit fails, the offline bidding documents will be adjusted to obtain new offline bidding documents.
[0014] Secondly, an artificial intelligence-based procurement decision-making system is provided, which is applied in a procurement decision-making platform. The system includes: Thirdly, a computer-readable storage medium is provided, on which a computer program capable of running on a processor is stored. When executed by the processor, the computer program implements an artificial intelligence-based procurement decision-making method as described in the first aspect. This includes modules for document templates, qualification requirements, evaluation methods, relevant elements, tender documents, and document review; wherein... The document template module is used to obtain project initiation information, send the project initiation information to the AI model to obtain a reference document template, and determine the target document template based on the reference document template. The qualification requirements module is used to send the target file template to the AI model to obtain reference qualification requirements, and to determine the target qualification requirements based on the reference qualification requirements; The bid evaluation method module is used to send the target qualification requirements to the AI model to obtain a reference bid evaluation method, and determine the target bid evaluation method based on the reference bid evaluation method. The relevant elements module is used to send the target evaluation method to the AI model to obtain reference relevant elements, and to determine the target relevant elements based on the reference relevant elements; The tender document module is used to send the target-related elements to the AI model to obtain a reference tender document, and to determine the target tender document based on the reference tender document. The document review module is used to send the target tender document to the AI model to obtain the review result of the target tender document. If the review result is that the review is passed, the target tender document is encrypted to obtain the procurement decision. The tender document module is also used to adjust the target tender document to obtain a new target tender document if the review result is that the review is not passed.
[0015] By employing the above method, this application first obtains project initiation information and sends it to an AI model to obtain a reference document template. Based on the reference document template, a target document template is determined. Then, the target document template is sent to the AI model to obtain reference qualification requirements, and the target qualification requirements are determined based on these requirements. Subsequently, the target qualification requirements are sent to the AI model to obtain reference evaluation methods, and the target evaluation methods are determined based on these methods. Next, the target evaluation methods are sent to the AI model to obtain reference relevant elements, and the target relevant elements are determined based on these elements. Then, the target relevant elements are sent to the AI model to obtain reference tender documents, and the target tender documents are determined based on these documents. Finally, the target tender documents are sent to the AI model to obtain the review results. If the review result is "approved," the target tender documents are encrypted to obtain a procurement decision. If the review result is "not approved," the target tender documents are adjusted to obtain a new target tender document. This AI-based procurement decision-making method is applied to a procurement decision-making platform. This platform is linked to an AI model, which uses its artificial intelligence to assist in initially determining some information, thereby reducing the workload of the electronic procurement platform and improving its efficiency. At the same time, the procurement decision-making platform will also conduct corresponding review and processing on the information obtained by the AI model, so that the procurement decision-making platform can quickly obtain accurate bidding and procurement decisions by performing only light calculations. Attached Figure Description
[0016] Figure 1 This is a flowchart of a procurement decision-making method based on artificial intelligence, provided in an embodiment of this application.
[0017] Figure 2 This is a flowchart of the method for determining the target file template based on the reference file template provided in this application.
[0018] Figure 3 This is a flowchart illustrating the AI-based procurement decision-making process provided in this application.
[0019] Figure 4 This is a schematic diagram of a procurement decision-making system based on artificial intelligence, provided in an embodiment of this application. Detailed Implementation
[0020] To better understand the purpose, technical solutions, and advantages of this application, it has been described and illustrated below with reference to the accompanying drawings and embodiments. However, those skilled in the art should understand that this application can be implemented without these details. It will be apparent to those skilled in the art that various modifications can be made to the embodiments disclosed in this application, and the general principles defined in this application can be applied to other embodiments and application scenarios without departing from the principles and scope of this application. Therefore, this application is not limited to the illustrated embodiments, but is consistent with the broadest scope claimed in this application.
[0021] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.
[0022] Figure 1 This is a flowchart illustrating a procurement decision-making method based on artificial intelligence, as provided in an embodiment of this application. Figure 1 As shown, an artificial intelligence-based procurement decision-making method includes the following steps: Step S100: Obtain project initiation information, send the project initiation information to the AI model to obtain a reference document template, and determine the target document template based on the reference document template.
[0023] This application describes an AI-based procurement decision-making method applied to a procurement decision-making platform. This platform is linked to an AI model, which first uses its artificial intelligence to assist in determining preliminary information, thereby reducing the workload of the electronic procurement platform and improving its efficiency. Simultaneously, the platform also performs corresponding review and processing on the information obtained from the AI model, enabling it to quickly and accurately determine bidding and procurement decisions using only minimal computation. This application describes the procurement decision-making platform from the perspective of the procurement decision-making platform itself.
[0024] The aforementioned project initiation information specifically refers to the information determined by the bidding party during the preliminary organizational meeting for this bidding process. This information can be in the form of meeting minutes, documents, and audio recordings. The bidding party can send this information to the procurement decision-making platform, allowing the platform to access it. The project initiation information includes project attributes, indicating the industry and specific work within that industry. The aforementioned AI models encompass multiple types, including AI template models, AI qualification requirement models, AI bid evaluation method models, AI related element models, AI bidding document models, and AI review models. Each type of AI model is trained offline using extensive historical data.
[0025] Once the procurement decision-making platform obtains project initiation information, it signifies the commencement of procurement decision-making. At this point, the platform sends the acquired project initiation information to the AI template model within its AI framework. To obtain more detailed procurement decisions for subsequent bidding processes, the AI template model retrieves a comprehensive document template from its database based on the industries involved in the project initiation information and designates this comprehensive document template as the reference document template. The AI template model then sends the obtained reference document template to the procurement decision-making platform, providing it with a reference document template. The procurement decision-making platform then reviews and modifies the obtained reference document template to determine the target document template. Figure 2 This is a flowchart illustrating the method for determining the target file template based on a reference file template provided in this application. For example... Figure 2 As shown, determining the target file template based on the reference file template includes the following steps: Step S101: Extract reference guide words from the reference document template, wherein the reference guide words include several reference guide sub-words.
[0026] Step S102: Determine whether all reference guide words belong to project attributes. If so, determine the reference file template as the target file template.
[0027] Step S103: If not, obtain the abnormal reference guidance sub-words that do not belong to the project attributes, and retrieve the correction reference guidance sub-words corresponding to the abnormal reference guidance sub-words from the preset historical file template library.
[0028] Step S104: Send the corrected reference guiding words, abnormal reference guiding words, and reference file template to the AI model to obtain the target file template.
[0029] The procurement decision-making platform uses keyword extraction technology to extract reference guiding words from the reference document template. Each item or column in the reference document template corresponds to a reference guiding sub-word, and the reference guiding words include several reference guiding sub-words.
[0030] The system uses a pre-built project attribute dictionary and a general attribute dictionary to perform semantic vectorization and similarity matching on each reference guiding word to determine whether it belongs to a project attribute. Specifically, a word embedding model is used to convert each reference guiding word into a semantic vector, and cosine similarity is calculated with the words in the attribute dictionary. If the highest similarity is greater than a preset similarity threshold, the reference guiding word is determined to belong to a project attribute, and the general attribute discrimination process is initiated simultaneously. A general attribute knowledge graph is also pre-built, whose nodes cover general terminology, legal clauses, standard fields, and business process stages in the bidding and procurement field. The discrimination process is as follows: First, semantic vector matching is performed, that is, using a sentence encoding model, such as Sentence-BERT, abnormal reference guiding words are converted into semantic vectors and cosine similarity is calculated with the node vectors in the knowledge graph. If the result is higher than a first threshold, the next step of verification is performed. Then, contextual consistency verification is performed, that is, the contextual context of the word in the reference document template is extracted and input into a pre-trained early warning classification model to determine whether the context belongs to the general and normative description scope of the bidding document. Finally, a comprehensive decision is made based on a pre-built general language pattern rule base. If the subword is verified to be of a general nature at both the semantic and contextual levels, it is classified as a general attribute. Otherwise, it is ultimately determined to be an abnormal reference guiding subword, and its abnormal type is recorded.
[0031] Then, by examining the attributes corresponding to each reference guide keyword, if each reference guide keyword corresponds to either a project attribute or a general attribute, then all reference guide keywords belong to the project attribute category. If at least one reference guide keyword corresponds to an attribute that is neither a project attribute nor a general attribute, then not all reference guide keywords belong to the project attribute category. Furthermore, different departments within the same industry may correspond to the same set of general attributes.
[0032] If it is determined that all reference guiding keywords belong to the project attributes, it means that the reference document template obtained above does not contain guiding keywords that do not belong to this bidding work. The reference document template sent by the AI model is accurate, and the reference document template is determined to be the template needed for subsequent procurement decision-making work, that is, the target document template.
[0033] If it is determined that not all reference guiding keywords belong to project attributes, it indicates that the reference document template obtained above contains guiding keywords that do not belong to this tender work. These reference guiding keywords, which belong to neither project attributes nor general attributes, are identified as abnormal reference guiding keywords. These abnormal reference guiding keywords may be abnormal because they do not belong to this tender work, or they may be abnormal because of extraction errors, such as the presence of an incorrect character.
[0034] To reduce the occurrence of missing reference guide words due to the direct deletion of abnormal reference guide words, the system also checks whether the historical reference guide words corresponding to the project attribute in the preset historical file template library have a greater overlap with the abnormal reference guide words than the preset overlap. If there is one, the historical reference guide word is identified as the corrected reference guide word; if not, the blank information is identified as the corrected reference guide word.
[0035] After identifying anomalous reference guide words and retrieving corresponding corrected reference guide words from the historical file template library, further refinement operations are performed to obtain the final target file template. This involves first constructing structured correction instructions, then injecting contextual information, followed by AI model execution and generation, and finally returning the result.
[0036] The construction of structured correction instructions is not simply about randomly sending correction reference guide words and abnormal reference guide words to the AI model. Instead, it involves building a structured set of correction instructions that explicitly identifies the operation type, location information, and content data. The operation type specifies whether to perform a replacement or deletion operation for each abnormal reference guide word. If a correction reference guide word with semantic similarity exceeding a preset threshold is successfully found in the historical file template library, the operation is a replacement. If no similarity word is found or the similarity is insufficient, the operation is a deletion. Location information refers to the specific position of each abnormal reference guide word within the reference file template. Content data refers to the specific content of the correction reference guide word in the instruction for replacement operations; for deletion operations, the instruction explicitly indicates the deletion target.
[0037] The contextual information injection is designed to help the AI model understand the correction intent more accurately. Project attributes are used as key contextual information and sent to the AI model along with the structured correction instructions and the original operation file template. This allows the AI model to better perform replacement or deletion operations based on the specific industry, type, and other attributes of the current project, ensuring that the generated content is highly consistent with the project attributes in terms of style and professionalism.
[0038] The AI model execution and generation process refers to the AI model receiving the aforementioned information packet. Based on structured correction instructions, the AI model performs precise replacement or deletion operations at specified locations in the reference document template. When performing replacement operations, the AI model does not mechanically fill in text; instead, it leverages its natural language understanding and generation capabilities to ensure that the corrected reference guiding words can be smoothly and compliantly integrated into the context of the original template, generating a semantically coherent and formatted target document template.
[0039] The result return process refers to the AI model returning the generated target document template to the procurement decision-making platform. This approach, on the one hand, organically combines and clearly divides the tasks of human review and judgment with the AI's semantic understanding and content generation capabilities by introducing structured correction instructions, overcoming the drawbacks of relying solely on AI or solely on human review. On the other hand, it specifically identifies the location and nature of errors and recommends corrective measures, greatly reducing the AI model's understanding burden and generation uncertainty, making the correction process more efficient, reliable, and predictable. Furthermore, through contextual information injection, the AI model's correction behavior becomes scenario-adaptive, ensuring that the final generated target document template is not only correct in wording but also deeply matched to project requirements in terms of domain knowledge and professional standards. This hybrid intelligent model, combining rule-based judgment and semantic understanding, significantly improves the accuracy and professionalism of the procurement decision-making platform's output.
[0040] Step S200: Send the target file template to the AI model to obtain reference qualification requirements, and determine the target qualification requirements based on the reference qualification requirements.
[0041] Obtaining the aforementioned target document template indicates that the procurement decision-making platform has completed the template work required for the bidding process. The next step is to determine the qualification requirements for the procurement decision. The procurement decision-making platform sends the obtained target document template to the AI qualification requirement model within its AI model. The AI qualification requirement model, based on the project requirements in the project initiation information and relevant historical bidding documents, first uses NLP (Natural Language Processing) technology to parse the basic project requirements and automatically extract key parameters, such as keywords related to project scale, technical complexity, and budget. Then, based on these key parameters, it matches the qualification dimension framework from historical bidding documents, such as basic qualifications, professional qualifications, and project experience, to construct a preliminary qualification requirement structure, i.e., obtaining reference qualification requirements, ensuring no core elements are omitted. For example, in a smart park project, after inputting requirements, the system can automatically divide the requirements into layers such as the IoT sensing layer, reducing manual processing time. The AI qualification requirement model then sends the obtained reference qualification requirements to the procurement decision-making platform, enabling the platform to obtain these requirements. The procurement decision-making platform then reviews and corrects the obtained reference qualification requirements to determine the target qualification requirements. The reference qualification requirements include several sub-requirements. Determining the target qualification requirements based on the reference qualification requirements involves the following steps: Step S201: Determine whether a supplementary qualification requirement has been received. If so, perform semantic fusion on the supplementary qualification requirement and add it to the reference qualification requirement to obtain a new reference qualification requirement.
[0042] Step S202: If not, use the reference qualification requirements as the new reference qualification requirements.
[0043] Step S203: Based on historical bidding information, obtain the reference compliant bidding information corresponding to each reference qualification sub-requirement, and determine the number of intersections of all reference compliant bidding information.
[0044] Step S204: Determine whether there are no conflicts among all the reference qualification sub-requirements and whether the number of intersections is not less than the preset bid quantity. If so, determine the reference qualification requirements as the target qualification requirements.
[0045] Step S205: If not, obtain the conflicting reference qualification sub-requirements and / or the adjustment reference qualification sub-requirements corresponding to the intersection quantity adjustment, and send the adjustment reference qualification sub-requirements and the reference qualification sub-requirements that do not belong to the adjustment reference qualification sub-requirements to the AI model to obtain the target qualification requirements.
[0046] The aforementioned supplementary qualification requirements refer to those sent by the bidding staff to the procurement decision-making platform. Whether these supplementary qualification requirements have been received can be determined by checking if the platform received such requests. If no requests were received, it indicates that the supplementary qualification requirements were not received. If they were received, it indicates that the supplementary qualification requirements were received. Upon receiving a supplementary qualification requirement, the multi-source information fusion module is activated to semantically align and structurally integrate the supplementary qualification requirements with the AI-generated reference qualification requirements, generating new reference qualification requirements. This process employs a BERT-based semantic similarity matching algorithm to identify overlaps, conflicts, or supplementary relationships between the supplementary and original requirements, ensuring logical consistency and freedom of redundancy after information fusion. If no supplementary qualification requirements are received, the reference qualification requirements sent by the AI model are designated as the new reference qualification requirements.
[0047] The aforementioned historical bidding information refers to information on bidders who have participated in bidding in the past. Based on this historical bidding information, a reverse matching search is performed on each reference qualification sub-requirement to obtain all potential bidding companies that meet that sub-requirement, forming a reference compliant bidding information list. Specifically, this reference compliant bidding information list refers to companies that meet the reference qualification sub-requirement.
[0048] Then, the intersection of the reference qualified bid information corresponding to each obtained reference qualification sub-requirement is calculated to obtain the intersection bid information that simultaneously satisfies all reference qualification sub-requirements. The number of such intersection bid information is then counted to obtain the intersection count. High-dimensional set operations and inverted index technology are used to achieve rapid matching and intersection calculation of qualifications for a large number of enterprises.
[0049] The number of intersections is then compared with the preset bid quantity, and a comparison is made to determine if there are any conflicts among all the reference qualification sub-requirements. The preset bid quantity can be determined based on actual circumstances. Considering that not all eligible intersection bids will participate in the bidding, and that a selection process is needed from among the bidders, the preset bid quantity is twice the bid quantity, calculated by multiplying the historical bid participation rate by the intersection quantity. A rule engine and knowledge graph are used to determine if there are any logical conflicts among all the reference qualification sub-requirements, such as a requirement for registered capital of no less than 10 million versus a requirement for registered capital of no more than 5 million.
[0050] If the number of intersections is not less than the preset number of bids and there is no conflict among all the reference qualification sub-requirements, it indicates that the reference qualification requirements obtained in the above steps S301 or S302 are accurate, and the reference qualification requirements are determined as the target qualification requirements.
[0051] When the number of intersections is less than the preset number of bids but there are no conflicts among all the reference qualification sub-requirements, it indicates that the reference qualification requirements obtained in steps S301 or S302 are too high. Bidding based on these qualification requirements will not yield the desired bidders. In this case, the reference qualification requirements obtained in steps S301 or S302 are determined to be inconsistent with the reference qualification sub-requirements, and a relaxed matching strategy is adopted to relax the non-core qualification sub-requirements, resulting in adjusted reference qualification sub-requirements. Here, "adjusted reference qualification sub-requirements" refers to the adjusted qualification requirements.
[0052] When the number of intersections is not less than the preset number of bids but there are conflicts among all the reference qualification sub-requirements, it indicates that there are conflicts among the reference qualification requirements obtained in steps S301 or S302 above. A conflict resolution algorithm based on decision trees can be used to automatically recommend adjustment schemes to determine the adjusted reference qualification sub-requirements corresponding to these conflicting qualification sub-requirements.
[0053] When the number of intersections is less than the preset number of bids and there are conflicts among all the reference qualification sub-requirements, it indicates that the reference qualification requirements obtained in steps S301 or S302 above are both conflicting and too demanding. In this case, for conflicting sub-requirements, a conflict resolution algorithm based on decision trees is used to automatically recommend adjustment schemes. For insufficient intersections, a relaxed matching strategy is adopted to relax non-core qualification sub-requirements, thereby obtaining adjusted reference qualification sub-requirements.
[0054] Finally, the adjusted reference qualification sub-requirements and the compliant reference qualification sub-requirements that do not fall under the adjusted reference qualification sub-requirements are sent to the AI qualification requirement model. This allows the AI qualification requirement model, based on reinforcement learning and historical successful cases, to integrate the adjusted and compliant reference qualification sub-requirements to obtain target qualification requirements that meet both project needs and have high bidding feasibility. This target qualification requirement is then sent to the procurement decision-making platform, providing the platform with a target document template. The compliant reference qualification sub-requirements refer to those sub-requirements that do not correspond to the adjusted reference qualification sub-requirements. By leveraging historical information to send detailed correction plans to the AI model, the model can accurately make corrections, reducing repeated corrections and improving the efficiency and accuracy of determining target qualification requirements. Furthermore, feasibility prediction based on historical data avoids situations where overly high qualifications result in no bids or overly low qualifications lead to quality risks. The combination of structured instructions and AI regeneration achieves closed-loop optimization through human-machine collaboration, significantly improving the accuracy and efficiency of procurement decisions.
[0055] Step S300: Send the target qualification requirements to the AI model to obtain the reference evaluation method, and determine the target evaluation method based on the reference evaluation method.
[0056] Obtaining the aforementioned target qualification requirements indicates that the procurement decision-making platform has completed the work of determining the qualification requirements needed for the bidding process. The next step is to determine the evaluation method for the procurement decision-making process. This involves the procurement decision-making platform sending the obtained target qualification requirements to the AI evaluation method model within the AI model. The AI evaluation method model uses multimodal NLP technology to deeply analyze unstructured information within the target qualification requirements, such as basic qualification requirements (type, presets, etc.), scale qualification requirements (project size, construction period, etc.), professional certification qualification requirements, and personnel configuration qualification requirements. It then extracts key feature values and quantifies them into computable weight factors, while simultaneously identifying implicit rules. This process leverages OCR and semantic understanding technologies to ensure accurate parsing across different file formats.
[0057] In addition, a knowledge graph of the project's attributes is built into a rule library for constructing evaluation methods, covering different scenarios. Specifically, for scale-driven scenarios, the lowest price method is selected for small-scale procurements, while a comprehensive scoring method is matched for large and complex projects, with weights dynamically adjusted according to preset rules. For qualification-oriented scenarios, post-qualification review and two-stage evaluation are triggered for projects with high qualification requirements to reduce invalid bids, and the weight of technical solutions is automatically increased for projects requiring technological innovation capabilities. For risk-adaptive scenarios, competitive negotiation evaluation is automatically recommended when keywords such as emergency disaster relief projects are identified. The AI evaluation method model then sends the obtained reference evaluation methods to the procurement decision-making platform, providing the platform with reference evaluation methods. The procurement decision-making platform then reviews and corrects the obtained reference evaluation methods to obtain the target evaluation method. The target reference evaluation method includes several reference sub-methods, each corresponding to a specific evaluation weight. Determining the target evaluation method based on the reference evaluation methods includes the following steps: Step S301: Sort the reference evaluation methods in descending order of evaluation weight to obtain a sequence of reference evaluation methods.
[0058] Step S302: Determine whether the reference evaluation sub-method sequence conforms to the preset sequence. If it does, determine the reference evaluation method as the target evaluation method.
[0059] Step S303: If it does not meet the requirements, generate an adjusted evaluation method based on the preset sequence, and send the adjusted evaluation method to the AI model to obtain the target evaluation method.
[0060] Specifically, all the aforementioned reference evaluation sub-methods can be sorted to obtain an ordered sequence of reference evaluation sub-methods. Then, this sequence is compared with a preset sequence to determine if the sequential relationships between each reference evaluation sub-method in the reference evaluation sub-method sequence are consistent with the preset sequential relationships in the preset sequence. If they are consistent, then the reference evaluation sub-method sequence conforms to the preset sequence, and the reference evaluation method obtained by the AI evaluation method model is accurate. In this case, the reference evaluation method is determined as the target evaluation method.
[0061] If all sub-methods are not consistent, then the sequence of reference evaluation methods does not conform to the preset sequence, and the reference evaluation method obtained by the AI evaluation method model is inaccurate. In this case, an adjusted evaluation method can be generated based on the preset sequence, specifying which reference evaluation sub-methods require weight adjustments. This adjusted evaluation method is then sent to the AI evaluation method model. The preset sequence is determined by the bidding professionals based on the latest requirements. After receiving the adjusted evaluation method, the AI evaluation method model adjusts the reference evaluation method accordingly to obtain the target evaluation method, which is then sent to the procurement decision-making platform. This allows the platform to obtain the target evaluation method. By leveraging historical information to send detailed correction plans to the AI model, the model can make accurate corrections, reducing the number of repeated corrections and improving the efficiency and accuracy of determining the target evaluation method.
[0062] Specifically, the above-mentioned adjusted evaluation method can adopt either a weight redistribution strategy based on constraint satisfaction or a case-driven type adjustment strategy.
[0063] The core of the constraint-based weight redistribution strategy is to treat the preset sequence as a hard constraint that must be satisfied. An optimization algorithm is then used to find a new set of weights that maximizes the conformity of the preset sequence with the new weights. First, an optimization problem is constructed, with the weights of each method as variables, and the objective is to ensure that the descending order of the weight sequence matches the preset sequence. An optimization model is then established. Next, constraints are introduced during the optimization process, such as the weights of core methods not falling below a certain minimum value, and the weights of non-core methods not exceeding a certain maximum value. Finally, a solution and generation process is employed, using optimization algorithms such as gradient descent and genetic algorithms to obtain an optimal weight combination that satisfies both the preset sequence order and various business constraints. This optimal combination is then used to generate the adjusted evaluation method.
[0064] The core of a case-driven analogy adjustment strategy is to utilize data from historical successful projects, matching similarity to find the most similar reference cases for the current project, and then referencing their weighting schemes. First, a case library is built, storing the project attributes, evaluation method sequences, and weights of each sub-method for historical projects. Then, similarity matching is performed, calculating the similarity between the current project's attributes and historical projects in the case library, identifying the K most similar historical projects. Finally, adjustment suggestions are generated by aggregating and analyzing the weighted data of these K historical projects, such as by taking a weighted average, and using the weights of each sub-method as adjustment suggestions for the current project, thereby generating a data-driven, more reliable adjustment evaluation method.
[0065] This automated decision-making process, based on optimization algorithms or data, involves experts formulating high-level strategic rules, while AI performs complex, quantitative calculations and optimizations. This collaborative approach surpasses traditional methods that rely solely on AI or solely on human intervention, significantly improving the accuracy, efficiency, and objectivity of decision-making.
[0066] Step S400: Send the target evaluation method to the AI model to obtain reference relevant elements, and determine the target relevant elements based on the reference relevant elements.
[0067] Obtaining the aforementioned target evaluation method indicates that the procurement decision-making platform has completed the work of determining the evaluation method required for the bidding process. The next step is to further determine the relevant elements for the procurement decision-making process. Specifically, the procurement decision-making platform sends the obtained target evaluation method to the AI-related element model within its AI model. Upon receiving the target evaluation method, the AI-related element model automatically generates reference elements, including the clarification deadline, bid opening time, and bid security amount, based on relevant policy documents. The AI-related element model then sends these reference elements to the procurement decision-making platform, enabling the platform to obtain the relevant reference elements. The procurement decision-making platform then reviews and corrects these reference elements to obtain the target relevant elements. The process of determining the target relevant elements based on the reference relevant elements includes the following steps: Step S401: Determine the evaluation criteria based on the supplementary qualification requirements and the preset sequence, and determine whether the reference related elements meet the evaluation criteria. If they do, determine the reference related elements as the target related elements.
[0068] Step S402: If it does not meet the criteria, determine the adjustment element information that needs to be adjusted based on the evaluation element standard, and send the adjustment element information to the AI model to obtain the target related elements.
[0069] The supplementary qualification requirements are substituted into a preset correspondence between information and relevant elements to obtain the evaluation criteria for the supplementary qualification requirements. The preset correspondence between information and relevant elements was determined through extensive offline testing. The evaluation criteria are based on the relevant elements corresponding to the additional qualification requirements provided by the bidding staff. Then, the reference relevant elements obtained from the AI-related elements are compared with the evaluation criteria obtained above. If the reference relevant elements meet the evaluation criteria, it indicates that the reference relevant elements obtained from the AI-related elements are accurate and contain the relevant elements required for this bidding process; therefore, the reference relevant elements are determined as the target relevant elements.
[0070] If the reference relevant elements do not meet the evaluation element criteria, it indicates that the reference relevant elements obtained by the AI are inaccurate. In this case, the procurement decision-making platform needs to make adjustments. This involves comparing the evaluation element criteria with the reference relevant elements to find the elements that the evaluation element criteria include more of than the reference relevant elements, and identifying these additional elements as adjustment element information. This adjustment element information is then sent to the AI relevant element model, allowing it to supplement the reference relevant elements to obtain the target relevant elements.
[0071] Subsequently, the AI-based relevant element model sends the obtained target-related elements to the procurement decision-making platform, enabling the platform to obtain these elements. This detailed correction plan is then sent to the AI model, facilitating accurate adjustments, reducing redundant corrections, and thus improving the efficiency and accuracy of determining the target evaluation method. Specifically, the AI model's correction process includes the following steps.
[0072] First, the AI model performs multi-dimensional conflict detection between the received adjustment element information and the original reference elements. The detection scope includes temporal logic conflicts, numerical logic conflicts, and business rule conflicts. Once a conflict is detected, the model will prioritize automatic resolution based on its built-in knowledge graph of bidding and tendering laws and implementation rules, as well as its historical success case database. For example, if the margin ratio is found to be excessive, the model will automatically correct it to the legal upper limit; if a temporal logic error is detected, the model will automatically adjust the relevant time nodes to ensure the rationality of the timeline.
[0073] After ensuring there are no conflicts, the AI model performs context-aware element fusion, that is, it understands the semantic role and position of each element in the tender document. The model utilizes a pre-trained tender document structure understanding model to accurately populate the adjusted element information into the corresponding structured fields in the relevant reference elements. This process employs semantic slot filling technology to ensure that the new elements and the context statements are syntactically and semantically seamlessly connected, generating fluent and professional text.
[0074] Next, the AI model incorporates a dynamically updated business rule base, which links the value rules of project attributes and related elements. While integrating new elements, the model again uses project initiation information and target evaluation methods as input to trigger matching within the rule base. For example, when the project attribute is identified as emergency disaster relief, even if the adjusted element information does not mention it, the model will automatically compress the announcement time to the shortest time limit allowed by policy and recommend element templates corresponding to competitive negotiations, achieving intelligent adjustment between elements.
[0075] Finally, after completing the conflict resolution, context fusion, and rule matching processes, the AI model generates a complete set of target-related elements with consistent internal logic and returns it to the procurement decision-making platform. This output is not merely a simple list of elements, but a structured data object that clearly identifies the type, value, document section, and associated business rule number of each element, facilitating subsequent processing and display by the procurement decision-making platform.
[0076] This multi-stage intelligent correction process, combining conflict detection and resolution, context-aware fusion, and dynamic rule matching, achieves autonomous correction under complex logic, reducing human-computer interaction and improving overall decision-making efficiency. Furthermore, conflict detection based on knowledge graphs and rule bases ensures that the generated target-related elements are consistent and compliant in terms of time, values, and business logic, thus reducing legal risks in bidding documents from the outset.
[0077] Step S500: Send the target-related elements to the AI model to obtain the reference tender document, and determine the target tender document based on the reference tender document.
[0078] Obtaining the aforementioned target-related elements indicates that the procurement decision-making platform has completed the task of determining the relevant elements required for the bidding process. The next step is to further determine the bidding documents for the procurement decision-making process. Specifically, the procurement decision-making platform sends the obtained target-related elements to the AI bidding document model within the AI model. The AI bidding document model automatically generates the bidding documents based on the target document template, target qualification requirements, target evaluation methods, and target-related elements. Then, the AI bidding document model sends the obtained reference bidding documents to the procurement decision-making platform, providing it with a reference bidding document. The procurement decision-making platform then reviews and corrects the obtained reference bidding documents to obtain the target bidding documents. The process of determining the target bidding documents based on the reference bidding documents includes the following steps: Step S501: Perform structured parsing on the reference tender document to obtain non-fixed clause fields and key parameter fields.
[0079] Step S502: Perform multimodal matching of the target investment requirements, target evaluation method, and target-related elements with non-fixed clause fields to generate clause filling content.
[0080] Step S503: Based on the preset knowledge graph of tender document compliance, perform compliance verification on key parameter fields.
[0081] Step S504: If the verification finds that the document is compliant, the target tender document will be determined by referring to the tender document.
[0082] Step S505: If the verification finds that the project does not comply with the requirements, retrieve the replacement parameters that match the project attributes from the historical bidding documents, update the key parameter fields with the replacement parameters to obtain the replacement key fields, and send the replacement key fields and the clause filling content to the AI model to obtain the target bidding documents.
[0083] The procurement decision-making platform performs deep structured analysis on the reference tender documents received from the AI-powered tender document model. This analysis process utilizes a pre-trained tender document structure model, such as those based on OCR and layout recognition technology, to decompose the reference tender documents into different logical units. Then, within each logical unit, two types of fields are identified and categorized: non-fixed clause fields and key parameter fields.
[0084] Next, a multimodal matching and filling process will be initiated. The previously determined target investment requirements, target evaluation methods, and target-related elements will be used as input. Through semantic similarity calculation, such as using models like Sentence-BERT to generate vectors and calculate cosine similarity, and with preset mapping rules, the insertion position will be found in the non-fixed clause fields of the reference bidding documents, and the corresponding target investment requirements, target evaluation methods, and target-related elements will be filled in. For example, the target investment requirements will be replaced with the placeholders corresponding to the bidder qualification requirements.
[0085] Simultaneously, a compliance check based on a knowledge graph will be initiated. This involves constructing a large-scale knowledge graph of tender document compliance. Nodes in this graph include, but are not limited to, specific clauses of laws and regulations such as the Bidding and Tendering Law and the Procurement Law; different project types, such as municipal engineering, IT services, and goods procurement; and various parameters, such as the bid bond ratio and the number of days for the public notice period. Nodes are connected by edges representing constraints, applicability, and conflicts. Matching and reasoning with the parsed key parameter fields against the knowledge graph will locate the current project type node, traverse its related legal and regulatory constraint nodes, and check whether the parameter value is within the allowed range. If a parameter conflict is found, such as identifying a key parameter value exceeding the legally defined range in the knowledge graph, or logical conflicts between different parameters, or a registration deadline later than the bid opening time, it will not simply report an error. Instead, it will automatically retrieve cases with similar attributes and scale to the current project from the stored historical tender document database, extract their corresponding compliance parameter values as alternative suggestions, and automatically replace them, thus achieving self-correction of the problem.
[0086] Finally, structured reorganization and output are performed. All the filled information is reassembled according to the standard logical structure of the tender document to obtain the target tender document. In this way, the procurement decision-making platform can accurately correct the reference tender document on its own, reducing the number of information interactions with the AI model and thus indirectly achieving energy savings.
[0087] Step S600: Send the target tender document to the AI model to obtain the review result of the target tender document. If the review result is that the review is passed, encrypt the target tender document to obtain the tender decision.
[0088] Step S700: If the audit result is that the audit fails, adjust the target tender document to obtain a new target tender document.
[0089] Obtaining the aforementioned target tender documents indicates that the procurement decision-making platform has completed the process of determining the tender documents required for the bidding process. The next step is to further verify the documents for procurement decision-making. Specifically, the procurement decision-making platform sends the obtained target tender documents to the AI document verification model within its AI framework. The document verification model reviews the document content based on relevant laws and regulations and project information to obtain a verification result, which is then sent to the procurement decision-making platform, thus providing the platform with the verification outcome.
[0090] If the review result is approved, the required procurement decision is obtained. To further ensure the accurate publication of the procurement decision, it is encrypted during transmission and then decrypted upon arrival at the designated address for publication. The encryption and decryption methods can be determined based on the actual situation, and no further limitations are specified here.
[0091] If the review result is unsuccessful, the target review document is returned to the logic in step S500 above, where the procurement decision-making platform determines the target bidding document based on the reference bidding document, to adjust the target bidding document and obtain a new one. This AI-based procurement decision-making method is applied to the procurement decision-making platform. The platform is linked to an AI model, which uses its artificial intelligence to initially determine some information, thereby reducing the workload of the electronic procurement platform and improving its efficiency. Simultaneously, the procurement decision-making platform also performs corresponding review processing on the information obtained from the AI model, enabling it to quickly obtain accurate bidding and procurement decisions with minimal computation.
[0092] Preferably, this application also obtains the revised document template corresponding to the target tender document, sends the revised document template to the AI model for storage, enriches the document templates stored in the AI model, and facilitates obtaining more accurate document templates when making other procurement decisions in the future. This reduces the modification and adjustment work of the procurement decision platform in determining the target document template based on the reference document template, thereby indirectly improving work efficiency.
[0093] Preferably, in this application, the procurement decision-making platform also obtains offline bidding documents and sends them to the AI model to obtain the review result. If the review result is "approved," the target bidding document is encrypted to obtain the procurement decision. If the review result is "failed," the offline bidding documents are adjusted to obtain new offline bidding documents. Specifically, bidding staff can also send the already finalized offline bidding documents (i.e., offline bidding documents) to the procurement decision-making platform, allowing the platform to obtain the offline bidding documents and then directly send them to the AI model for review to obtain the review result. This supports multiple forms of procurement decision-making and improves the diversity of procurement decision-making processes.
[0094] Figure 3 This is a flowchart illustrating the AI-based procurement decision-making process provided in this application. Figure 3 As shown, the procurement decision-making platform first receives project initiation information, then sends this information to the AI model via a push interface. The AI model selects a document template and pushes a reference document template to the procurement decision-making platform. Upon receiving the reference document template, the procurement decision-making platform reviews it to determine the target document template and sends it to the AI model. The AI model, upon receiving the target document template, generates reference qualification requirements and returns them to the procurement decision-making platform. The procurement decision-making platform, upon receiving the reference qualification requirements, reviews them to determine the target qualification requirements and sends them to the AI model. The AI model, upon receiving the target qualification requirements, generates reference evaluation methods and returns them to the procurement decision-making platform. The procurement decision-making platform, upon receiving the reference evaluation methods, reviews them to determine the target evaluation methods and sends them to the AI model. The AI model, upon receiving the target evaluation methods, generates reference related elements and returns them to the procurement decision-making platform. The procurement decision-making platform, upon receiving the reference related elements, reviews them to determine the target related elements and sends them to the AI model. The AI model, upon receiving the target related elements, generates a reference tender document and returns it to the procurement decision-making platform. After receiving the reference tender documents, the procurement decision-making platform reviews them to determine the target tender documents, which are then sent to the AI model. The AI model receives the target tender documents, reviews them, generates a review result, and returns it to the procurement decision-making platform. If the review result is unsatisfactory, the process returns to the tender document review step to re-determine the target tender documents. If the review result is satisfactory, the AI model generates structured data for the tender documents and returns both the tender documents and structured data to the procurement decision-making platform, enabling the platform to generate encrypted tender documents and publish them.
[0095] The procurement decision-making platform receives offline, pre-edited tender documents and sends them to an AI model for review. The model generates a review result, which is then returned to the procurement decision-making platform. Further processing is then based on this result. This AI-based procurement decision-making method is applied to the platform. The platform is linked to an AI model, which uses its artificial intelligence to initially determine certain information, thereby reducing the workload of the electronic procurement platform and improving its efficiency. Simultaneously, the platform also performs review and processing on the information obtained from the AI model, allowing it to quickly and accurately make bidding and procurement decisions with minimal computation.
[0096] Figure 4 This is a schematic diagram illustrating the connection of an artificial intelligence-based procurement decision-making system provided in an embodiment of this application. This system is applied in a procurement decision-making platform. Figure 4 As shown, an artificial intelligence-based procurement decision-making system includes: a document template module, a qualification requirements module, a bid evaluation method module, a relevant element module, a tender document module, and a document review module.
[0097] The system comprises several modules: a document template module, a qualification requirements module, and a document review module. The document template module retrieves project initiation information, sends it to the AI model to obtain reference document templates, and determines the target document template based on these templates. The qualification requirements module sends the target document template to the AI model to obtain reference qualification requirements, and determines the target qualification requirements based on these templates. The evaluation method module sends the target qualification requirements to the AI model to obtain reference evaluation methods, and determines the target evaluation method based on these methods. The relevant elements module sends the target evaluation method to the AI model to obtain reference relevant elements, and determines the target relevant elements based on these elements. The tender document module sends the target relevant elements to the AI model to obtain reference tender documents, and determines the target tender document based on these reference documents. The document review module sends the target tender document to the AI model to obtain the review result; if the review result is "approved," the target tender document is encrypted to obtain the procurement decision. The tender document module also adjusts the target tender document to obtain a new one if the review result is "failed."
[0098] The other functions performed by the above-mentioned document template module, qualification requirements module, bid evaluation method module, relevant element module, tender document module, and document review module, as well as the technical details of each function, are the same as or similar to the corresponding features in the artificial intelligence-based procurement decision-making method described above, so they will not be repeated here.
[0099] This application also provides a computer storage medium storing a computer program that, when run on a computer, enables the computer to execute the steps in the artificial intelligence-based procurement decision-making method described above.
[0100] It should be understood that although the steps in the flowcharts in the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order requirement for the execution of these steps, and they can be performed in other orders.
[0101] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.
Claims
1. A procurement decision-making method based on artificial intelligence, characterized in that, The method includes: Obtain project initiation information, send the project initiation information to the AI model to obtain a reference document template, and determine the target document template based on the reference document template; The target file template is sent to the AI model to obtain reference qualification requirements, and the target qualification requirements are determined based on the reference qualification requirements. The target qualification requirements are sent to the AI model to obtain a reference evaluation method, and the target evaluation method is determined based on the reference evaluation method. The target evaluation method is sent to the AI model to obtain reference relevant elements, and the target relevant elements are determined based on the reference relevant elements. The target-related elements are sent to the AI model to obtain a reference tender document, and the target tender document is determined based on the reference tender document. The target tender document is sent to the AI model to obtain the review result of the target tender document. If the review result is that the review is passed, the target tender document is encrypted to obtain the procurement decision. If the audit result is that the audit fails, the target tender document shall be adjusted to obtain a new target tender document.
2. The method according to claim 1, characterized in that, The project initiation information includes project attributes, and determining the target file template based on the reference file template includes: Extract reference guide words from the reference document template, wherein the reference guide words include several reference guide sub-words; Determine whether all referenced guiding subkeys belong to the project attribute; if so, determine the referenced file template as the target file template. If not, obtain abnormal reference guide words that do not belong to the project attributes, and retrieve the correction reference guide words corresponding to the abnormal reference guide words from the preset historical file template library; The corrected reference guide words, the abnormal reference guide words, and the reference file template are sent to the AI model to obtain the target file template.
3. The method according to claim 1, characterized in that, The reference qualification requirements include several reference qualification sub-requirements, and determining the target qualification requirements based on the reference qualification requirements includes: Determine whether a supplementary qualification requirement has been received. If so, perform semantic fusion on the supplementary qualification requirement and add it to the reference qualification requirement to obtain a new reference qualification requirement. If not, the aforementioned reference qualification requirements shall be used as the new reference qualification requirements; Based on historical bidding information, obtain the reference compliant bidding information corresponding to each reference qualification sub-requirement, and determine the number of intersections of all reference compliant bidding information; Determine whether there are no conflicts among all the reference qualification sub-requirements and whether the number of intersections is not less than the preset number of bids. If so, determine the reference qualification requirement as the target qualification requirement. If not, obtain the adjusted reference qualification sub-requirements corresponding to the number of conflicting reference qualification sub-requirements and / or the number of intersections, and send the adjusted reference qualification sub-requirements and the compliant reference qualification sub-requirements that do not belong to the adjusted reference qualification sub-requirements to the AI model to obtain the target qualification requirements.
4. The method according to claim 3, characterized in that, The reference evaluation method includes several reference evaluation sub-methods, and each reference evaluation sub-method corresponds to an evaluation weight. The step of determining the target evaluation method based on the reference evaluation method includes: The reference bid evaluation sub-methods are sorted in descending order of their bid evaluation weights to obtain a sequence of reference bid evaluation sub-methods. Determine whether the reference evaluation method sequence conforms to the preset sequence; if it does, determine the reference evaluation method as the target evaluation method. If it does not meet the requirements, an adjusted evaluation method is generated based on the preset sequence, and the adjusted evaluation method is sent to the AI model to obtain the target evaluation method.
5. The method according to claim 4, characterized in that, The determination of target relevant elements based on the reference relevant elements includes: Based on the supplementary qualification requirements, the evaluation criteria are determined, and it is determined whether the reference relevant elements meet the evaluation criteria. If they do, the reference relevant elements are determined as target relevant elements. If it does not meet the criteria, the adjustment element information that needs to be adjusted for the reference related elements is determined based on the evaluation element criteria, and the adjustment element information is sent to the AI model to obtain the target related elements.
6. The method according to claim 5, characterized in that, The determination of the target tender document based on the reference tender document includes: The reference tender document is structured and parsed to obtain non-fixed clause fields and key parameter fields; The target qualification requirements, target evaluation methods, and target-related elements are matched with the non-fixed clause fields using multimodal matching to generate clause-filling content; Based on a pre-defined knowledge graph of tender document compliance, the key parameter fields are validated for compliance. If the verification finds that the reference tender document is compliant, the reference tender document will be determined as the target tender document; If the verification finds that the project does not meet the requirements, the replacement parameters that match the project attributes are retrieved from the historical bidding documents. The replacement parameters are used to update the key parameter field to obtain the replacement key field. The replacement key field and the clause fill content are sent to the AI model to obtain the target bidding document.
7. The method according to claim 1, characterized in that, The method further includes: Obtain the revised document template corresponding to the target tender document, and send the revised document template to the AI model for storage.
8. The method according to claim 1, characterized in that, The method further includes: Obtain offline bidding documents, send the offline bidding documents to the AI model to obtain the review result of the offline bidding documents, and if the review result is that the review is passed, encrypt the target bidding documents to obtain the procurement decision; If the audit result is that the audit fails, the offline bidding documents will be adjusted to obtain new offline bidding documents.
9. A procurement decision-making system based on artificial intelligence, characterized in that, The system is used in a procurement decision-making platform and includes: a document template module, a qualification requirements module, a bid evaluation method module, a relevant element module, a tender document module, and a document review module; among which... The document template module is used to obtain project initiation information, send the project initiation information to the AI model to obtain a reference document template, and determine the target document template based on the reference document template. The qualification requirements module is used to send the target file template to the AI model to obtain reference qualification requirements, and to determine the target qualification requirements based on the reference qualification requirements; The bid evaluation method module is used to send the target qualification requirements to the AI model to obtain a reference bid evaluation method, and determine the target bid evaluation method based on the reference bid evaluation method. The relevant elements module is used to send the target evaluation method to the AI model to obtain reference relevant elements, and to determine the target relevant elements based on the reference relevant elements; The tender document module is used to send the target-related elements to the AI model to obtain a reference tender document, and to determine the target tender document based on the reference tender document. The document review module is used to send the target tender document to the AI model to obtain the review result of the target tender document. If the review result is that the review is passed, the target tender document is encrypted to obtain the procurement decision. The tender document module is also used to adjust the target tender document to obtain a new target tender document if the review result is that the review is not passed.
10. A computer-readable storage medium having a computer program stored thereon that can run on a processor, characterized in that, When the computer program is executed by the processor, it implements an artificial intelligence-based procurement decision-making method as described in any one of claims 1 to 8.
Citation Information
Patent Citations
Intelligent bidding document generation method and system, computer device and storage medium
CN115760258A
Method for realizing automatic bid inviting and purchasing for engineering analysis
CN118551937A
Automatic bid invitation file auditing method and system based on artificial intelligence
CN119624385A
Autonomous bidder solicitation and selection system
US20210304297A1
Cited By
A method and system for intelligent matching and recommendation of bidding information
CN122388269A
