A tender auxiliary checking method, device and equipment

Abstract

The present application relates to the field of data processing, and particularly relates to a bidding document auxiliary checking method, device and equipment. The method comprises the following steps: extracting key information of each bidding document in historical bidding document data to obtain a key information set of each bidding document; calculating the performance importance of each key information; calculating a core embodiment factor of each key information in each bidding document according to the performance importance of each key information; calculating the clustering distance between each bidding document and any other bidding document according to the core embodiment factor of each key information, taking the clustering distance as a clustering parameter of a fuzzy C-means clustering method, using the fuzzy C-means clustering method to cluster the bidding documents in the historical bidding document data to obtain a clustering result; using the clustering result to train a neural network to obtain a bidding document auxiliary model, inputting a target bidding document into the bidding document auxiliary model, and outputting a target auxiliary checking reference document through the bidding document auxiliary model. The present application can improve the efficiency of bidding document checking.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically to a method, apparatus, and equipment for assisting in the verification of tender documents. Background Technology

[0002] The tender document verification system based on a generative large model is a tool that uses artificial intelligence technology to improve the quality and efficiency of tender document preparation. Through intelligent analysis and automatic checking functions, it helps companies reduce errors and risks in tender documents and increase the success rate. It collects historical tender document data to train a generative large model for the tender document verification task.

[0003] Different types of tender documents have their own exclusive terminology, rules, and verification focuses (e.g., engineering tender documents emphasize construction specifications, while service tender documents emphasize delivery standards). Tender document classification before training the generative large-scale model for tender document verification tasks allows the model to learn specific domain patterns from similar data, avoiding "generalization bias" caused by mixed data. However, in reality, there are many "hybrid tender documents" that fall between two or more categories. Due to the complexity of project requirements, these tender documents may contain content modules from different domains simultaneously. For example, road construction tenders may include engineering content (road construction plans, subgrade treatment processes, construction schedules, etc.) and goods content (procurement of building materials such as asphalt and cement, as well as engineering equipment such as road rollers and pavers). This can lead to significant classification errors in the tender documents, thus affecting the training effect of the generative large-scale model for tender document verification tasks and resulting in larger errors in tender document verification. Summary of the Invention

[0004] This invention provides a method, apparatus, and equipment for assisting in the verification of tender documents, in order to solve existing problems.

[0005] In a first aspect, one embodiment of the present invention provides a method for assisting in the verification of tender documents, the method comprising the following steps:

[0006] Extract key information from each bid in the historical bid data to obtain a key information set for each bid. Each key information in each key information set includes the key information order and the initial tag.

[0007] Calculate the performance importance of each key piece of information in each tender document within the key information set;

[0008] Based on the perceived importance of each key piece of information, calculate the core manifestation factor of each key piece of information in each tender document;

[0009] Based on the core representation factors of each key information in each tender document, the clustering distance of each tender document with any other tender document is calculated. The clustering distance is used as the clustering parameter of the fuzzy C-means clustering method. The fuzzy C-means clustering method is used to cluster the tender documents in the historical tender document data to obtain the clustering results.

[0010] The clustering results are used to train the neural network to obtain the tender document auxiliary model. The target tender document is input into the tender document auxiliary model, and the target auxiliary verification reference document is output by the tender document auxiliary model.

[0011] Optionally, the performance importance of each key piece of information in each tender document within the key information set is calculated, specifically including:

[0012] The target bid is identified from historical bid data, the target key information set of the target bid is identified from the key information set, the target key information is identified from the target key information set, and the semantic similarity of the target key information with other key information is calculated. The other key information refers to the key information in the target key information set other than the target key information.

[0013] Based on semantic similarity, reference key information, non-reference key information, and verification key information are determined from the target key information set. Based on the verification key information, reference key information, and non-reference key information, the information high-value specificity and information penetration carrying capacity of the target key information are calculated.

[0014] The performance importance of the target key information is determined by the product of its informational value specificity and its informational pervasiveness.

[0015] Each key piece of information in the set of key target information is identified as a key target information, and the performance importance of each key piece of information is obtained separately.

[0016] Each bid document in the historical bid document data is identified as a target bid document, and the performance importance of each key piece of information in each bid document is obtained.

[0017] Optionally, based on the performance importance of each key piece of information, the core manifestation factor of each key piece of information in each tender document is calculated, specifically including:

[0018] Determine a set of key information for comparison from the set of key information, and calculate the semantic similarity between each key information in the set of key information for comparison and the target key information.

[0019] Key information with semantic similarity greater than a preset second similarity threshold is identified as key information for mapping.

[0020] When comparing tender documents containing key information related to the target tender, the tender documents are identified as relevant historical tender documents of the target tender.

[0021] Based on the performance importance of each key piece of information, calculate the degree of correlation between each relevant historical tender and the target key information;

[0022] Calculate the initial specific novelty of the target key information based on the degree of correlation between each relevant historical tender and the target key information;

[0023] Based on the initial tags of the target bid, identify the main relevant historical bid and the sub-relevant historical bid from the relevant historical bids, and calculate the type specificity of the target key information based on the main relevant historical bid and the sub-relevant historical bid.

[0024] When the comparative tender does not contain the mirroring key information of the target key information, the comparative tender is identified as an irrelevant historical tender of the target tender, and the initial special novelty of the target key information is set to the first preset value. The comparative tender is the tender corresponding to the set of comparative key information in the historical tender data.

[0025] Based on the initial specificity and type specificity of the target key information, calculate the core embodiment factor of each key information in each tender document.

[0026] Optionally, reference key information, non-reference key information, and verification key information are determined from the target key information set based on semantic similarity, specifically including:

[0027] Other key information with semantic similarity greater than a preset first similarity threshold is identified as reference key information;

[0028] Other key information with semantic similarity less than or equal to a preset first similarity threshold is identified as non-reference key information;

[0029] The target key information and reference key information are combined to obtain the verification key information.

[0030] Optionally, based on the key information examined, the reference key information, and the non-reference key information, the information value specificity and information pervasiveness of the target key information are calculated, specifically including:

[0031] The quantity of key information obtained for inspection;

[0032] The ratio of the number of key information items to the number of key information items in the target key information set is determined as the proportion of key information items to be inspected.

[0033] Obtain the mean of the semantic similarity between the target key information and each reference key information to get the first mean;

[0034] The mean of the semantic similarity between the target key information and each non-reference key information is obtained to obtain the second mean.

[0035] The difference between the first mean and the second mean is determined as the feature difference of the target key information;

[0036] The product of the proportion of key information and the feature difference is determined as the information high-value specificity of the target key information.

[0037] Based on the key information obtained from the inspection, the information permeability of the target key information is obtained.

[0038] Optionally, based on the key information examined, the information permeability of the target key information is obtained, specifically including:

[0039] The key information sequence of each key information in the test is obtained sequentially to obtain the ordinal value sequence;

[0040] Based on the ordinal value sequence, obtain the interval value sequence, and determine the target interval value sequence from the interval value sequence;

[0041] Based on the target interval value sequence, calculate the information penetration carrying capacity of key target information.

[0042] Optionally, based on the main relevant historical tender documents and the sub-relevant historical tender documents, the type specificity of the target key information is calculated, specifically including:

[0043] When relevant historical bids contain key information about the target, the type specificity of the key information about the target is calculated based on the degree of association between each relevant historical bid and the key information about the target.

[0044] When the relevant historical tender documents do not contain the main relevant historical tender documents containing the key information of the target, the type specificity of the key information of the target is set to the second preset value.

[0045] Optionally, based on the initial specificity and type specificity of the target key information, the core embodiment factor of each key information in each tender document is calculated, specifically including:

[0046] Based on the initial specificity and type specificity of the target key information, calculate the specificity of the target key information, and based on the specificity of the target key information, calculate the core embodiment factor of the target key information;

[0047] Each key piece of information in the set of key target information is identified as the key target information, and the core manifestation factors of each key piece of information are obtained respectively.

[0048] Each bid document in the historical bid document data is identified as a target bid document, and the core manifestation factors of each key information in each bid document are obtained.

[0049] Secondly, one embodiment of the present invention provides a tender document auxiliary verification device, the device comprising:

[0050] The data acquisition unit is used to extract key information from each bid in the historical bid data to obtain a key information set for each bid. Each key information in each key information set includes the key information order and the initial label.

[0051] The data processing unit is used to calculate the performance importance of each key information in each tender document within the key information set; calculate the core manifestation factor of each key information in each tender document based on the performance importance of each key information; calculate the clustering distance between each tender document and any other tender document based on the core manifestation factor of each key information in each tender document; use the clustering distance as the clustering parameter of the fuzzy C-means clustering method; and use the fuzzy C-means clustering method to cluster the tender documents in the historical tender document data to obtain the clustering results.

[0052] The auxiliary verification unit is used to train the neural network using the clustering results to obtain the tender document auxiliary model. The target tender document is input into the tender document auxiliary model, and the target auxiliary verification reference document is output through the tender document auxiliary model.

[0053] Thirdly, the present invention also provides an electronic device comprising a processor, the processor being configured to execute a computer program stored in a memory to implement the steps of any of the above-described tender document verification methods.

[0054] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described tender document verification methods.

[0055] Fifthly, the present invention provides a computer program product, comprising a computer program: when the computer program is executed by a processor, it implements the tender document verification method as described in the first aspect above.

[0056] The beneficial effects of the technical solution of the present invention are:

[0057] In this embodiment of the invention, by analyzing the performance importance, specificity, and novelty of each key information in the key information set of historical tender documents, core reflective factors are determined and used as weights. This ensures the accuracy of historical tender document classification. After classification, the model can learn specific domain patterns from similar data, avoiding "generalization bias" caused by mixed data. For example, when processing commercial tender data, the model can more accurately grasp the pricing calculation logic and qualification document verification rules; when processing technical tender data, it can gain a deeper understanding of the technical parameter matching requirements of different industries. Ultimately, the powerful generation and understanding capabilities of the large language model are utilized to improve the efficiency of tender document verification. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, 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.

[0059] Figure 1 A flowchart illustrating a tender document verification method provided in one embodiment of the present invention;

[0060] Figure 2 This is a schematic diagram of the structure of a tender document verification device provided in one embodiment of the present invention;

[0061] Figure 3 This is a schematic diagram of a tender document verification device provided in one embodiment of the present invention. Detailed Implementation

[0062] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a tender document verification method proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0063] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0064] The following description, in conjunction with the accompanying drawings, details the specific scheme of the tender document auxiliary verification method provided by the present invention.

[0065] This invention provides a method, apparatus, and device for auxiliary verification of tender documents. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of a tender document verification method provided by an embodiment of the present invention, which includes the following steps:

[0066] S101. Extract the key information of each bid from the historical bid data to obtain a key information set for each bid. Each key information in each key information set includes the key information order and the initial label.

[0067] For example, historical bid data may include bids from a predetermined time period in the past. Historical bids may include successful and unsuccessful bids, as well as relevant tender documents, laws and regulations, industry standards, and other materials. Each historical bid is the same as each bid in the historical bid data.

[0068] Then, Natural Language Processing (NLP) technology (a well-known technique) is used to parse each historical tender document, extracting key information (such as project name, technical requirements, evaluation criteria, commercial terms, delivery date, etc.) to form a key information set for each historical tender document. Following the order of reading, each key information in the key information set is assigned a key information order (e.g., sequence number 1, 2, 3, 4, ...).

[0069] Finally, a manual tagging method can be used to assign an initial tag to each historical bid based on its name.

[0070] For example, in a specific embodiment, the initial tags are three common types: "Goods Procurement," "Services," and "Engineering Construction." For instance, the initial tag for a tender document titled "Tender Document from XX Technology Co., Ltd. for the Procurement of Smart Campus Servers and Supporting Software for XX School" is "Goods Procurement." The initial tag for a tender document titled "Tender Document from XX Advertising Media Co., Ltd. for the 2024 Urban Tourism Image Promotion and Publicity Service Project of XX Municipal Bureau of Culture and Tourism" is "Services." The initial tag for a tender document titled "Tender Document from XX Municipal Engineering Co., Ltd. for the Construction of Standard Factory Buildings and Supporting Infrastructure in XX Industrial Park" is "Engineering Construction."

[0071] S102. Calculate the performance importance of each key information in each tender document in the key information set.

[0072] In this embodiment, calculating the performance importance of each key piece of information in each tender document within the key information set specifically includes:

[0073] The target bid is identified from historical bid data, the target key information set of the target bid is identified from the key information set, the target key information is identified from the target key information set, and the semantic similarity of the target key information with other key information is calculated. The other key information refers to the key information in the target key information set other than the target key information.

[0074] Based on semantic similarity, reference key information, non-reference key information, and verification key information are determined from the target key information set. Based on the verification key information, reference key information, and non-reference key information, the information high-value specificity and information penetration carrying capacity of the target key information are calculated.

[0075] The performance importance of the target key information is determined by the product of its informational value specificity and its informational pervasiveness.

[0076] Each key piece of information in the set of key target information is identified as a key target information, and the performance importance of each key piece of information is obtained separately.

[0077] Each bid document in the historical bid document data is identified as a target bid document, and the performance importance of each key piece of information in each bid document is obtained.

[0078] Optionally, reference key information, non-reference key information, and verification key information are determined from the target key information set based on semantic similarity, specifically including:

[0079] Other key information with semantic similarity greater than a preset first similarity threshold is identified as reference key information;

[0080] Other key information with semantic similarity less than or equal to a preset first similarity threshold is identified as non-reference key information;

[0081] The target key information and reference key information are combined to obtain the verification key information.

[0082] Optionally, based on the key information examined, the reference key information, and the non-reference key information, the information value specificity and information pervasiveness of the target key information are calculated, specifically including:

[0083] The quantity of key information obtained for inspection;

[0084] The ratio of the number of key information items to the number of key information items in the target key information set is determined as the proportion of key information items to be inspected.

[0085] Obtain the mean of the semantic similarity between the target key information and each reference key information to get the first mean;

[0086] The mean of the semantic similarity between the target key information and each non-reference key information is obtained to obtain the second mean.

[0087] The difference between the first mean and the second mean is determined as the feature difference of the target key information;

[0088] The product of the proportion of key information and the feature difference is determined as the information high-value specificity of the target key information.

[0089] Based on the key information obtained from the inspection, the information permeability of the target key information is obtained.

[0090] Optionally, based on the key information examined, the information permeability of the target key information is obtained, specifically including:

[0091] The key information sequence of each key information in the test is obtained sequentially to obtain the ordinal value sequence;

[0092] Based on the ordinal value sequence, obtain the interval value sequence, and determine the target interval value sequence from the interval value sequence;

[0093] Based on the target interval value sequence, calculate the information penetration carrying capacity of key target information.

[0094] For example, different types of tender documents differ significantly in their core clauses (such as technical parameters, acceptance criteria, and price composition). Classification allows the model to learn the specific rules for each field. "Hybrid tender documents," falling between two or more categories, contain content modules from different fields due to the complexity of project requirements. Therefore, it's necessary to analyze the performance characteristics of each key piece of information in each historical tender document. The more important a particular key piece of information is in a tender document, the more attention should be paid to that key piece of information during subsequent tender matching.

[0095] The semantic similarity of target key information to other key information can be calculated using the path-length method (a well-known technique).

[0096] The path length method first integrates all key information from all key information sets to form a unique global dictionary, ensuring consistency in vector dimensions. Second, according to the order of the global dictionary, each key information is mapped to a vector, and the semantic similarity between two key information pieces is obtained by multiplying the weights of corresponding dimensions of the two vectors using a dot product and then summing the results. The semantic similarity value ranges from [value missing]. .

[0097] The preset first similarity threshold can be set based on historical experience or manually. There is no specific limitation on the value of the preset first similarity threshold. It can be modified according to the actual situation. In a preferred embodiment, it can be set to 0.8.

[0098] Taking the key information set of the t-th historical bid as an example, that is, the t-th historical bid is determined as the target bid, the key information set of the t-th historical bid is determined as the target key information set, and any key information in the key information set of the t-th historical bid is determined as the target key information. A preset first similarity threshold is set to 0.8. Any key information in the key information set of the t-th historical bid is denoted as the target key information. All other key information with a semantic similarity greater than 0.8 with the target key information is denoted as reference key information. The target key information and all reference key information are collectively referred to as verification key information. All other key information besides verification key information is denoted as non-reference key information.

[0099] Key information in tender documents often carries high-value information and has distinct industry-specific characteristics. This means that core keywords appear repeatedly in key sections such as requirements, terms, and commitments, and are distributed across different chapters to ensure consistency across all aspects, rather than being concentrated in a non-core paragraph. Furthermore, core keywords are often industry-specific terms, such as "on-site supervision" and "concrete strength grade" in engineering tenders, and "SLA (Service Level Agreement)" and "Operation and Maintenance Response Time" in IT service tenders, exhibiting extremely low cross-industry reusability.

[0100] Based on the key information examined, reference key information, and non-reference key information, the formula for calculating the information price specificity of target key information can be:

[0101]

[0102] in, Information indicating key target information is highly valuable and specific. This represents the number of key information items in the key information set of the t-th historical tender. This represents the number of key information items for verification in the key information set of the t-th historical tender. Let represent the mean semantic similarity between the target key information and the reference key information in the key information set of the t-th historical tender. This represents the mean semantic similarity between the target key information and the non-reference key information in the key information set of the t-th historical tender.

[0103] In this formula, core and key information will be repeatedly mentioned in the requirements description, terms explanation, and commitment sections to ensure accurate and unambiguous expression. The larger the value of C, the more frequently key information with similar meaning to the target key information appears. The closer P1 is to 1 and the closer P2 is to 0, the more semantically similar the key information with similar meaning to the target key information is, and the less semantically similar the key information with dissimilar meaning is, indicating that the target keyword has strong industry-specific attributes. Therefore, the larger C is, the higher the value of the target key information and the stronger its specificity.

[0104] Furthermore, it is necessary to analyze whether key information with similar meaning to the target key information runs through the key modules of the tender document to ensure consistency of information in all aspects. For example, "quality acceptance specifications" may appear simultaneously in "technical requirements," "acceptance process," and "liability for breach of contract."

[0105] For example, obtaining an interval value sequence based on an ordinal value sequence, and determining a target interval value sequence from the interval value sequence, specifically includes:

[0106] The difference between the (a+1)th element and the ath element in the ordinal numerical sequence is determined as the interval value of the (a+1)th element.

[0107] Obtain the interval value of each element in the ordinal value sequence to get the interval value sequence, where the (a+1)th element is not the first element in the ordinal value sequence;

[0108] The number of key information items in the target key information set is obtained, and the product of the preset threshold coefficient and the target key information set is determined as the interval threshold.

[0109] Intervals in the interval value sequence that are greater than the interval value threshold are identified as target interval values.

[0110] The sequence consisting of all target interval values ​​is determined as the target interval value sequence.

[0111] Optionally, the preset threshold coefficient can be set based on historical experience or manually. There is no specific limitation on the value of the preset threshold coefficient, which can be modified according to the actual situation. In a preferred embodiment, it can be set to 5%.

[0112] The specific steps are as follows:

[0113] In the key information set of the t-th historical tender, the key information sequence of all the key information for verification is obtained in the order of reading, forming an ordinal value sequence.

[0114] In the ordinal value sequence, the difference between the order of the next key information and the order of the previous key information is calculated sequentially to form the interval value sequence. ,in, , as well as These represent the order of the 1st, 2nd, and 3rd key information items in the ordinal numerical sequence, respectively. The interval value reflects the number of key information items that exist between adjacent key information items in the historical tender documents according to the reading order.

[0115] A preset threshold coefficient of 5% is set, and the number of all interval values ​​greater than 5%×S in the interval value sequence, S2, is counted, where S is the number of key information items in the key information set of the t-th historical tender document. In this embodiment, for interval values ​​greater than 5%×S, it is considered that there is a large amount of key information between adjacent key information items in the historical tender document according to the reading order, that is, adjacent key information items may span some chapters, resulting in low information penetration.

[0116] Based on the target interval value sequence, the formula for calculating the information penetration carrying capacity of key target information can be:

[0117]

[0118] in, Information carrying capacity that represents key target information. This represents the number of key information items in the key information set of the t-th historical tender. This indicates the order of the most significant key information in the ordinal numerical sequence. This indicates the order of the first key piece of information in the ordinal numerical sequence. This indicates the number of target interval values ​​in the target interval value sequence. This indicates the number of interval values ​​in the interval value sequence.

[0119] In the formula, The larger the value, the more widely key information with similar meaning to the target key information is distributed in the t-th historical tender. The smaller the value of D, the fewer instances of cross-chapter situations occur between adjacent key information in the t-th historical tender document according to the reading order. Therefore, the larger the value of D, the more key information with similar significance to the target key information appears throughout the entire historical tender document and is distributed across different chapters, ensuring information consistency at each stage.

[0120] Therefore, the formula for calculating the performance importance of key target information can be:

[0121]

[0122] in, This indicates the performance importance of key target information. This represents the normalization function.

[0123] Among these, the more frequently key information with similar meanings appears in the tender documents, the more often it is distributed across different chapters, and the stronger the specificity of its meaning, the more likely it is to be the core key information in the tender documents. Following this method, the relative importance of each key piece of information within the key information set of each historical tender document can be obtained.

[0124] S103. Based on the importance of each key piece of information, calculate the core manifestation factor of each key piece of information in each tender document.

[0125] In this embodiment, based on the importance of each key piece of information, the core manifestation factor of each key piece of information in each tender document is calculated, specifically including:

[0126] Determine a set of key information for comparison from the set of key information, and calculate the semantic similarity between each key information in the set of key information for comparison and the target key information.

[0127] Key information with semantic similarity greater than a preset second similarity threshold is identified as key information for mapping.

[0128] When comparing tender documents containing key information related to the target tender, the tender documents are identified as relevant historical tender documents of the target tender.

[0129] Based on the performance importance of each key piece of information, calculate the degree of correlation between each relevant historical tender and the target key information;

[0130] Calculate the initial specific novelty of the target key information based on the degree of correlation between each relevant historical tender and the target key information;

[0131] Based on the initial tags of the target bid, identify the main relevant historical bid and the sub-relevant historical bid from the relevant historical bids, and calculate the type specificity of the target key information based on the main relevant historical bid and the sub-relevant historical bid.

[0132] When the comparative tender does not contain the mirroring key information of the target key information, the comparative tender is identified as an irrelevant historical tender of the target tender, and the initial special novelty of the target key information is set to the first preset value. The comparative tender is the tender corresponding to the set of comparative key information in the historical tender data.

[0133] Based on the initial specificity and type specificity of the target key information, calculate the core embodiment factor of each key information in each tender document.

[0134] Optionally, based on the main relevant historical tender documents and the sub-relevant historical tender documents, the type specificity of the target key information is calculated, specifically including:

[0135] When relevant historical bids contain key information about the target, the type specificity of the key information about the target is calculated based on the degree of association between each relevant historical bid and the key information about the target.

[0136] When the relevant historical tender documents do not contain the main relevant historical tender documents containing the key information of the target, the type specificity of the key information of the target is set to the second preset value.

[0137] Based on the initial specificity and type specificity of the target key information, calculate the core embodiment factor of each key information in each tender document, specifically including:

[0138] Based on the initial specificity and type specificity of the target key information, calculate the specificity of the target key information, and based on the specificity of the target key information, calculate the core embodiment factor of the target key information;

[0139] Each key piece of information in the set of key target information is identified as the key target information, and the core manifestation factors of each key piece of information are obtained respectively.

[0140] Each bid document in the historical bid document data is identified as a target bid document, and the core manifestation factors of each key information in each bid document are obtained.

[0141] Based on the initial tags of the target tender, identify the main relevant historical tender and the sub-relevant historical tender from the relevant historical tender documents, specifically including:

[0142] The initial label of the target bid is determined as the target label. Among the relevant historical bids, the bids with the same initial label as the target label are determined as the primary relevant historical bids, and the bids with different initial labels as the target labels are determined as the secondary relevant historical bids.

[0143] For example, S102 analyzed the key information within the same tender document. Important key information in tender documents is often industry-specific terminology, appearing frequently only in tenders within that specific field, with extremely low cross-industry reusability. For instance, in engineering tenders, terms like "on-site supervision," "concrete strength grade," and "construction organization design" only appear in building and municipal engineering tenders, rarely appearing in other industry tenders (such as software procurement). In service tenders, terms like "Service Level Agreement," "response timeliness," and "maintenance inspection frequency" are specialized terms specific to IT and property services tenders, completely different from those used in goods procurement tenders. In goods tenders, terms like "warranty period," "technical parameter deviation table," and "delivery acceptance standards" only apply to equipment and material procurement tenders and are not core to engineering tenders. Therefore, further analysis of the performance of key information across different historical tenders is needed to determine the core factors that embody key information.

[0144] Optionally, the preset second similarity threshold can be set based on historical experience or manually. There is no specific limitation on the value of the preset second similarity threshold, which can be modified according to the actual situation. In a preferred embodiment, it can be set to 0.9.

[0145] Taking the target key information in the key information set of the t-th historical bid as an example, we obtain the semantic similarity between the target key information and each key information in the key information set of the x-th historical bid. All key information in the key information set of the x-th historical bid with a semantic similarity greater than 0.9 is recorded as the mapping key information. That is, the x-th historical bid is determined as the comparison bid, and the key information set of the x-th historical bid is determined as the comparison key information set.

[0146] If there is relevant key information, then the x-th historical bid is identified as a related historical bid of the t-th historical bid. If there is no relevant key information, then the x-th historical bid of the t-th historical bid is identified as an irrelevant historical bid of the t-th historical bid.

[0147] Based on the above, all historical bids except for the t-th historical bid can be classified (only for the t-th historical bid).

[0148] Based on the perceived importance of each key piece of information, the correlation between each relevant historical tender document and the target key information is calculated. The calculation formula can be:

[0149]

[0150] in, This indicates the degree of correlation between the b-th relevant historical tender and the key information of the target. This represents the number of mapping key information items in the key information set of the b-th relevant historical tender. This represents the number of key information items in the key information set of the b-th relevant historical tender. This represents the sum of the performance importance of all key information reflected in the key information set of the b-th relevant historical tender. This represents the sum of the performance importance of all key information in the key information set of the b-th relevant historical tender.

[0151] In the formula, the more key information reflected in the key information set of the b-th related historical tender, and the greater the importance of the reflected key information, the greater the correlation between each related historical tender and the target key information.

[0152] Because some key information in certain tender documents rarely appears in other tender documents, this situation is particularly common in special needs projects or industry innovation tenders. The core reason is the difference between the project's "personalized requirements" and "general standards." Specifically, when some projects need to meet the unique needs of the tendering party, for example, in a tender for a "customized monitoring system for nuclear power plants," core information such as "radiation-resistant equipment parameters" and "nuclear-grade qualification certification" would almost never appear in ordinary security monitoring tender documents. When tender projects involve new technologies and new scenarios, core information may be rare because the industry has not yet formed common terminology. For example, in the "Metaverse Exhibition Hall Construction Project," "real-time interaction latency requirements for virtual scenes" and "digital asset ownership confirmation scheme" have no corresponding information in traditional exhibition hall decoration tender documents, and this key information is particularly important in the corresponding tender documents. Therefore, it is necessary to analyze the specialization and novelty of key information in historical tender documents.

[0153] When no relevant historical tender documents exist, it means that the target key information only appears in the t-th historical tender document. Therefore, the initial specific novelty of the target key information is set to a first preset value (the first preset value can be 1, and can be set according to the actual situation; no specific restriction is made here). When relevant historical tender documents exist, the formula for calculating the initial specific novelty of the target key information can be:

[0154]

[0155] in, This indicates the initial specific novelty of the target's key information when relevant historical tender documents exist. This indicates the degree of correlation between the i-th relevant historical tender and the key information of the target. Indicates the number of relevant historical tenders. This represents an exponential function.

[0156] In the formula, Used to obtain the inverse proportional normalized value of the data. The smaller the value, the fewer relevant historical tender documents there are, and the weaker the correlation between the relevant historical tender documents and the target key information, indicating a stronger initial novelty of the target key information.

[0157] Furthermore, when the initial specificity of the target key information is relatively small, i.e., when there are many strongly related historical bids, the related historical bids may all be bids of the same type as the t-th historical bid, indicating that the target key information is an industry-specific term. However, there may also be related historical bids of different types from the t-th historical bid, leading to a wider and more general application scope of the target key information. Therefore, further analysis of the type specificity of the target key information is needed.

[0158] Among all related historical bids of the t-th historical bid, all related historical bids with the same initial label ("Goods Procurement", "Services", "Engineering Construction") as the t-th historical bid are designated as primary related historical bids, and all related historical bids with different initial labels from the t-th historical bid are designated as secondary related historical bids. The initial label of the t-th historical bid is the target label.

[0159] When no relevant historical bid documents exist, it means that all relevant historical bid documents are of the same type as the t-th historical bid document, thus ensuring the type specificity of the target key information. Set to the second preset value (the second preset value can be 1, and can be set according to the actual situation; no specific restriction is made here). When there are related historical tender documents, the type specificity of the target key information... The calculation formula is:

[0160]

[0161] in, Indicates the number of relevant historical tenders. This indicates the number of relevant historical tenders. This indicates the number of initial tags of the relevant historical tender documents (in this embodiment, there are 3 types of initial tags). This represents the sum of the correlations between all relevant historical tender documents and key target information. This represents the normalization function.

[0162] In the formula, the greater the proportion of related historical bids among related historical bids, the more types of related historical bids there are, and the greater the correlation between related historical bids and target key information, the less likely the target key information belongs to the same type of bid. Therefore, the larger U is, the greater the type specificity of the target key information.

[0163] Based on the initial specificity and type specificity of the target key information, calculate the core embodiment factor of each key information in each tender document, specifically including:

[0164] Calculate the specific novelty of the target key information based on its initial specificity and type specificity;

[0165] Based on the specific novelty of the target's key information, calculate the core embodiment factor of the target's key information.

[0166] The formula for calculating the specific novelty of the target key information, based on its initial specificity and type specificity, can be as follows:

[0167]

[0168] in, This indicates the specificity and novelty of the target's key information. When relevant historical bids exist, the fewer the relevant historical bids, the weaker the correlation between the relevant historical bids and the target's key information, and the fewer and less varied the sub-relevant historical bids among the relevant historical bids, the stronger the specificity and novelty of the target's key information.

[0169] Based on the specific novelty of the target's key information, the formula for calculating the core embodiment factor of the target's key information can be:

[0170]

[0171] in, The core factors that represent the key information of the target. The importance of the target key information is determined by its performance within its own historical tender documents, and its rarity in other historical tender documents. This indicates that the target key information is important, specific, or novel core information within the tender document.

[0172] From the above, we can obtain the core elements of each key information in the key information set of each historical tender.

[0173] S104. Based on the core representation factor of each key information in each bid document, calculate the clustering distance between each bid document and any other bid document. Use the clustering distance as the clustering parameter of the fuzzy C-means clustering method. Use the fuzzy C-means clustering method to cluster the bid documents in the historical bid document data to obtain the clustering results.

[0174] In this embodiment, based on the core representation factors of each key information in each tender document, the clustering distance of each tender document with any other tender document is calculated, specifically including:

[0175] Calculate the cosine similarity between the first and second bids based on the core representation factors of each key information in each bid document;

[0176] The inverse proportional value of the cosine similarity is determined as the cluster distance between the first and second bids;

[0177] Any bid in the historical bid data is designated as the first bid and any bid is designated as the second bid. The cluster distance of each bid to any other bid is then obtained.

[0178] For example, the weighted cosine similarity (a common calculation) of the key information sets of any two historical bids is obtained by using the core manifestation factor of each key information set in the key information set of each historical bid as the weight. The inverse proportional value of the weighted cosine similarity is used as the clustering distance. Fuzzy C-means clustering (a common technique) is used to divide all historical bids into several classes, and the classification result of historical bids is obtained.

[0179] The weighted cosine similarity ranges from -1 to 1. The inverse proportional value is calculated by subtracting the difference in weighted cosine similarity from 1. Key information with larger core characteristics is assigned greater weight to highlight its contribution, ensuring the similarity calculation results better align with the actual business needs of bid comparison. This avoids judgment bias caused by treating all information "equally," thereby improving classification accuracy. Furthermore, the fuzzy C-means clustering method allows a historical bid to belong to multiple categories, ensuring the accuracy of classifying "mixed-category bids."

[0180] S105. Use the clustering results to train the neural network to obtain the tender document auxiliary model. Input the target tender document into the tender document auxiliary model and output the target auxiliary verification reference document through the tender document auxiliary model.

[0181] In this embodiment, the neural network can be selected from the GPT series (a well-known technology). Based on the classification results of historical tender documents, the neural network learns specific knowledge and rules for checking each type of tender document. Supervised learning, reinforcement learning, and other methods are used to adjust the model's parameters and structure based on the training effect of the neural network, improving the accuracy and generalization ability of the neural network, thus obtaining a tender document auxiliary model.

[0182] The system retrieves the latest uploaded tender documents (i.e., the target tender documents) from the user, inputs them into the tender document assistance model, and outputs a target auxiliary verification reference document. Specifically, the tender document assistance model retrieves, integrates, and rewrites relevant content from the database based on the key requirements, automatically generating highly adapted drafts of core chapters such as technical proposals, implementation plans, business responses, and qualification review forms. Then, it performs a compliance check against the scoring criteria, points out potentially missed scoring points or risky answers, and provides optimization suggestions, such as highlighting one's own advantages and avoiding potential risks, to assist the bidding personnel in finalizing the draft.

[0183] In summary, in this embodiment of the invention, by analyzing the performance importance, specialization, and novelty of each key information in the key information set of historical tender documents, core reflective factors are determined and used as weights. This ensures the accuracy of historical tender document classification. After classification, the model can learn specific domain patterns from similar data, avoiding "generalization bias" caused by mixed data. For example, when processing commercial tender data, the model can more accurately grasp the pricing calculation logic and qualification document verification rules; when processing technical tender data, it can gain a deeper understanding of the technical parameter matching requirements of different industries. Ultimately, the powerful generation and understanding capabilities of the large language model are utilized to improve the efficiency of tender document verification.

[0184] This invention also proposes a tender document verification device; please refer to [link / reference]. Figure 2 The diagram shows a schematic of the structure of a tender document auxiliary verification device provided in an embodiment of the present invention. The system includes: a data acquisition unit 101, a data processing unit 102, and an auxiliary verification unit 103.

[0185] The data acquisition unit 101 is used to extract key information of each bid from the historical bid data to obtain a key information set for each bid. Each key information in each key information set includes the key information order and the initial label.

[0186] The data processing unit 102 is used to calculate the performance importance of each key information in each tender document in the key information set; calculate the core manifestation factor of each key information in each tender document based on the performance importance of each key information; calculate the clustering distance of each tender document with any other tender document based on the core manifestation factor of each key information in each tender document; use the clustering distance as the clustering parameter of the fuzzy C-means clustering method; and use the fuzzy C-means clustering method to cluster the tender documents in the historical tender document data to obtain the clustering results.

[0187] The auxiliary verification unit 103 is used to train the neural network using the clustering results to obtain the tender document auxiliary model, input the target tender document into the tender document auxiliary model, and output the target auxiliary verification reference document through the tender document auxiliary model.

[0188] This invention also proposes a tender document verification device; please refer to [link / reference]. Figure 3 The diagram shows a schematic of the structure of a tender document verification device provided in an embodiment of the present invention. The device includes a processor 201, a communication interface 202, a memory 203 and a communication bus 204, wherein the processor 201, the communication interface 202 and the memory 203 communicate with each other through the communication bus 204.

[0189] The memory 203 stores a computer program, which, when executed by the processor 201, causes the processor 201 to perform the steps of any of the tender document verification methods provided in the embodiments of this application.

[0190] Since the above-mentioned electronic equipment solves the problem in a similar way to the tender document verification method, the implementation of the above-mentioned electronic equipment can be found in the embodiments of the method, and the repeated parts will not be described again.

[0191] The communication bus mentioned in the aforementioned electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus. Communication interface 202 is used for communication between the aforementioned electronic device and other devices. The memory can include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory can also be at least one storage device located remotely from the aforementioned processor.

[0192] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be 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.

[0193] Based on the above embodiments, this invention also provides a computer-readable storage medium storing a computer program executable by a processor. When the program runs on the processor, it causes the processor to perform the steps of any of the tender document verification methods provided in this application.

[0194] Based on the same inventive concept, embodiments of this application provide a computer program product, including a computer program, which, when executed by a processor, implements the steps of any of the tender document verification methods provided in embodiments of this application.

[0195] Since the principle of the computer-readable storage medium in solving the problem is similar to that of the tender document verification method, the implementation of the computer-readable storage medium can be found in the embodiments of the method, and repeated details will not be repeated.

[0196] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0197] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0198] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0199] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0200] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0201] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0202] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assisting in the verification of tender documents, characterized in that, include: Extract key information from each bid in the historical bid data to obtain a key information set for each bid. Each key information in each key information set includes the key information order and the initial tag. Calculate the performance importance of each key piece of information in each tender document within the key information set; Based on the perceived importance of each key piece of information, calculate the core manifestation factor of each key piece of information in each tender document; Based on the core representation factors of each key information in each tender document, the clustering distance of each tender document with any other tender document is calculated. The clustering distance is used as the clustering parameter of the fuzzy C-means clustering method. The fuzzy C-means clustering method is used to cluster the tender documents in the historical tender document data to obtain the clustering results. The clustering results are used to train the neural network to obtain the tender document assistance model. The target tender document is input into the tender document assistance model, and the target auxiliary verification reference document is output by the tender document assistance model. The calculation of the core manifestation factor of each key piece of information in each tender document based on the performance importance of each key piece of information specifically includes: Determine a set of key information for comparison from the set of key information, and calculate the semantic similarity between each key information in the set of key information for comparison and the target key information. Key information with semantic similarity greater than a preset second similarity threshold is identified as key information for mapping. When comparing tender documents containing key information related to the target tender, the comparison tender documents are identified as relevant historical tender documents of the target tender. Based on the performance importance of each key piece of information, calculate the degree of correlation between each relevant historical tender and the target key information; Calculate the initial specific novelty of the target key information based on the degree of correlation between each relevant historical tender and the target key information; Based on the initial tags of the target bid, identify the main relevant historical bid and the sub-relevant historical bid from the relevant historical bids, and calculate the type specificity of the target key information based on the main relevant historical bid and the sub-relevant historical bid. When the comparative tender does not contain the mirroring key information of the target key information, the comparative tender is identified as an irrelevant historical tender of the target tender, and the initial special novelty of the target key information is set to the first preset value. The comparative tender is the tender corresponding to the set of comparative key information in the historical tender data. Based on the initial specificity and type specificity of the target key information, calculate the core embodiment factor of each key information in each tender document.

2. The method for auxiliary verification of tender documents according to claim 1, characterized in that, The calculation of the performance importance of each key piece of information in each tender document within the key information set specifically includes: The target bid is identified from historical bid data, the target key information set of the target bid is identified from the key information set, the target key information is identified from the target key information set, and the semantic similarity of the target key information with other key information is calculated. The other key information refers to the key information in the target key information set other than the target key information. Based on semantic similarity, reference key information, non-reference key information, and verification key information are determined from the target key information set. Based on the verification key information, reference key information, and non-reference key information, the information high-value specificity and information penetration carrying capacity of the target key information are calculated. The performance importance of the target key information is determined by the product of its informational value specificity and its informational pervasiveness. Each key piece of information in the set of key target information is identified as a key target information, and the performance importance of each key piece of information is obtained separately. Each bid document in the historical bid document data is identified as a target bid document, and the performance importance of each key piece of information in each bid document is obtained; The calculation of the information value specificity and information pervasiveness of the target key information based on the key information being examined, referenced key information, and non-referenced key information specifically includes: The quantity of key information obtained for inspection; The ratio of the number of key information items to the number of key information items in the target key information set is determined as the proportion of key information items to be inspected. Obtain the mean of the semantic similarity between the target key information and each reference key information to get the first mean; The mean of the semantic similarity between the target key information and each non-reference key information is obtained to obtain the second mean. The difference between the first mean and the second mean is determined as the feature difference of the target key information; The product of the proportion of key information and the feature difference is determined as the information high-value specificity of the target key information. Based on the key information obtained from the inspection, the information permeability of the target key information is obtained.

3. The method for auxiliary verification of tender documents according to claim 2, characterized in that, The step of determining reference key information, non-reference key information, and verification key information from the target key information set based on semantic similarity specifically includes: Other key information with semantic similarity greater than a preset first similarity threshold is identified as reference key information; Other key information with semantic similarity less than or equal to a preset first similarity threshold is identified as non-reference key information; The target key information and reference key information are combined to obtain the verification key information.

4. The method for auxiliary verification of tender documents according to claim 2, characterized in that, The information permeability of obtaining target key information based on key inspection information specifically includes: The key information sequence of each key information in the test is obtained sequentially to obtain the ordinal value sequence; Based on the ordinal value sequence, obtain the interval value sequence, and determine the target interval value sequence from the interval value sequence; Based on the target interval value sequence, calculate the information penetration carrying capacity of key target information.

5. The method for auxiliary verification of tender documents according to claim 1, characterized in that, The calculation of the type specificity of target key information based on the main relevant historical tender documents and the sub-relevant historical tender documents specifically includes: When relevant historical bids contain key information about the target, the type specificity of the key information about the target is calculated based on the degree of association between each relevant historical bid and the key information about the target. When the relevant historical tender documents do not contain the main relevant historical tender documents containing the key information of the target, the type specificity of the key information of the target is set to the second preset value.

6. The method for auxiliary verification of tender documents according to claim 1, characterized in that, The calculation of the core embodiment factor of each key piece of information in each tender document, based on the initial specificity and type specificity of the target key information, specifically includes: Based on the initial specificity and type specificity of the target key information, calculate the specificity of the target key information, and based on the specificity of the target key information, calculate the core embodiment factor of the target key information; Each key piece of information in the set of key target information is identified as the key target information, and the core manifestation factors of each key piece of information are obtained respectively. Each bid document in the historical bid document data is identified as a target bid document, and the core manifestation factors of each key information in each bid document are obtained.

7. A tender document verification aid device, characterized in that, include: The data acquisition unit is used to extract key information from each bid in the historical bid data to obtain a key information set for each bid. Each key information in each key information set includes the key information order and the initial label. The data processing unit is used to calculate the performance importance of each key piece of information in each tender document within the key information set; and to calculate the core manifestation factor of each key piece of information in each tender document based on its performance importance. Specifically, this includes: determining a set of comparative key information from the key information set; calculating the semantic similarity between each comparative key piece of information in the comparative key information set and the target key information; identifying comparative key information with a semantic similarity greater than a preset second similarity threshold as mapping key information; when a comparative tender document contains mapping key information of the target key information, identifying the comparative tender document as a related historical tender document of the target tender document; calculating the degree of association between each related historical tender document and the target key information based on the performance importance of each key piece of information; calculating the initial specific novelty of the target key information based on the degree of association between each related historical tender document and the target key information; and calculating the initial specific novelty of the target key information based on the initial tags of the target tender document. The relevant historical bid documents are used to identify primary and secondary relevant historical bid documents. Based on these documents, the type specificity of the target key information is calculated. When the comparison bid documents do not contain the mapping key information of the target key information, the comparison bid documents are identified as irrelevant historical bid documents of the target bid documents, and the initial specificity and novelty of the target key information are set to a first preset value. Here, the comparison bid documents are the bid documents corresponding to the set of comparison key information in the historical bid document data. Based on the initial specificity and novelty and type specificity of the target key information, the core manifestation factor of each key information in each bid document is calculated. Based on the core manifestation factor of each key information in each bid document, the clustering distance of each bid document with any other bid document is calculated. The clustering distance is used as the clustering parameter of the fuzzy C-means clustering method. The fuzzy C-means clustering method is used to cluster the bid documents in the historical bid document data to obtain the clustering results. The auxiliary verification unit is used to train the neural network using the clustering results to obtain the tender document auxiliary model. The target tender document is input into the tender document auxiliary model, and the target auxiliary verification reference document is output through the tender document auxiliary model.

8. A tender document verification aid device, characterized in that, include: At least one processor; and memory that is communicatively connected to at least one processor; The memory stores instructions that can be executed by at least one processor, which, when executed by at least one processor, enables the at least one processor to perform the method of any one of claims 1-6.

Images