A bidding document retrieval question and answer method and device based on directory association verification, electronic equipment and storage medium

Abstract

This invention discloses a tender document retrieval question-answering method, apparatus, electronic device, and storage medium based on directory association verification, belonging to the field of intelligent document retrieval. The method includes: acquiring the tender document text and the main question statement; parsing the association relationships between the tender document directories and generating auxiliary question statements based on the main question statement; retrieving the response text and directory paths corresponding to the main question statement and auxiliary question statements respectively; determining the directory path association relationship by combining the directory association relationship; comparing the statement association relationship with the directory path association relationship; and outputting the corresponding response text when they match. Therefore, by implementing this invention, the problems of existing technologies, such as the lack of cross-verification of the document's physical directory structure and question logic, which leads to contextual misalignment in retrieval results and low retrieval accuracy, can be solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent document retrieval technology, specifically to a method, apparatus, electronic device, and storage medium for bid document retrieval question answering based on directory association verification. Background Technology

[0002] With the digital transformation of bidding and tendering processes, the demand for automated processing of long documents such as tender documents is increasing. Tender documents are typically lengthy and highly technical, containing a tightly structured hierarchy of clauses covering technical, commercial, and legal aspects. When faced with such complex documents, tender document retrieval-based question-and-answer technology, with its directory association verification, can help users navigate through massive amounts of text and quickly and accurately locate key clauses.

[0003] However, existing tender document question-and-answer systems suffer from low retrieval accuracy and are prone to contextual misalignment in practical applications. The main reason for this is that current retrieval methods mostly rely on a flat semantic similarity calculation between a single question and candidate paragraphs, thus severing the inherent hierarchical logic of the tender document. Because existing technologies lack a cross-validation mechanism between the extended logic of the question's intent and the physical directory structure of the tender document, when different chapters contain clauses with similar expressions but different applicable conditions, the system struggles to verify whether a single retrieved text belongs to a structural branch consistent with the question's logic. This can easily lead to paragraphs deviating from the structural context being output as the final result, affecting the reliability of the question-and-answer results. Summary of the Invention

[0004] This invention provides a bid document retrieval question-answering method, device, electronic device, and storage medium based on directory association verification. It can solve the problem in the prior art that the lack of cross-verification between the physical directory structure of the document and the question logic leads to contextual misalignment and low retrieval accuracy in the search results.

[0005] An embodiment of the present invention provides a tender document retrieval question-answering method based on directory association verification, comprising: Retrieve the tender document text and the main question statement to be processed; The tender document text to be processed is parsed to determine the type of directory relationship between the tender documents. Based on the main question statement, extract related words from the preset keyword dictionary; combine the related words to generate auxiliary question statements, and determine the type of sentence association between the main question statement and the auxiliary question statements; Based on the main question statement and auxiliary question statements, a text search is performed on the tender document text to be processed, generating the first response text corresponding to the main question statement and the first directory path of the first response text in the tender document text to be processed, as well as the second response text corresponding to the auxiliary question statements and the second directory path of the second response text in the tender document text to be processed. Based on the directory association type between each tender document directory, the path association type between the tender document directory corresponding to the first directory path and the tender document directory corresponding to the second directory path is extracted; Compare the statement association type with the path association type; if the statement association type and the path association type are the same, output the first response text.

[0006] Furthermore, the tender document text to be processed is parsed to determine the types of directory relationships between the tender document directories, including: The tender document text to be processed is combined in pairs to generate tender document index pairs; For each pair of tender documents, determine the first and second directory headings included in the current pair of tender documents; Perform text semantic parsing on the first and second directory titles to determine the first semantic category of the first directory title and the second semantic category of the second directory title. Structural feature identification is performed on the first directory title and the second directory title to determine the first directory level of the first directory title and the second directory level of the second directory title; The first directory title and the second directory title are divided into subject categories to determine the first subject to which the first directory title belongs and the second subject to which the second directory title belongs. Keyword attributes are extracted from the first and second directory titles to determine the interaction logic between them. Based on the first semantic category, the second semantic category, the first directory level, the second directory level, the first subject, the second subject, and the interaction logic, determine the type of directory association relationship between the current tender directory and the corresponding directory; The directory association type corresponding to each tender document directory is used as the directory association type between each tender document directory.

[0007] Furthermore, the pre-defined set of relationship types includes subordinate relationships, parallel relationships, temporal relationships, causal relationships, complementary relationships, and responsive relationships; Based on the first semantic category, the second semantic category, the first directory level, the second directory level, the first subject, the second subject, and the interaction logic, determine the type of directory association relationship between the current tender document directory and the corresponding directory, including: When there is an overlap relationship between the first semantic category and the second semantic category, the current tender document directory is determined to be a subordinate relationship to the corresponding directory. If the current tender document directory pair does not belong to a subordinate relationship and the interaction logic is executed sequentially, then the directory association type corresponding to the current tender document directory pair is determined to be a sequential relationship. If the current tender document directory pair does not belong to a subordinate relationship, and the interaction logic is a basis-result correspondence logic, then the directory association type corresponding to the current tender document directory pair is determined to be a causal relationship. If the current tender document directory does not belong to a subordinate relationship and the interaction logic is a requirement and response correspondence logic, then the directory association type corresponding to the current tender document directory is determined to be a response relationship. If the current tender document directory does not belong to the subordinate, chronological, causal, or response relationship, and the first and second subjects are consistent, and the interaction logic is complementary, then the corresponding directory association type of the current tender document directory is determined to be complementary. If the current tender document directory does not belong to any of the above association types, the first semantic category and the second semantic category are independent of each other, and the first directory level and the second directory level are the same, then the directory association type corresponding to the current tender document directory is determined to be a parallel relationship. If the current tender document catalog does not meet the determination criteria for any of the above association types, it is determined that the current tender document catalog does not have an association.

[0008] Furthermore, based on the main statement of the question, related words are extracted from a pre-defined keyword dictionary, including: Feature extraction is performed on the main question statement to generate initial action keywords and initial entity keywords contained in the main question statement; Calculate the similarity between the initial action keywords and each action word in the preset keyword dictionary, and generate semantic similarity for each action; Calculate the similarity between the initial entity keywords and each entity word in the preset keyword dictionary, and generate the semantic similarity of each entity; The action with the highest semantic similarity score is selected from all action semantic similarities to generate the maximum action semantic similarity score; the action words corresponding to the maximum action semantic similarity score are established as standard action keywords. The entity with the highest semantic similarity score is selected from all entity semantic similarity scores to generate the maximum entity semantic similarity score; the entity words corresponding to the maximum entity semantic similarity score are established as standard entity keywords. Extract a set of related action words that are associated with the standard action keywords from the pre-defined keyword dictionary; Extract a set of related entity words that are associated with standard entity keywords from a pre-defined keyword dictionary; Combine the set of related action words and the set of related entity words to generate each related word.

[0009] Furthermore, the related words are combined to generate auxiliary question statements, and the type of statement relationship between the main question statement and the auxiliary question statements is determined, including: Each related keyword is paired with its associated action words and associated entity words to generate keyword combinations. Input each keyword combination into the preset statement generation template to generate each alternative auxiliary question statement; Get the selection command; Based on the selection instructions, determine the auxiliary question statement from each candidate auxiliary question statement, and determine the target action vocabulary and target entity vocabulary contained in the auxiliary question statement; Extract the action relationship types between target action words and standard action keywords, as well as the entity relationship types between target entity words and standard entity keywords from the preset keyword dictionary; Based on the action relationship type and entity relationship type, determine the statement association type between the main question statement and the auxiliary question statements.

[0010] Furthermore, based on the main question statement and auxiliary question statements, text retrieval is performed on the tender document text to be processed, generating the first response text corresponding to the main question statement and the first directory path of the first response text in the tender document text to be processed, as well as the second response text corresponding to the auxiliary question statements and the second directory path of the second response text in the tender document text to be processed, including: Calculate the similarity between the main statement of the question and each text paragraph in the tender document to be processed, and generate the first text similarity score; The text with the highest numerical similarity score is used to generate the maximum first text similarity score. The text paragraph with the highest first text similarity will be used as the first response text corresponding to the main question statement; Based on the tender document text to be processed, locate the first directory node to which the first response text belongs in the tender document text to be processed; Based on the first directory node, determine the hierarchical sequence of the first directory node and generate the first directory path corresponding to the first response text in the tender document text to be processed; Calculate the similarity between the auxiliary question statement and each text paragraph in the tender document to be processed, and generate the similarity scores for each second text. The second text with the highest numerical similarity is used to generate the maximum second text similarity. The text paragraph with the highest second text similarity is designated as the second response text corresponding to the auxiliary question statement; Based on the tender document text to be processed, locate the second directory node to which the second response text belongs in the tender document text to be processed; Determine the hierarchical sequence of the second directory node based on the second directory node; Based on the hierarchical sequence of the second directory node, the second directory path corresponding to the second response text in the tender document text to be processed is generated.

[0011] Furthermore, based on the directory association types between the various tender document directories, the path association types between the tender document directories corresponding to the first directory path and the tender document directories corresponding to the second directory path are extracted, including: Based on the first directory path, extract the last-level directory nodes contained in the first directory path and establish them as the first tender document directory; Based on the second directory path, extract the last-level directory nodes contained in the second directory path and establish them as the second tender document directory; Extract the target directory relationship type between the first and second bid directories from the directory association types among the various bid directories; Define the target directory relationship type as a path association relationship type.

[0012] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0013] One embodiment of the present invention provides a tender document retrieval question-answering device based on directory association verification, comprising: an acquisition module, a directory parsing module, a statement association module, a text retrieval module, and a verification output module; The acquisition module is used to acquire the tender document text to be processed and the main question statement; The directory parsing module is used to parse the various directories contained in the tender document text to be processed, and to determine the directory association type between the various directories. The statement association module is used to extract related words from a preset keyword dictionary based on the main question statement; combine the related words to generate auxiliary question statements and determine the statement association relationship type between the main question statement and the auxiliary question statements; perform text retrieval on the tender text to be processed based on the main question statement and the auxiliary question statements to generate the first response text corresponding to the main question statement and the first directory path corresponding to the first response text in the tender text to be processed, as well as the second response text corresponding to the auxiliary question statement and the second directory path corresponding to the second response text in the tender text to be processed. The text retrieval module is used to extract the path association type between the tender directory corresponding to the first directory path and the tender directory corresponding to the second directory path based on the directory association type between each tender directory. The verification output module is used to compare the statement association type with the path association type; when the statement association type and the path association type are consistent, the first response text is output.

[0014] Based on the above method embodiments, the present invention provides corresponding electronic device embodiments.

[0015] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the tender retrieval question-answering method based on directory association verification as described in any of the above-described method embodiments.

[0016] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments.

[0017] One embodiment of the present invention provides a storage medium storing a computer program thereon, wherein, when the computer program is running, it controls the device where the storage medium is located to execute any of the above-described method embodiments of the tender retrieval question-answering method based on directory association verification.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a method, apparatus, electronic device, and storage medium for tender document retrieval-based question answering based on directory association verification. The method acquires the tender document text to be processed and the main question statement; parses the tender document directories in the text to be processed, determining the directory association relationships between the directories; extracts related words based on the main question statement and generates auxiliary question statements, determining the statement association relationship between the main question statement and the auxiliary question statements; retrieves the tender document text based on both the main question statement and the auxiliary question statements, obtaining the corresponding response text and its directory path; determines the path association relationship between the directory paths based on the directory association relationship; compares the statement association relationship with the path association relationship, and outputs the response text corresponding to the main question statement when they match.

[0019] This application's technical solution establishes the statement association type of the question intent by extending and generating auxiliary question statements, and simultaneously parses the tender document structure to establish the path association type of the directory where each response text is located, thus constructing a cross-validation mechanism between the question logic and the physical directory of the tender document. By comparing the statement association type and the path association type, and outputting the first response text when they match, this application overcomes the context misalignment defect caused by the lack of structural context verification due to the reliance on a single flat matching method in existing technologies. This solution ensures that the output retrieval text belongs to the directory branch that conforms to the question logic, thereby improving the retrieval accuracy and result reliability of the tender document question-and-answer system. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a tender document retrieval-based question-and-answer method based on directory association verification, provided by an embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the structure of a tender document retrieval question-and-answer device based on directory association verification provided in an embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 As shown, to address the problem in existing technologies where the lack of cross-validation between the document's physical directory structure and the question logic leads to contextual misalignment and low search accuracy, an embodiment of the present invention provides a tender document retrieval question-answering method based on directory association validation, comprising at least the following steps: Step S1: Obtain the tender document text and the main question statement to be processed; Specifically, the system receives the initial tender document and initial question string sent by the user. The initial tender document typically contains unformatted text and redundant characters. A text cleaning operation is performed on the initial tender document. This operation removes extra spaces, special characters, and formatting characters from the initial tender document, outputting the processed tender text. The same text cleaning operation is then used to structure the initial question string, outputting the main question statement.

[0024] To improve the accuracy of subsequent directory parsing and text retrieval, the process of obtaining the tender documents to be processed also includes determining the tender document type. A keyword library for each tender document type is pre-built. This pre-built keyword library includes construction tender documents, procurement tender documents, and service tender documents. A dedicated keyword set is established for each tender document type. The dedicated keyword set refers to a pre-defined set of characteristic business terms summarized and organized based on the characteristics of tender documents in different specific industries. To achieve accurate category classification, the dedicated keyword set configured for construction tender documents specifically covers standard industry terms such as "industrial park," "road," "building," "construction," and "infrastructure"; the dedicated keyword set configured for procurement tender documents specifically covers standard procurement terms such as "supply," "equipment procurement," "materials," "equipment," and "materials"; and the dedicated keyword set configured for service tender documents specifically covers standard business terms such as "operation and maintenance," "supervision," "consulting," "training," and "design."

[0025] Extract the title of the tender document from the text to be processed. Segment the title using a Chinese word segmentation algorithm to obtain a title word sequence. A Chinese word segmentation algorithm is a computational model that re-groups and synthesizes a word sequence from a continuous sequence of Chinese characters according to language norms. Match the title word sequence with all dedicated keyword sets, and calculate the number of keyword hits for different tender document types. The formula for calculating the number of keyword hits is as follows: In the formula, This indicates the tender document text to be processed, corresponding to the first... The number of keyword hits for each type of tender document. This indicates the total number of segments contained in the title segmentation sequence. Indicates the sequence number of the word segment. This indicates the first word in the title segmentation sequence. Each word. The serial number indicates the type of tender document. Indicates the first A set of exclusive keywords corresponding to each type of tender document. This represents the decision function. When word segmentation... Belongs to a dedicated set of keywords Internally, the value of the decision function is set to one; when word segmentation... Not part of the exclusive keyword set Internally, the value of the decision function is set to zero.

[0026] Compare the keyword hit counts for all bid document types. Filter out the bid document type with the highest keyword hit count. Establish the bid document type corresponding to the highest keyword hit count as the target bid document type for the bid document to be processed. If multiple highest keyword hit counts exist, or if the keyword hit counts for all bid document types are zero, the target bid document type for the bid document to be processed is defaulted to a general bid document. The target bid document type is used to assist in accurately applying the preset matching strategy.

[0027] By performing the above acquisition and processing steps, the standardized extraction and type identification of the original input data were completed, eliminating the negative impact of interfering characters on semantic understanding, and laying a highly standardized data foundation for the accurate parsing of directory relationships and text retrieval question answering.

[0028] Step S2: Parse the various bid directories contained in the bid document text to be processed, and determine the type of directory association relationship between the various bid directories; In a preferred embodiment, the various bid directories contained in the bid document text to be processed are parsed to determine the type of directory association between the various bid directories, including: The tender document text to be processed is combined in pairs to generate tender document index pairs; For each pair of tender documents, determine the first and second directory headings included in the current pair of tender documents; Perform text semantic parsing on the first and second directory titles to determine the first semantic category of the first directory title and the second semantic category of the second directory title. Structural feature identification is performed on the first directory title and the second directory title to determine the first directory level of the first directory title and the second directory level of the second directory title; The first directory title and the second directory title are divided into subject categories to determine the first subject to which the first directory title belongs and the second subject to which the second directory title belongs. Keyword attributes are extracted from the first and second directory titles to determine the interaction logic between them. Based on the first semantic category, the second semantic category, the first directory level, the second directory level, the first subject, the second subject, and the interaction logic, determine the type of directory association relationship between the current tender directory and the corresponding directory; The directory association type corresponding to each tender document directory is used as the directory association type between each tender document directory.

[0029] In a preferred embodiment, the preset set of association types includes subordinate relationships, parallel relationships, temporal relationships, causal relationships, complementary relationships, and response relationships; Based on the first semantic category, the second semantic category, the first directory level, the second directory level, the first subject, the second subject, and the interaction logic, determine the type of directory association relationship between the current tender document directory and the corresponding directory, including: When there is an overlap relationship between the first semantic category and the second semantic category, the current tender document directory is determined to be a subordinate relationship to the corresponding directory. If the current tender document directory pair does not belong to a subordinate relationship and the interaction logic is executed sequentially, then the directory association type corresponding to the current tender document directory pair is determined to be a sequential relationship. If the current tender document directory pair does not belong to a subordinate relationship, and the interaction logic is a basis-result correspondence logic, then the directory association type corresponding to the current tender document directory pair is determined to be a causal relationship. If the current tender document directory does not belong to a subordinate relationship and the interaction logic is a requirement and response correspondence logic, then the directory association type corresponding to the current tender document directory is determined to be a response relationship. If the current tender document directory does not belong to the subordinate, chronological, causal, or response relationship, and the first and second subjects are consistent, and the interaction logic is complementary, then the corresponding directory association type of the current tender document directory is determined to be complementary. If the current tender document directory does not belong to any of the above association types, the first semantic category and the second semantic category are independent of each other, and the first directory level and the second directory level are the same, then the directory association type corresponding to the current tender document directory is determined to be a parallel relationship. If the current tender document catalog does not meet the determination criteria for any of the above association types, it is determined that the current tender document catalog does not have an association.

[0030] In a preferred embodiment, before parsing the various bid catalogs contained in the bid document text to be processed, the method further includes a step of standardizing the bid catalogs: extracting a preset catalog template corresponding to the target bid document type from the data center; comparing each initial catalog title in the bid document text to be processed with each standardized catalog language in the catalog template, and calculating semantic similarity; selecting the standardized catalog language with the highest semantic similarity to replace the corresponding initial catalog title, so as to eliminate parsing errors caused by non-standard catalog naming in different bid documents.

[0031] Specifically, deep semantic understanding technology is used to parse the first and second directory titles, outputting the first semantic category corresponding to the first directory title and the second semantic category corresponding to the second directory title. Directory hierarchy regular expression matching rules and syntactic structure analysis are used to identify the first and second directory titles, extracting the first directory level of the first directory title and the second directory level of the second directory title. Hierarchical clustering algorithm is used to classify the first and second directory titles, establishing the first and second topics to which the first directory title belongs. Named entity recognition model is used to extract the key lexical attributes within the first and second directory titles, inferring the interaction logic between them.

[0032] The current tender list pairs are evaluated based on a predefined set of relationship types. This predefined set of relationship types includes subordinate, parallel, temporal, causal, complementary, and responsive relationships.

[0033] When the first semantic category completely covers the second semantic category, or the second semantic category completely covers the first semantic category, the current tender directory is determined to have a subordinate relationship with the corresponding directory. After establishing the subordinate relationship, the subordinate relationship degree between the first directory title and the second directory title is calculated. The calculation logic for the subordinate relationship degree is to obtain the proportion of the semantic vector of the first directory title covered by the semantic vector of the second directory title, or the proportion of the semantic vector of the second directory title covered by the semantic vector of the first directory title. The formula is expressed as: In the formula, Indicates the degree of dependence or association; This represents the set of semantic features corresponding to the first directory title; This represents the set of semantic features corresponding to the second directory title; This indicates the number of characteristic elements contained in the set. This indicates the intersection operation.

[0034] When the current tender document directory pair does not have a subordinate relationship, and the aforementioned interaction logic manifests as the sequential execution logic of business process stages, the directory association type corresponding to the current tender document directory pair is determined to be a sequential relationship. Sequential keywords are extracted from the first and second directory titles and assigned corresponding sequential weights to calculate the sequential relationship degree. The sequential weights are provided by a pre-constructed sequential keyword weight dictionary; for example, weights can be assigned to various sequential keywords as follows: preparation weight 0.1, implementation weight 0.3, acceptance weight 0.6, and maintenance weight 0.9, etc. The formula is expressed as: In the formula, Indicates the degree of temporal correlation. This represents the first weight value corresponding to the time-series keywords in the first directory title. This represents the second weight value corresponding to the time-series keywords in the second directory title. This represents a minimal constant to prevent the denominator from being zero.

[0035] When the current tender document directory pair does not have a subordinate relationship, and the interaction logic is a deductive logic based on the correspondence between the basis and the result, the directory association type corresponding to the current tender document directory pair is determined to be a causal relationship. Supporting keywords are extracted from the first and second directory titles, and the causal relationship degree is calculated based on the mapping relationship of these supporting keywords.

[0036] The demand response keyword mapping dictionary internally constructs a standard question-and-answer matching database for the specific context of bidding and tendering. This mapping dictionary specifically stores logical key-value pairs of associated data, such as "demand" mapping to "response", "requirement" mapping to "satisfaction", "indicator" mapping to "achievement", and "commitment" mapping to "fulfillment".

[0037] The formula is expressed as: In the formula, Indicates the degree of causal relationship. Supporting keywords for the first directory title. These are supporting keywords for the second section heading. This represents a mapping determination function. It outputs a value of one when there is a mapping match for the supporting keyword, and a value of zero when there is no match.

[0038] When the current tender document directory pair does not have a subordinate relationship, and the interaction logic is a correspondence between requirements and responses, the directory association type corresponding to the current tender document directory pair is determined to be a response relationship. Response-type keywords are extracted and combined with a requirement-response keyword mapping dictionary to calculate the response relationship correlation degree. The formula is expressed as: In the formula, This indicates the degree of relevance in the response relationship. Responsive keywords that represent the title of the first directory. Responsive keywords that indicate the title of the second directory. This represents the response determination function; a successful match outputs a value of 1, and a failure outputs a value of 0.

[0039] When the current tender document directory does not belong to a subordinate, chronological, causal, or response relationship, and the first and second related topics are completely identical, and the interaction logic exhibits content complementarity, the directory association type of the current tender document directory pair is determined to be complementary. In this case, the complementary relationship association degree is calculated by weighted summation of semantic similarity and topic domain overlap. The formula is expressed as: In the formula, Indicates the degree of correlation in a complementary relationship. This represents the semantic similarity score between the first directory title and the second directory title. This indicates the degree of overlap between the topic domains of the first and second topics. This represents the preset weight coefficient corresponding to semantic similarity. This represents the preset weight coefficient corresponding to the topic domain overlap. In a specific embodiment, the weights of both semantic similarity and topic domain overlap are preset to 0.5 (i.e., , ).

[0040] When the current tender document directory does not belong to any of the above association types, the first semantic category and the second semantic category are independent and have no overlap, and the first directory level and the second directory level are at the same level, the directory association type corresponding to the current tender document directory is determined to be a parallel relationship. The degree of parallel relationship association is calculated based on the overlap between the subject areas of the first directory title and the subject areas of the second directory title. The formula is expressed as: In the formula, Indicates the degree of correlation between parallel relationships. This represents the set of features contained in the first subject. This represents the set of features contained in the second subject. This indicates the union operation.

[0041] If the current tender document catalog does not meet any of the above-mentioned criteria for determining the type of association, it is directly determined that there is no association between the current tender document catalog and no relationship degree analysis is performed on the data without association. All independently calculated association degrees are normalized to unify the units of measurement.

[0042] Iterate through all bid document directory pairs in the bid document text to be processed, integrate the corresponding directory association types of all identified bid document directory pairs, and finally generate the directory association types between each bid document directory.

[0043] By performing the above-mentioned multi-dimensional directory relationship parsing and association quantification operations, the limitations of conventional retrieval relying solely on local text features are overcome. The flat arrangement of tender documents is transformed into a structured knowledge network with rigorous logical connections, which greatly improves the accuracy and reliability when dealing with complex cross-clause questions and answers.

[0044] Step S3: Based on the main question statement, extract related words from the preset keyword dictionary; combine the related words to generate auxiliary question statements, and determine the type of sentence association between the main question statement and the auxiliary question statements; In a preferred embodiment, based on the main question statement, related words are extracted from a preset keyword dictionary, including: Feature extraction is performed on the main question statement to generate initial action keywords and initial entity keywords contained in the main question statement; Calculate the similarity between the initial action keywords and each action word in the preset keyword dictionary, and generate semantic similarity for each action; Calculate the similarity between the initial entity keywords and each entity word in the preset keyword dictionary, and generate the semantic similarity of each entity; The action with the highest semantic similarity score is selected from all action semantic similarities to generate the maximum action semantic similarity score; the action words corresponding to the maximum action semantic similarity score are established as standard action keywords. The entity with the highest semantic similarity score is selected from all entity semantic similarity scores to generate the maximum entity semantic similarity score; the entity words corresponding to the maximum entity semantic similarity score are established as standard entity keywords. Extract a set of related action words that are associated with the standard action keywords from the pre-defined keyword dictionary; Extract a set of related entity words that are associated with standard entity keywords from a pre-defined keyword dictionary; Combine the set of related action words and the set of related entity words to generate each related word.

[0045] In a preferred embodiment, the associated words are combined to generate an auxiliary question statement, and the statement association type between the main question statement and the auxiliary question statement is determined, including: Each related keyword is paired with its associated action words and associated entity words to generate keyword combinations. Input each keyword combination into the preset statement generation template to generate each alternative auxiliary question statement; Get the selection command; Based on the selection instructions, determine the auxiliary question statement from each candidate auxiliary question statement, and determine the target action vocabulary and target entity vocabulary contained in the auxiliary question statement; Extract the action relationship types between target action words and standard action keywords, as well as the entity relationship types between target entity words and standard entity keywords from the preset keyword dictionary; Based on the action relationship type and entity relationship type, determine the statement association type between the main question statement and the auxiliary question statements.

[0046] Specifically, after receiving the main question statement, part-of-speech tagging (POS) is used to extract features from it. POS tagging is a natural language processing method that assigns a specific part-of-speech tag to each word in a text sequence according to grammatical rules and contextual information. POS tagging is used to accurately identify the initial action keywords and initial entity keywords contained in the main question statement.

[0047] After extraction, the similarity between the initial action keywords and the various action words distributed in the preset keyword dictionary is calculated, and the semantic similarity of each action is output. The preset keyword dictionary includes an action keyword dictionary and an entity keyword dictionary. The specific construction logic of the preset keyword dictionary is as follows: massive amounts of historical tender documents, bidding documents, contract terms, and frequently asked questions and answers in bidding are obtained from the data center, and these texts are classified. Then, high-frequency words in various standard texts are extracted, and semantically consistent words are uniformly mapped and merged to obtain a standard vocabulary. In the actual application of automated tender processing scenarios, the action words pre-stored in the action keyword dictionary specifically cover standardized operational action words such as "provide," "pay," "affix," "seal," and "sign"; the entity words pre-stored in the entity keyword dictionary specifically cover standardized objective entity words such as "business license," "deposit," "official seal," "legal representative," and "audit report."

[0048] The formula for calculating action semantic similarity is as follows: In the formula, Indicates the semantic similarity of actions. This represents the multidimensional feature vector corresponding to the initial action keyword. This represents the multidimensional feature vector corresponding to action words in the predefined keyword dictionary. This represents the vector dot product operation. This represents the magnitude of the multidimensional feature vector corresponding to the initial action keyword. This represents the magnitude of the multidimensional feature vector corresponding to action words in the predefined keyword dictionary.

[0049] Using the same calculation principle, the similarity between the initial entity keywords and each entity word distributed in the preset keyword dictionary is calculated, and the semantic similarity of each entity is output. The formula for calculating the semantic similarity of entities is as follows: In the formula, Indicates the semantic similarity of entities. This represents the multidimensional feature vector corresponding to the initial entity keywords. This represents the multidimensional feature vector corresponding to the entity words in the predefined keyword dictionary. This represents the magnitude of the multidimensional feature vector corresponding to the initial entity keyword. This represents the magnitude of the multidimensional feature vector corresponding to the entity words in the preset keyword dictionary.

[0050] Compare all action semantic similarities and select the action semantic similarity with the highest value as the maximum action semantic similarity. Establish the action vocabulary bound to the maximum action semantic similarity as the standard action keyword. Similarly, iterate through all entity semantic similarities and extract the entity semantic similarity with the highest value to generate the maximum entity semantic similarity. Establish the entity vocabulary corresponding to the maximum entity semantic similarity as the standard entity keyword. This standardization process effectively eliminates the semantic bias risk caused by colloquial expressions.

[0051] Next, based on standard action keywords and standard entity keywords, sets of related action words and sets of related entity words with preset logical dimensions are extracted from a pre-defined keyword dictionary. Elements within these sets are combined to generate related words. Then, the related action words and related entity words within each related word are extracted and cross-matched pairwise to generate keyword combinations.

[0052] Input all keyword combinations into the preset statement generation template to generate corresponding alternative auxiliary question statements. The preset statement generation template refers to a pre-defined standardized supplementary question-and-answer sentence structure that includes entity placeholders and action placeholders.

[0053] As a specific implementation of sentence structure, the preset sentence generation template is configured with fixed text frames such as "Confirm the relevant [action placeholder] requirements for [entity placeholder]" and "Verify the [action placeholder] standards for [entity placeholder]". During the generation operation, the extracted related action words and related entity words are automatically filled into the corresponding placeholder coordinates, thereby generating a structurally complete independent question and answer sentence.

[0054] Obtain the selection command triggered by the user interface. Based on the content of the selection command, select a specific auxiliary question statement from among the candidate auxiliary question statements, and identify the target action words and target entity words embedded in the selected auxiliary question statement.

[0055] Source matching is performed in a pre-defined keyword dictionary to extract the action relationship type formed by the target action words and standard action keywords, and the entity relationship type formed by the target entity words and standard entity keywords. Based on the obtained action relationship type and entity relationship type, comprehensive mapping inference is performed to determine the statement association type between the main question statement and the auxiliary question statements. The statement association type covers subordinate relationship, parallel relationship, temporal relationship, causal relationship, complementary relationship, and response relationship.

[0056] By standardizing and parsing the main question statement and extending it to generate auxiliary question statements with related logical support, the above processing steps effectively expand the semantic coverage dimension of the original question. At the same time, they construct a multi-perspective question logic benchmark with clear primary and secondary elements, thereby significantly reducing the probability of information omission caused by a single question and laying a solid data foundation for subsequent accurate text cross-verification based on the directory logic structure.

[0057] Step S4: Based on the main question statement and auxiliary question statements, perform text retrieval on the tender document text to be processed, and generate the first response text corresponding to the main question statement and the first directory path of the first response text in the tender document text to be processed, as well as the second response text corresponding to the auxiliary question statements and the second directory path of the second response text in the tender document text to be processed. In a preferred embodiment, text retrieval is performed on the tender document text to be processed based on the main question statement and auxiliary question statements to generate a first response text corresponding to the main question statement and a first directory path corresponding to the first response text in the tender document text to be processed, as well as a second response text corresponding to the auxiliary question statements and a second directory path corresponding to the second response text in the tender document text to be processed, including: Calculate the similarity between the main statement of the question and each text paragraph in the tender document to be processed, and generate the first text similarity score; The text with the highest numerical similarity score is used to generate the maximum first text similarity score. The text paragraph with the highest first text similarity will be used as the first response text corresponding to the main question statement; Based on the tender document text to be processed, locate the first directory node to which the first response text belongs in the tender document text to be processed; Based on the first directory node, determine the hierarchical sequence of the first directory node and generate the first directory path corresponding to the first response text in the tender document text to be processed; Calculate the similarity between the auxiliary question statement and each text paragraph in the tender document to be processed, and generate the similarity scores for each second text. The second text with the highest numerical similarity is used to generate the maximum second text similarity. The text paragraph with the highest second text similarity is designated as the second response text corresponding to the auxiliary question statement; Based on the tender document text to be processed, locate the second directory node to which the second response text belongs in the tender document text to be processed; Determine the hierarchical sequence of the second directory node based on the second directory node; Based on the hierarchical sequence of the second directory node, the second directory path corresponding to the second response text in the tender document text to be processed is generated.

[0058] Specifically, after obtaining the main question statement and auxiliary question statements, a deep text retrieval operation is initiated on the tender document text to be processed. The tender document text is segmented to extract individual text segments. Using a pre-deployed deep semantic coding model, multi-dimensional feature vectors corresponding to the main question statement and each text segment are generated. As a preferred implementation, the deep semantic coding model can employ pre-trained language models such as BERT or Sentence-BERT. A deep semantic coding model refers to a deep learning network architecture capable of mapping natural language sequences into a high-dimensional spatial representation with continuous numerical features. Based on the extracted multi-dimensional feature vectors, the similarity between the main question statement and each text segment in the tender document text is calculated, and all first text similarities are output and summarized. The formula for calculating the first text similarity is as follows: In the formula, This indicates the first text similarity. This represents the multidimensional feature vector corresponding to the main statement of the question. This represents the multidimensional feature vector corresponding to a single text paragraph in the tender document to be processed. This represents the magnitude of the multidimensional feature vector corresponding to the main query statement. This represents the magnitude of the multidimensional feature vector corresponding to a single text paragraph in the tender document to be processed.

[0059] In a preferred embodiment, to further improve the accuracy of the retrieval, while using a deep semantic coding model to calculate vector semantic similarity, keyword exact matching technology can also be combined. That is, the main query statement is matched literally with each text paragraph to obtain an exact matching score; then, the exact matching score is weighted and fused with the first text similarity to obtain the final comprehensive retrieval score, which serves as the final basis for selecting the first response text.

[0060] All the collected first-text similarity scores are iterated and compared numerically. The first-text similarity score with the highest value is selected as the maximum first-text similarity score. The text paragraph bound to the maximum first-text similarity score is extracted from the tender document text to be processed. This extracted text paragraph is directly established as the first response text corresponding to the main question statement. The first response text represents the objective content fragment that best matches the main question statement in terms of semantic matching.

[0061] Based on the physical structure information of the tender document to be processed, the absolute paragraph position of the first response text within the tender document is obtained. The overlap between the absolute paragraph position of the first response text and the text coverage areas of each tender directory in the tender document to be processed is compared to locate the tender directory containing the absolute paragraph position and at the bottom of the hierarchy. This located tender directory is established as the first directory node to which the first response text belongs. Based on the first directory node, the corresponding parent directory nodes at each level are extracted upwards. The parent directory nodes at each level are concatenated with the first directory node according to the hierarchical distribution order to determine the hierarchical sequence of the first directory node, thereby generating the first directory path corresponding to the first response text in the tender document to be processed. The hierarchical sequence refers to the structured chain of ownership formed by the top-level root directory to the bottom-level directory.

[0062] Parallel processing logic is employed to extract multi-dimensional feature vectors corresponding to the auxiliary question statements. The similarity between the auxiliary question statements and each text paragraph in the tender document to be processed is calculated to generate all second text similarity scores. The formula for calculating the second text similarity is as follows: In the formula, Indicates the similarity between the second and third texts; This represents the multidimensional feature vector corresponding to the auxiliary question statement; This represents the multidimensional feature vector corresponding to a single text paragraph in the tender document to be processed. This represents the magnitude of the multidimensional feature vector corresponding to the auxiliary question statement; This represents the magnitude of the multidimensional feature vector corresponding to a single text paragraph in the tender document to be processed.

[0063] Compare all the second text similarities and select the one with the highest value as the maximum second text similarity. Then, designate the text paragraph corresponding to the maximum second text similarity as the second response text for the auxiliary question statement.

[0064] Based on the global directory tree information of the tender document to be processed, the actual second directory node to which the second response text belongs within the tender document is located. According to the second directory node, all corresponding parent levels are traced level by level to determine the hierarchical sequence of the second directory node. Based on the hierarchical sequence of the second directory node, a concatenation operation is performed in descending order of hierarchy to generate the second directory path corresponding to the second response text within the tender document to be processed. The second directory path is used to represent the precise coordinates of the second response text within the physical structure of the tender document.

[0065] Through the rigorous text retrieval and path tracing steps described above, the unstructured natural language question-and-answer intent was transformed into a precise location operation within a structured document space. This not only outputs a highly semantically consistent response result but also simultaneously binds the actual physical structure and hierarchical relationship of the response text, providing complete and objective data evidence for subsequent logical cross-validation using a directory architecture.

[0066] Step S5: Based on the directory association type between each tender document directory, extract the path association type between the tender document directory corresponding to the first directory path and the tender document directory corresponding to the second directory path; In a preferred embodiment, based on the directory association type between each tender document directory, the path association type between the tender document directory corresponding to the first directory path and the tender document directory corresponding to the second directory path is extracted, including: Based on the first directory path, extract the last-level directory nodes contained in the first directory path and establish them as the first tender document directory; Based on the second directory path, extract the last-level directory nodes contained in the second directory path and establish them as the second tender document directory; Extract the target directory relationship type between the first and second bid directories from the directory association types among the various bid directories; Define the target directory relationship type as a path association relationship type.

[0067] Specifically, the process involves obtaining the directory relationship types between the pre-parsed tender document directories. For the generated first directory path, a node hierarchy parsing operation is performed. Since the first directory path is a structured long sequence generated by concatenating the top-level root node to the bottom-level last-level node, all node positions within the first directory path are traversed in descending order of hierarchy. The last-level directory node at the bottom level that directly contains the first response text is extracted. This last-level directory node within the first directory path is established as the first tender document directory. A last-level directory node refers to a leaf-end node object in the tree-like document structure that does not derive further sub-levels, representing the most direct physical affiliation attribute of the retrieved text paragraph.

[0068] The second directory path is processed using the same node parsing logic. Traversal and retrieval are performed downwards along the hierarchical distribution of the second directory path to extract the final-level directory node within it. This final-level directory node is explicitly designated as the second tender directory. The operations of establishing the first and second tender directories successfully reduce the dimensionality of complex and varying-length path sequences into comparison nodes based on a single benchmark, avoiding the computational overhead caused by directly comparing lengthy paths.

[0069] After completing the dimensionality reduction transformation at the node level, the global directory relationship data set already constructed in the previous steps is invoked. In this data set covering the directory relationship types between various tender document directories, the first and second tender document directories are used as joint retrieval identifiers for cross-matching. Specifically, the process of extracting the target directory relationship type from the set of directory relationship types between various tender document directories employs a matrix addressing matching mechanism. The relationship types between each level of the tender document directories are pre-determined through the aforementioned directory parsing steps and stored in a structured database to construct a relationship mapping matrix storing the global relationship network data. The position indexes corresponding to the first and second tender document directories are input to accurately extract and lock the target directory relationship type. The extraction and matching formula for the target directory relationship type is as follows: In the formula, Indicates the target directory relationship type. This represents a pre-constructed association mapping matrix used to store the directory association types between various tender directories. This represents the row dimension index value corresponding to the first tender document directory in the association mapping matrix. This represents the column dimension index value corresponding to the second tender document directory in the association mapping matrix.

[0070] The target directory relationship type objectively reflects the inherent logical connection between the first and second response texts within the original physical structure of the tender document. The extracted target directory relationship type is formally established as the path association relationship type. This path association relationship type will serve as the absolute benchmark for subsequent cross-validation mechanisms between execution intent and physical structure.

[0071] By performing the above node-based source tracing and matrix addressing extraction operations, the abstract and incomputable physical location of the search segment was successfully transformed into a standard relational feature that can be used for mathematical logic comparison, which greatly improved the execution efficiency and objective accuracy of the subsequent cross-validation process.

[0072] Step S6: Compare the statement association type with the path association type; when the statement association type and the path association type are consistent, output the first response text.

[0073] Specifically, the identified statement and path relationship types are extracted, followed by a rigorous logical consistency comparison. To perform cross-validation along the logical dimension, a feature matching function is introduced to quantify the comparison results. The feature matching function is calculated as follows: In the formula, This represents the numerical value of the output result of the logical comparison verification. This indicates the type of statement relationship between the main question statement and the auxiliary question statements. This indicates the type of path association between the first tender directory and the second tender directory. This represents the feature matching determination function. When the statement association type and the path association type are completely consistent, the feature matching determination function outputs a value of one; when the statement association type and the path association type are inconsistent, the feature matching determination function outputs a value of zero.

[0074] The subsequent distribution action is determined based on the output value of the logical comparison verification. When the output value of the logical comparison verification is one, it indicates that the statement association type and the path association type are consistent. This consistency indicates that the actual logical position of the first response text corresponding to the main question statement within the physical structure of the tender document to be processed perfectly matches the pre-constructed question logic intent. At this point, the verification passes, and the first response text is executed as the search answer corresponding to the main question statement, and the corresponding original text segment to which the first response text belongs in the tender document to be processed is simultaneously displayed to the user.

[0075] When the output value of the logical comparison verification is zero, it indicates that the statement association type and the path association type are inconsistent. This inconsistency means that even if the first response text has a high degree of matching with the main query statement at the literal semantic level, the constraints represented by the chapter position of the first response text have seriously deviated from the user's true query intent, constituting a contextual misalignment. In this case, the verification fails, the output link of the first response text is immediately intercepted, and a search failure message is generated and displayed.

[0076] For example, consider a specific verification scenario: If a user's main question is "What is the total construction period of the road construction project?", and the auxiliary question is "What are the schedule requirements for the lighting construction project within the road construction project?", then there is a subordinate relationship between the main question and the auxiliary question. During retrieval, the first response text may yield two results: Result A is "Total construction period 120 days," with the corresponding first directory path being "Instructions to Bidders → Contract Terms → Total Construction Period Requirements → Schedule Requirements" (this path is parallel to the path containing the auxiliary question); Result B is "Total construction period 180 days," with the corresponding first directory path being "Instructions to Bidders → Contract Terms → Total Construction Period Requirements" (this path is subordinate to the path containing the auxiliary question). At this point, a comparison reveals that the path association (parallel) of result A is inconsistent with the statement association (subordinate), indicating that 120 days may be the construction period for partial civil engineering or lighting work, resulting in a contextual misalignment, and thus the result is blocked; while the path association (subordinate) of result B is consistent with the statement association (subordinate), thus the verification passes, and the correct answer "total construction period 180 days" is finally output.

[0077] By implementing the cross-validation mechanism between the question intent logic and the physical structure of the text, a rigorous data closed-loop filtering condition was successfully constructed, completely eliminating the defect of irrelevant answers caused by single flat semantic comparison, and significantly improving the accuracy of long text retrieval question-and-answer results and the reliability of the final output response content.

[0078] Based on the above method embodiments, the present invention provides corresponding apparatus embodiments.

[0079] like Figure 2 As shown, an embodiment of the present invention provides a tender document retrieval question-answering device based on directory association verification, including: an acquisition module 101, a directory parsing module 102, a statement association module 103, a text retrieval module 104, and a verification output module 105; The acquisition module 101 is used to acquire the tender document text to be processed and the main question statement; The directory parsing module 102 is used to parse the various directories contained in the tender document text to be processed, and to determine the directory association relationship type between the various directories. The statement association module 103 is used to extract related words from a preset keyword dictionary based on the main question statement; combine the related words to generate auxiliary question statements and determine the statement association relationship type between the main question statement and the auxiliary question statements; perform text retrieval on the tender text to be processed based on the main question statement and the auxiliary question statements, and generate the first response text corresponding to the main question statement and the first directory path corresponding to the first response text in the tender text to be processed, as well as the second response text corresponding to the auxiliary question statement and the second directory path corresponding to the second response text in the tender text to be processed. The text retrieval module 104 is used to extract the path association type between the tender directory corresponding to the first directory path and the tender directory corresponding to the second directory path based on the directory association type between each tender directory. The verification output module 105 is used to compare the statement association type with the path association type; when the statement association type and the path association type are consistent, the first response text is output.

[0080] It should be noted that the embodiments of the device described above correspond to the embodiments of the present invention described above, and can realize the tender retrieval question-answering method based on directory association verification described above in any one of the present invention. Furthermore, the embodiments of the device described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between modules indicates that they have a communication connection, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without creative effort.

[0081] Based on the above-described method embodiments of the present invention, a corresponding embodiment of an electronic device is provided.

[0082] An embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the tender retrieval question-answering method based on directory association verification as described in any one of the present invention, or the processor executes the computer program to implement the functions of each module in the above-described device embodiments.

[0083] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the terminal device.

[0084] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0085] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0086] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0087] Based on the above method embodiments, the present invention provides corresponding storage medium embodiments; Another embodiment of the present invention provides a storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute any of the above-described directory association verification-based tender retrieval question-answering methods of the present invention.

[0088] The aforementioned storage medium is a computer-readable storage medium, and the computer program includes computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0089] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0090] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method of RFP (Request for Proposal) search question and answer based on directory association verification, characterized in that, include: Retrieve the tender document text and the main question statement to be processed; The tender document text to be processed is parsed to determine the type of directory relationship between the tender documents. Based on the main statement of the question, extract the relevant words from the preset keyword dictionary; Combine the related words to generate auxiliary question statements, and determine the type of statement association between the main question statement and the auxiliary question statements; Based on the main question statement and auxiliary question statements, a text search is performed on the tender document text to be processed, generating the first response text corresponding to the main question statement and the first directory path of the first response text in the tender document text to be processed, as well as the second response text corresponding to the auxiliary question statements and the second directory path of the second response text in the tender document text to be processed. Based on the directory association type between each tender document directory, the path association type between the tender document directory corresponding to the first directory path and the tender document directory corresponding to the second directory path is extracted; Compare the types of relational relationships between statements and the types of relational relationships between paths; When the statement association type and the path association type are the same, output the first response text.

2. The catalog association verification based bid request-answer method of claim 1, wherein, The tender document text to be processed is parsed to determine the types of directory relationships between the tender document directories, including: The tender document text to be processed is combined in pairs to generate tender document index pairs; For each pair of tender documents, determine the first and second directory headings included in the current pair of tender documents; Perform text semantic parsing on the first and second directory titles to determine the first semantic category of the first directory title and the second semantic category of the second directory title. Structural feature identification is performed on the first directory title and the second directory title to determine the first directory level of the first directory title and the second directory level of the second directory title; The first directory title and the second directory title are divided into subject categories to determine the first subject to which the first directory title belongs and the second subject to which the second directory title belongs. Keyword attributes are extracted from the first and second directory titles to determine the interaction logic between them. Based on the first semantic category, the second semantic category, the first directory level, the second directory level, the first subject, the second subject, and the interaction logic, determine the type of directory association relationship between the current tender directory and the corresponding directory; The directory association type corresponding to each tender document directory is used as the directory association type between each tender document directory.

3. The catalog association verification based bid request-answer method of claim 2, wherein, The preset set of relationship types includes subordinate, parallel, temporal, causal, complementary, and responsive relationships; Based on the first semantic category, the second semantic category, the first directory level, the second directory level, the first subject, the second subject, and the interaction logic, determine the type of directory association relationship between the current tender document directory and the corresponding directory, including: When there is an overlap relationship between the first semantic category and the second semantic category, the current tender document directory is determined to be a subordinate relationship to the corresponding directory. If the current tender document directory pair does not belong to a subordinate relationship and the interaction logic is executed sequentially, then the directory association type corresponding to the current tender document directory pair is determined to be a sequential relationship. If the current tender document directory pair does not belong to a subordinate relationship, and the interaction logic is a basis-result correspondence logic, then the directory association type corresponding to the current tender document directory pair is determined to be a causal relationship. If the current tender document directory does not belong to a subordinate relationship and the interaction logic is a requirement and response correspondence logic, then the directory association type corresponding to the current tender document directory is determined to be a response relationship. If the current tender document directory does not belong to the subordinate, chronological, causal, or response relationship, and the first and second subjects are consistent, and the interaction logic is complementary, then the corresponding directory association type of the current tender document directory is determined to be complementary. If the current tender document directory does not belong to any of the above association types, the first semantic category and the second semantic category are independent of each other, and the first directory level and the second directory level are the same, then the directory association type corresponding to the current tender document directory is determined to be a parallel relationship. If the current tender document catalog does not meet the determination criteria for any of the above association types, it is determined that the current tender document catalog does not have an association.

4. The tender document retrieval question-answering method based on directory association verification as described in claim 3, characterized in that, Based on the main statement of the question, extract related words from the pre-defined keyword dictionary, including: Feature extraction is performed on the main question statement to generate initial action keywords and initial entity keywords contained in the main question statement; Calculate the similarity between the initial action keywords and each action word in the preset keyword dictionary, and generate semantic similarity for each action; Calculate the similarity between the initial entity keywords and each entity word in the preset keyword dictionary, and generate the semantic similarity of each entity; The action with the highest semantic similarity score is selected from all action semantic similarities to generate the maximum action semantic similarity score; the action words corresponding to the maximum action semantic similarity score are established as standard action keywords. The entity with the highest semantic similarity score is selected from all entity semantic similarity scores to generate the maximum entity semantic similarity score; the entity words corresponding to the maximum entity semantic similarity score are established as standard entity keywords. Extract a set of related action words that are associated with the standard action keywords from the pre-defined keyword dictionary; Extract a set of related entity words that are associated with standard entity keywords from a pre-defined keyword dictionary; Combine the set of related action words and the set of related entity words to generate each related word.

5. The tender document retrieval question-answering method based on directory association verification as described in claim 4, characterized in that, Combine related words to generate auxiliary question statements, and determine the type of statement relationship between the main question statement and the auxiliary question statements, including: Each related keyword is paired with its associated action words and associated entity words to generate keyword combinations. Input each keyword combination into the preset statement generation template to generate each alternative auxiliary question statement; Get the selection command; Based on the selection instructions, determine the auxiliary question statement from each candidate auxiliary question statement, and determine the target action vocabulary and target entity vocabulary contained in the auxiliary question statement; Extract the action relationship types between target action words and standard action keywords, as well as the entity relationship types between target entity words and standard entity keywords from the preset keyword dictionary; Based on the action relationship type and entity relationship type, determine the statement association type between the main question statement and the auxiliary question statements.

6. The tender document retrieval question-answering method based on directory association verification as described in claim 5, characterized in that, Based on the main question statement and auxiliary question statements, a text search is performed on the tender document text to be processed, generating the first response text corresponding to the main question statement and the first directory path of the first response text in the tender document text to be processed, as well as the second response text corresponding to the auxiliary question statements and the second directory path of the second response text in the tender document text to be processed, including: Calculate the similarity between the main statement of the question and each text paragraph in the tender document to be processed, and generate the first text similarity score; The text with the highest numerical similarity score is used to generate the maximum first text similarity score. The text paragraph with the highest first text similarity will be used as the first response text corresponding to the main question statement; Based on the tender document text to be processed, locate the first directory node to which the first response text belongs in the tender document text to be processed; Based on the first directory node, determine the hierarchical sequence of the first directory node and generate the first directory path corresponding to the first response text in the tender document text to be processed; Calculate the similarity between the auxiliary question statement and each text paragraph in the tender document to be processed, and generate the similarity scores for each second text. The second text with the highest numerical similarity is used to generate the maximum second text similarity. The text paragraph with the highest second text similarity is designated as the second response text corresponding to the auxiliary question statement; Based on the tender document text to be processed, locate the second directory node to which the second response text belongs in the tender document text to be processed; Determine the hierarchical sequence of the second directory node based on the second directory node; Based on the hierarchical sequence of the second directory node, the second directory path corresponding to the second response text in the tender document text to be processed is generated.

7. The tender document retrieval question-answering method based on directory association verification as described in claim 6, characterized in that, Based on the directory association types between the various tender document directories, the path association types between the tender document directories corresponding to the first directory path and the tender document directories corresponding to the second directory path are extracted, including: Based on the first directory path, extract the last-level directory nodes contained in the first directory path and establish them as the first tender document directory; Based on the second directory path, extract the last-level directory nodes contained in the second directory path and establish them as the second tender document directory; Extract the target directory relationship type between the first and second bid directories from the directory association types among the various bid directories; Define the target directory relationship type as a path association relationship type.

8. A tender document retrieval question-and-answer device based on directory association verification, characterized in that, include: The module includes an acquisition module, a directory parsing module, a statement association module, a text retrieval module, and a verification output module. The acquisition module is used to acquire the tender document text to be processed and the main question statement; The directory parsing module is used to parse the various directories contained in the tender document text to be processed, and to determine the directory association type between the various directories. The statement association module is used to extract related words from a preset keyword dictionary based on the main statement of the question. Combine the related words to generate auxiliary question statements, and determine the type of statement association between the main question statement and the auxiliary question statements; Based on the main question statement and auxiliary question statements, a text search is performed on the tender document text to be processed, generating the first response text corresponding to the main question statement and the first directory path of the first response text in the tender document text to be processed, as well as the second response text corresponding to the auxiliary question statements and the second directory path of the second response text in the tender document text to be processed. The text retrieval module is used to extract the path association type between the tender directory corresponding to the first directory path and the tender directory corresponding to the second directory path based on the directory association type between each tender directory. The verification output module is used to compare the statement association type with the path association type; When the statement association type and the path association type are the same, output the first response text.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the tender retrieval question-answering method based on directory association verification as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the storage medium is located to perform the tender retrieval question-answering method based on directory association verification as described in any one of claims 1 to 7.

Images