Bid document abstract generation method and system based on large language model

Abstract

The invention provides a bidding file abstract generation method and system based on a large language model, and relates to the technical field of natural language process.The method comprises the steps that firstly, domain terms are extracted from an input to-be-processed bidding file set to construct a term library, and file units are split to form a domain semantic chain; inputting the term library into a pre-trained large language model to adjust the semantic analysis weight so as to obtain a semantic analysis model adaptive to the field; inputting semantic nodes in the semantic chain into the semantic analysis model to extract core expressions so as to form a dynamic abstract fragment set; determining an abstract fragment association sequence according to a semantic node connection relationship, and joining to obtain a preliminary fusion abstract text; and finally, semantic feedback of the reader is received, abstract fragment expressions and sequences are adjusted, a semantic chain is updated, and a final abstract is generated and output, so that the high-quality bidding document abstract can be efficiently generated.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and more specifically, to a method and system for generating bid document summaries based on a large language model. Background Technology

[0002] In the bidding field, tender documents contain a wealth of complex and specialized information, covering project proposals, technical specifications, commercial terms, and more. For readers, quickly extracting key information from this vast amount of content is an extremely challenging task. Traditional methods of generating tender document summaries primarily rely on manual extraction, which not only requires significant manpower and time but also struggles to guarantee accuracy and comprehensiveness due to variations in human understanding and subjectivity. With the development of natural language processing technology, some automatic summarization methods based on general language models have begun to be applied to tender document processing. However, these general models lack a deep understanding of the specialized knowledge required for the tender document domain and cannot accurately grasp the key terms and semantic logical relationships within the tender documents. This results in inconsistent quality of the generated summaries, failing to adequately meet the needs of tender document readers for quickly and accurately obtaining core information. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for generating bid document summaries based on a large language model, the method comprising: Domain terms are extracted from the input set of tender documents to be processed to form a domain terminology library for tender documents. Based on the terminology associations in each tender document unit, each tender document unit is split into semantic nodes. The semantic nodes are connected through terminology associations to form a domain semantic chain for tender documents. The domain terminology database of the tender document is input into the pre-trained large language model, and the semantic parsing weights of the pre-trained large language model are adjusted to adapt the pre-trained large language model to the semantic parsing requirements of the tender document domain, thus obtaining a domain-adapted semantic parsing model. Each semantic node in the semantic chain of the tender document domain is input into the domain-adapted semantic parsing model. The core expression content of each semantic node is extracted through the domain-adapted semantic parsing model to form a dynamic summary fragment corresponding to each semantic node. All dynamic summary fragments are integrated to obtain a dynamic summary fragment set. Based on the connection relationship of semantic nodes in the semantic chain of the tender document domain, the association order of each dynamic summary fragment in the dynamic summary fragment set is determined, and the semantic connection processing of each dynamic summary fragment is performed according to the association order to obtain the preliminary fused summary text; The system receives semantic feedback from the tender document reader regarding the preliminary fused summary text, adjusts the content and order of each dynamic summary segment based on the semantic feedback, updates the semantic chain of the tender document domain, generates the final tender document summary, and outputs it to the target receiving end.

[0004] In another aspect, the present invention also provides a tender document summary generation system based on a large language model, including a processor and a machine-readable storage medium connected to the processor. The machine-readable storage medium is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the machine-readable storage medium to implement the above-mentioned method.

[0005] Based on the above, this invention extracts domain terms from a set of tender documents and constructs a domain terminology database. Simultaneously, it decomposes tender document units based on terminology relationships to form domain semantic chains. The domain terminology database is input into a pre-trained large language model, and the semantic parsing weights are adjusted to obtain a semantic parsing model adapted to the semantic parsing needs of the tender document domain. This allows the semantic parsing model to deeply understand the professional semantics in the tender documents, improving the accuracy of semantic parsing. Using this semantic parsing model, the core content of semantic nodes is extracted to form a dynamic summary fragment set, accurately capturing the key information of each semantic node. The association order of the dynamic summary fragments is determined based on the semantic node connection relationships, and semantic connection processing is performed to obtain a preliminary fused summary text, ensuring the coherence and logic of the summary. Reader semantic feedback is received to adjust the content and association order of the summary fragments and update the domain semantic chain to generate the final summary. This achieves personalized customization and continuous optimization of the summary. The final generated tender document summary accurately and comprehensively reflects the core content of the tender documents, greatly facilitating readers to quickly obtain key information and improving the efficiency and quality of bidding work. Attached Figure Description

[0006] Figure 1 This is a schematic diagram of the execution flow of the bid document summary generation method based on a large language model provided in an embodiment of the present invention.

[0007] Figure 2 This is a schematic diagram of exemplary hardware and software components of the tender document summary generation system based on a large language model provided in an embodiment of the present invention.

[0008] In the diagram, 100 is the tender document summary generation system based on a large language model, 110 is the network port, 120 is the processor, 130 is the communication bus, 140 is the storage medium, and 150 is the I / O interface. Detailed Implementation

[0009] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for generating bid document summaries based on a large language model, provided in one embodiment of the present invention. The following is a detailed description of this method for generating bid document summaries based on a large language model.

[0010] Step S110: Extract domain terms from the input set of tender documents to be processed to form a domain terminology library for tender documents. Based on the terminology associations in each tender document unit, split each tender document unit into semantic nodes and connect the semantic nodes through terminology associations to form a domain semantic chain for tender documents.

[0011] In this embodiment, the set of tender documents to be processed is the tender documents for the "Industrial Plant Steel Structure Installation Project," which includes four tender document units: project requirements response document, technical implementation plan document, commercial quotation document, and construction organization plan document. A semantic chain tailored to the industrial steel structure installation field needs to be formed through terminology extraction, semantic node decomposition, and association construction.

[0012] Step S111: Scan the text of each tender document unit in the set of tender documents to be processed sentence by sentence, identify words with tender document domain attributes, including words describing project requirements, words describing technical solutions, words describing commercial terms, and words describing implementation plans, and complete the extraction of domain terms from the set of tender documents to be processed.

[0013] The text scanning program was activated to analyze each of the four tender document units sentence by sentence. In the project requirements response documents, terms such as "steel structure bearing capacity level," "column spacing and span requirements," and "roof load standards" were identified. These terms directly correspond to the core performance requirements of steel structures for industrial plants.

[0014] In the technical implementation plan documents, identify technical terms such as "steel component welding process", "high-strength bolt connection standard", "steel structure hoisting plan" and "anti-corrosion coating process", which cover the key technical aspects of steel structure installation.

[0015] In the commercial quotation document, identify terms such as "unit price of steel components", "installation labor cost", "performance bond ratio", and "payment cycle agreement" to reflect the core commercial content of the bid.

[0016] In the construction organization plan document, identify implementation plan terms such as "construction schedule", "construction team configuration", "large equipment entry plan", and "acceptance node setting" to reflect the time and resource planning for construction execution.

[0017] All identified terms are stored in a temporary terminology buffer to ensure coverage of the core domain terms for each unit of the tender document.

[0018] Step S112: Count the frequency of the identified words with the domain of the tender document, retain the words whose frequency meets the core terminology standard of the tender document domain, delete the duplicate words, and form a terminology database for the tender document domain.

[0019] The vocabulary frequency statistics module is invoked to count the words in the temporary terminology buffer. For example, "steel component welding process" appears multiple times in the technical solution document, "high-strength bolt connection standard" is mentioned in multiple technical chapters, "construction schedule arrangement" is repeatedly emphasized in the construction plan, while "temporary material storage area" only appears once.

[0020] The core terminology standard for bidding documents is defined as "appearing in at least two bidding document units or appearing at a frequency no less than a certain benchmark in a single unit." Vocabulary is selected based on this standard. Duplicate words and low-frequency words that do not meet the standard are removed. The selected vocabulary is then categorized into four main groups: "Project Requirements," "Technical Solutions," "Commercial Terms," ​​and "Construction Organization."

[0021] For example, project requirements include "steel structure bearing capacity level" and "column spacing and span requirements", technical solutions include "steel component welding process" and "anti-corrosion coating process", commercial terms include "steel component unit price" and "payment cycle agreement", and construction organization includes "construction schedule" and "large equipment entry plan", ultimately forming a structured domain terminology library for tender documents, stored in JSON format.

[0022] Step S113: Using the terms in the domain terminology library of the tender documents as splitting markers, divide each tender document unit into multiple text segments. Each text segment contains at least one domain term and its corresponding description. Each text segment serves as a semantic node.

[0023] Using terms from the domain terminology database as anchors, the text of each tender document unit is segmented. In the technical implementation plan document, using "steel component welding process" as the segmentation marker, continuous text containing this term along with expressions such as "welding material type," "welding current control," and "weld inspection standard" is divided into one segment; using "high-strength bolt connection standard" as the segmentation marker, text containing this term along with expressions such as "bolt preload requirements," "connection surface treatment method," and "torque detection method" is divided into another segment.

[0024] In the construction organization plan document, the text containing the term "construction schedule" and expressions such as "foundation construction phase duration", "steel component installation phase progress", and "roof capping node time" are divided into segments, using "construction schedule" as the dividing mark; the text containing the term "large equipment arrival plan" and expressions such as "tower crane arrival time", "truck crane operating range", and "equipment installation and commissioning cycle" are divided into segments, using "large equipment arrival plan" as the dividing mark.

[0025] Each segmented text contains at least one domain term and its content is complete and independent; each segment is a semantic node. A unique identifier is assigned to each semantic node, following the rule of "category abbreviation-serial number". For example, semantic nodes for technical solutions are identified as "JS-001" and "JS-002", and semantic nodes for construction organizations are identified as "SG-001" and "SG-002".

[0026] Step S114: Compare the domain terms in different semantic nodes, identify semantic node combinations containing the same domain terms, and determine the term association relationship between semantic node combinations containing the same domain terms. The term association relationship includes term reference relationship, term subordination relationship and term supplement relationship.

[0027] The following sub-steps complete the association analysis: Step S1141: Traverse the text content of each semantic node, filter out words belonging to the domain terminology library of the tender document from the text content, and form a domain terminology set corresponding to each semantic node.

[0028] The text content of each semantic node is read one by one, and the terms belonging to the domain terminology database are filtered out using a term matching algorithm. For example, the text of semantic node "JS-001" is "The welding process of steel components adopts submerged arc automatic welding, the welding material is low alloy steel welding wire, the welding current is controlled within a reasonable range, and the weld needs to be ultrasonically tested," and the filtered term is "steel component welding process," forming the term set {steel component welding process}; the text of semantic node "JS-002" is "The weld quality grade of the steel component welding process needs to reach level two, and the deformation after welding needs to be controlled within the allowable range," and the filtered term is also "steel component welding process," forming the term set {steel component welding process}; the text of semantic node "JS-003" is "The standard for high-strength bolt connection requires the bolt preload to reach the design value, the connection surface needs to be sandblasted and rust-removed, and the torque value needs to be checked after installation," and the filtered term is "high-strength bolt connection standard," forming the term set {high-strength bolt connection standard}.

[0029] Step S1142: Compare the domain term sets of any two semantic nodes item by item, record the semantic node pairs that have the same domain terms, and form a semantic node combination containing the same domain terms.

[0030] Perform pairwise comparisons on the term sets of all semantic nodes: compare the term sets of “JS-001” and “JS-002”, and find that they both contain “steel component welding process”, and record the semantic node pair (JS-001, JS-002); compare “JS-004” (term set {steel component welding process, weld rework process}) with “JS-001”, and find that it also contains the same term, and record the node pair (JS-004, JS-001).

[0031] Organize all node pairs containing the same term to form a list of semantic node combinations containing the same domain term, and label each combination with the common domain term.

[0032] Step S1143: Analyze the grammatical role of the same domain term in the text content of two semantic nodes in a semantic node combination containing the same domain term. If the same domain term in one semantic node is the main description and the same domain term in the other semantic node is the referential description, then it is determined that there is a term reference relationship between the two semantic nodes.

[0033] Taking node pair (JS-005, JS-006) as an example, the text of JS-005 is "The main steel column installation adopts a dual-machine lifting scheme, and the lifting point of the main steel column is set at 1 / 3 of the column height", with the term set {main steel column installation}; the text of JS-006 is "During the hoisting process, the verticality of the column body needs to be monitored throughout, and adjustments should be made immediately if the deviation exceeds the allowable value", with the term set {main steel column installation} ("it" refers to "main steel column installation").

[0034] Analysis of grammatical roles reveals that "main steel column installation" in JS-005 is the main description, while "its" in JS-006 is the referential description. Therefore, it is determined that there is a terminological referential relationship between the two.

[0035] Step S1144: Analyze the scope of the expression content corresponding to the same domain term in the semantic node combination containing the same domain term. If the expression content of the same domain term in one semantic node is a whole concept, and the expression content of the same domain term in another semantic node is a local concept, then it is determined that there is a term subordination relationship between the two semantic nodes.

[0036] Taking node pair (JS-003, JS-007) as an example, the text of JS-003 is "The high-strength bolt connection standard requires the bolt preload to reach the design value, the connection surface needs to be sandblasted to remove rust, and the torque value needs to be checked after installation," which describes the overall requirements of the "high-strength bolt connection standard." The text of JS-007 is "The preload control in the high-strength bolt connection standard requires the use of a torque wrench, tightened to the design value in three stages," which describes the local requirements of "preload control" in the "high-strength bolt connection standard."

[0037] Because the content of the expression has a relationship between the whole and the part, it is determined that there is a term subordinate relationship between the two semantic nodes.

[0038] Step S1145: Analyze the completeness of the expression content corresponding to the same domain term in the semantic node combination containing the same domain term. If the expression content of the same domain term in two semantic nodes covers different aspects and together constitutes a complete concept, then it is determined that there is a term supplement relationship between the two semantic nodes.

[0039] Taking node pair (JS-001, JS-004) as an example, the text of JS-001 describes the welding method, materials and testing requirements of "steel component welding process"; the text of JS-004 describes the weld quality grade and deformation control requirements of "steel component welding process".

[0040] Both cover different aspects of the welding process, and when combined, they form a complete description of the welding process. Therefore, it is determined that there is a terminology complementarity relationship between the two semantic nodes.

[0041] Step S1146: Add a corresponding term association type label to each semantic node combination containing the same domain term to form a semantic node association list.

[0042] For each semantic node combination, add a label indicating "reference," "subordination," or "supplementary relationship" based on the analysis results above. For example, add a "supplementary relationship" label to the node pair (JS-001, JS-002), a "subordination" label to the node pair (JS-003, JS-007), and a "reference" label to the node pair (JS-005, JS-006).

[0043] All annotated node pairs are organized into a list of semantic node relationships, which includes three fields: node identifier pairs, common terms, and relationship type.

[0044] Step S115: Using semantic nodes as chain nodes and the terminological relationships between semantic nodes as chain connecting edges, arrange the chain nodes according to the business description order of the tender document units to form the initial tender document domain semantic chain.

[0045] Based on the business logic sequence of the tender document units (project requirements—technical solutions—commercial quotations—construction organization), each semantic node is arranged as a chain node in sequence. For example, first arrange the semantic nodes of project requirements (such as XQ-001, XQ-002), then arrange the semantic nodes of technical solutions (such as JS-001, JS-002, JS-003), then arrange the semantic nodes of commercial quotations, and finally arrange the semantic nodes of construction organization.

[0046] Based on the list of semantic node relationships, chain edges are added between nodes with existing relationships, and the relationship type is marked on the edges. For example, a link marked "supplementary relationship" is added between JS-001 and JS-002, and a link marked "subordinate relationship" is added between JS-003 and JS-007, forming the initial bid document domain semantic chain, which is stored in a graph structure data format.

[0047] Step S116: Check whether there are isolated chain nodes without chain connection edges in the initial bid document domain semantic chain. If so, analyze the potential association between the domain terms in the isolated chain node and the domain terms in other chain nodes, and add chain connection edges based on the potential association to form a complete bid document domain semantic chain.

[0048] Traverse all chain nodes in the initial semantic chain and check for any isolated nodes that are not connected to any other nodes. For example, the semantic node "SG-005" (term set {construction safety protection measures}) in the construction organization class has no connecting edges with other nodes and is therefore determined to be an isolated node.

[0049] Analyzing the potential relationship between the domain term "construction safety protection measures" of this isolated node and other node terms reveals that the description of the technical solution node "JS-008" (term set {high-altitude welding operation plan}) involves high-altitude operation safety, and there is a potential "complementary relationship" between the two (construction safety protection measures supplement the safety requirements of high-altitude welding operations).

[0050] Based on this potential relationship, a link labeled "supplementary relationship" is added between SG-005 and JS-008. This process is repeated until all nodes have at least one link, forming a complete semantic chain for the tender document domain with no isolated nodes.

[0051] Step S120: Input the domain terminology database of the tender document into the pre-trained large language model, adjust the semantic parsing weights of the pre-trained large language model to adapt the pre-trained large language model to the semantic parsing requirements of the tender document domain, and obtain the domain-adapted semantic parsing model.

[0052] A general pre-trained large language model was selected as the base model. Through domain terminology injection and weight adjustment, it was adapted to meet the semantic parsing requirements of tender documents for steel structure installation of industrial plants. The specific process is as follows.

[0053] Step S121: Select core terms from the domain terminology database of the tender documents, and match the corresponding standard expression text of the tender document domain for each core term to form a domain-adapted corpus pair composed of core terms and standard expression text.

[0054] Core terms were selected from the terminology database for tender documents. The selection criteria were "terms that cover the four major categories and are core expressions in the tender documents", such as "steel component welding process", "high-strength bolt connection standard", "construction schedule", and "steel component unit price".

[0055] Match each core term with a relevant industry standard description, which must conform to the standard wording conventions for industrial steel structure installation bidding. For example, the standard description for "steel component welding process" is: "The steel component welding process should be determined according to the component material and stress requirements. Submerged arc welding or gas shielded welding should be preferred. The welding material must match the base material. The weld quality grade should meet the design requirements. Non-destructive testing is required after welding." The standard description for "high-strength bolt connection standard" is: "High-strength bolt connections should comply with relevant industry standards. The surface treatment of the connection should meet the design roughness requirements. The bolt preload should be applied using a torque wrench according to the prescribed procedure. Torque verification is required after installation."

[0056] Each core term is combined with its matching standard expression text to form a domain-adapted corpus pair, with each corpus pair containing two fields: "core term - standard expression".

[0057] Step S122: Convert the domain-adapted corpus according to the input format requirements of the pre-trained large language model, and input them in batches into the corpus input module of the pre-trained large language model.

[0058] The input format for the pre-trained large language model requires an "instruction-content" structure. Therefore, the domain-adapted corpus pairs are converted to this structure: the instruction field is uniformly set to "parse tender document domain terms: [core terms]", and the content field is the corresponding standard expression text. For example, a certain corpus pair is converted to "Instruction: Parse tender document domain terms: steel component welding process; Content: The steel component welding process should be determined according to the component material and stress requirements...".

[0059] All converted corpus pairs are grouped according to the batch size required by the model, and then input into the corpus input module in batches through the model's corpus input interface to ensure that the input data format meets the model's parsing requirements.

[0060] Step S123: Call the parameter viewing interface of the pre-trained large language model to obtain the set of weight parameters related to term recognition and semantic understanding in the semantic parsing layer.

[0061] By using the API provided by the pre-trained large language model, the parameter viewing function is invoked to locate the cluster of neurons responsible for term recognition and semantic understanding in the model's semantic parsing layer. The weight parameters corresponding to these neurons are extracted to form a weight parameter set, which contains information such as parameter identifier, current parameter value, and corresponding semantic parsing function description.

[0062] For example, extracting the weight parameters of neurons responsible for "process term recognition" and "connection relationship semantic understanding" ensures that the acquired parameters cover the core aspects of term recognition and semantic understanding.

[0063] Step S124: Based on the core terms in the domain-adapted corpus, increase the weight parameter value for identifying domain terms in the semantic parsing layer and decrease the weight parameter value for identifying general domain terms, so that the semantic parsing layer prioritizes identifying domain terms in the bid documents.

[0064] For the core terminology types in the domain-adapted corpus, the corresponding term recognition weight parameters in the semantic parsing layer are located. For example, for technical terms such as "steel component welding process" and "high-strength bolt connection standard", the weight parameters responsible for technical term recognition are found; for business terms such as "steel component unit price" and "payment cycle agreement", the weight parameters responsible for business term recognition are found.

[0065] Increase the values ​​of these weighting parameters while decreasing the values ​​of the weighting parameters responsible for recognizing general domain terms (such as "ordinary bolts" and "general welding," which are not core bidding terms). For example, adjust the technical terminology recognition parameter value to a certain proportion higher than the general terminology recognition parameter value to ensure that the model prioritizes capturing core terms from the bidding document domain when parsing the text.

[0066] Step S125: Encapsulate the adjusted semantic parsing weight parameters into an independent domain adaptation layer module. The domain adaptation layer module includes a term weight adjustment submodule, a domain semantic correction submodule, and a representation specification submodule.

[0067] The domain adaptation layer module is built using the following sub-steps: Step S1251: Write the adjusted identification weight parameters of the domain terms in the tender document into the program of the term weight adjustment submodule, so that when the term weight adjustment submodule receives input text, it can enhance the identification strength of the domain terms in the tender document based on the identification weight parameters and output text containing domain term tags.

[0068] The terminology weight adjustment submodule is written in Python, and the adjusted weight parameters are imported into the program as a configuration file. After receiving the input text, the program segments the text using a word segmentation algorithm, and then performs terminology recognition scoring on each segment based on the imported weight parameters. Segments with scores higher than a set threshold are marked as terms in the tender document domain and annotated in the output text with special markers (such as square brackets). For example, if the input text is "Steel component welding adopts submerged arc automatic welding", the output will be "[Steel component welding] adopts submerged arc automatic welding".

[0069] Step S1252: Import the semantic rule library of the tender document domain. The semantic rule library of the tender document domain contains common semantic expression logic rules in the tender document domain, so that the semantic correction submodule of this domain can compare the semantic expression of the input text with the rules in the semantic rule library of the tender document domain and correct the expression content that does not conform to the semantic rules of the domain.

[0070] The semantic rule library for tender documents includes logical rules for the expression of tender documents for industrial steel structure installation, such as "the technical solution should first clarify the process type, then explain the parameter requirements, and finally explain the testing standards", "the commercial terms should first clarify the cost items, then explain the calculation method, and finally explain the payment method", and "the construction plan should first clarify the stage division, then explain the duration of each stage, and finally explain the node acceptance requirements", etc.

[0071] The semantic rule base is imported into the domain semantic correction submodule in XML format. The submodule program compares the expression logic of the input text with the corresponding rules in the rule base. For example, if the input text is "The weld seam needs to be ultrasonically tested, the steel component welding process adopts submerged arc automatic welding, and the welding current is controlled within a reasonable range", the comparison finds that its expression order does not conform to the rule of "first clarifying the process type, then stating the parameter requirements, and finally stating the testing standards". The submodule will automatically adjust the expression order to "[Steel component welding] adopts submerged arc automatic welding, the welding current is controlled within a reasonable range, and the weld seam needs to be ultrasonically tested", ensuring that it conforms to the semantic expression specifications of the tender documents.

[0072] Step S1253: Store the standard expression template for the tender document field. The standard expression template for the tender document field covers the standardized expression format of project requirements, technical solutions, and commercial terms, so that the expression specification sub-module can adjust the corrected semantic expression to conform to the format of the standard expression template for the tender document field.

[0073] The standard wording templates for tender documents are categorized as follows: Project requirement templates include the format "[Requirement Terminology]: Must meet [Specific Standards], corresponding [Performance Indicators] requirements"; Technical solution templates include the format "[Process Terminology]: Use [Process Methods], [Parameter Requirements], [Testing Standards]"; Commercial terms templates include the format "[Fee Terminology]: [Unit Price Standard], [Calculation Basis], [Payment Method]".

[0074] The above templates are stored in the template library of the expression specification submodule. After receiving the text output by the domain semantic correction submodule, the submodule matches the corresponding standard template according to the domain terminology categories contained in the text and adjusts the expression format. For example, the corrected text is "[Steel component welding] Submerged arc welding is used, the welding current is controlled within a reasonable range, and the weld needs to be ultrasonically tested." After matching the technical solution template, it is adjusted to "[Steel component welding process]: Submerged arc welding is used, the welding current is controlled within a reasonable range, and the weld inspection standard is ultrasonic testing," making the expression more in line with the standard format of the tender document.

[0075] Step S1254: After receiving the input text, the constraint domain adaptation layer module first calls the term weight adjustment submodule to process it, and then inputs the processing result into the domain semantic correction submodule. Finally, it inputs the correction result into the expression specification submodule.

[0076] In the main program of the domain adaptation layer module, set up flow control logic to clarify the calling order of sub-modules: when the module receives external input text, it first triggers the processing flow of the term weight adjustment sub-module to complete the domain term tagging; the tagged text is automatically passed to the domain semantic correction sub-module for expression logic correction; the corrected text is then passed to the expression standardization sub-module to complete the format adjustment.

[0077] This process is achieved through function calls within the program. For example, the main program calls the term weight adjustment submodule through the "term_weight_adjust(input_text)" function, passes the returned result as a parameter to the "semantic_correction(marked_text)" function, and then passes the semantic correction result to the "format_standardization(corrected_text)" function, ensuring that the submodules work together in sequence.

[0078] Step S1255: After integrating the program code of the terminology weight adjustment submodule, the domain semantic correction submodule, and the expression standardization submodule, input the text fragment of the tender document containing non-standard expressions into the domain adaptation layer module. Check whether the output text processed by the terminology weight adjustment submodule, the domain semantic correction submodule, and the expression standardization submodule in sequence meets the requirements of the tender document domain terminology tagging, semantic rules, and expression standardization. If it meets the requirements, the construction of the domain adaptation layer module is completed, and the test of the submodule collaboration effect is completed.

[0079] The Python code of the three sub-modules is integrated into a complete domain adaptation layer module program, generating an executable file. Text fragments from tender documents containing non-standard expressions are selected as test cases. For example, the test text is "Welding steel components uses submerged arc welding; the current must be appropriate, and the weld seam must be inspected." This text suffers from unclear terminology, disordered wording, and non-standard formatting.

[0080] The test text is input into the domain adaptation layer module: the terminology weight adjustment submodule outputs "[Welding steel components] use submerged arc welding, the current should be appropriate, and the weld should be inspected"; the domain semantic correction submodule adjusts the expression order to "[Welding steel components] use submerged arc welding, the current should be controlled within a reasonable range, and the weld needs to be inspected"; the expression specification submodule matches the template and outputs "[Steel component welding process]: use submerged arc welding, the welding current should be controlled within a reasonable range, and the weld inspection standard should meet the design requirements".

[0081] Check the output text: the terminology is accurately marked, the expression logic conforms to semantic rules, the format fits the standard template, the sub-module collaboration effect is determined to be satisfactory, and the domain adaptation layer module construction is completed.

[0082] Step S126: Embed the domain adaptation layer module between the input layer and the semantic parsing layer of the pre-trained large language model, so that the pre-trained large language model processes the input text first through the domain adaptation layer module.

[0083] By using the plugin interface of the pre-trained large language model, the constructed domain adaptation layer module is embedded into the model architecture, with the module's embedding position clearly defined as between the input layer and the semantic parsing layer. When the model receives external input text, the input layer first transmits the text to the domain adaptation layer module. After term tagging, semantic correction, and format standardization, the processed text is then transmitted to the semantic parsing layer for subsequent semantic understanding and feature extraction.

[0084] Modify the model's call flow code by adding a call instruction for the domain adaptation layer module to the text output function of the input layer. This ensures that the text flow path is "input layer — domain adaptation layer module — semantic parsing layer," achieving seamless integration of the domain adaptation function with the original model.

[0085] Step S127: Select the test bid document text input into the pre-trained large language model embedded in the domain adaptation layer module, compare the consistency of the output semantic parsing result with the standard parsing result of the bid document domain, and if the consistency meets the preset adaptation standard, then determine the domain-adapted semantic parsing model.

[0086] Historical tender documents for industrial plant steel structure installation projects were selected as test text, containing four parts: project requirements, technical solutions, commercial terms, and construction organization. The test text was input into a pre-trained large language model embedded in the domain adaptation layer module. The semantic parsing results output by the pre-trained large language model were obtained, including domain terminology recognition results, semantic relationship analysis results, and core content extraction results.

[0087] Simultaneously, experts in the bidding field were invited to manually parse the same test text, generating standard parsing results for the bidding document field. The consistency between the model's parsing results and the standard parsing results was compared, and evaluated from three dimensions: terminology recognition accuracy, semantic relationship judgment accuracy, and core content extraction completeness rate.

[0088] The preset adaptation standard is set as "the accuracy of all three dimensions is not lower than a certain percentage". If the accuracy of term recognition, semantic relationship judgment and core content extraction of the model parsing results all reach the standard, the model is judged to be adapted to meet the standard and the domain-adapted semantic parsing model is obtained. If it does not meet the standard, return to step S124 to adjust the semantic parsing weight parameters and repeat the adaptation and testing process until the preset standard is met.

[0089] Step S130: Input each semantic node in the semantic chain of the tender document domain into the domain-adapted semantic parsing model, extract the core expression content of each semantic node through the domain-adapted semantic parsing model, form a dynamic summary fragment corresponding to each semantic node, and integrate all dynamic summary fragments to obtain a dynamic summary fragment set.

[0090] Taking the complete semantic chain of the tender document as input, each semantic node is processed one by one through the semantic parsing model adapted to the domain, generating dynamic summary fragments and integrating them. The specific process is as follows.

[0091] Step S131: Read the semantic nodes corresponding to each chain node in the semantic chain of the tender document domain in sequence according to the arrangement order of the chain nodes, form a semantic node sequence, and complete the extraction of the semantic node sequence of the semantic chain of the tender document domain.

[0092] The semantic chain parser is invoked to read the graph structure data of the semantic chain in the tender document domain. Semantic nodes corresponding to each chain node are extracted according to the order of the chain nodes (project requirements - technical solutions - commercial quotations - construction organization). The identifier, text content and category of each semantic node are recorded.

[0093] The extracted semantic nodes are arranged in order to form a semantic node sequence, for example, the sequence is "XQ-001, XQ-002, JS-001, JS-002, SW-001, SW-002, SG-001, SG-002, ...". The position of each node in the sequence is consistent with the order of arrangement in the semantic chain, ensuring that the processing order conforms to the business logic of the tender document.

[0094] Step S132: Input the text content of each semantic node in the semantic node sequence into the text input port of the domain-adapted semantic parsing model in sequence, thus completing the input of semantic nodes into the domain-adapted semantic parsing model one by one.

[0095] Through the text input interface of the domain-adapted semantic parsing model, the text content of each semantic node is input into the model one by one in the order of the semantic node sequence. During the input process, a unique input identifier is added to each semantic node, which is consistent with the original identifier of the semantic node (such as "JS-001" or "SG-001"), to ensure that the model output can be associated with the corresponding semantic node.

[0096] For example, first input the text content of XQ-001: "The bearing capacity of the steel structure must meet the Class A requirements, the column spacing span is designed to be 12 meters, and the roof load standard is 0.7kN / m²", and input the identifier "XQ-001"; then input the text content of XQ-002: "The seismic resistance level of the steel structure must reach Class VIII, and the fire resistance level must meet the Class I standard", and input the identifier "XQ-002", and so on to complete the input of all semantic nodes.

[0097] Step S133: Enable the core information extraction mode in the domain-adapted semantic parsing model. This core information extraction mode is used to filter out sentences that embody the core meaning of semantic nodes from the input semantic node text content based on the domain semantic rules of the tender document.

[0098] Enable the core information extraction mode and filter core sentences through the following sub-steps: Step S1331: In the function selection interface of the semantic parsing model after domain adaptation, check the core information extraction option and set the running parameters of the core information extraction mode. The running parameters include the number of core sentences to be filtered and the proportion of core terms to be retained.

[0099] Log in to the management interface of the domain-adapted semantic parsing model, find the "Core Information Extraction" option in the function module list and check it to enter the parameter settings page. Set the core sentence filtering quantity to "1-3 core sentences per semantic node", and the specific quantity can be adjusted adaptively according to the text length of the semantic node; set the core term retention ratio to "100%", that is, ensure that all domain terms are retained in the core sentences.

[0100] Save the parameter settings, and the model will automatically enable the core information extraction mode. Subsequent input semantic node text will be processed according to this mode.

[0101] Step S1332: From the built-in rule base of the domain-adapted semantic parsing model, call the target rules related to the identification of core information in the tender document domain. The target rules include the sentence priority rule where the core term is located and the sentence priority rule guided by business logic keywords.

[0102] After the model enables the core information extraction mode, it automatically calls the core information identification rules of the bidding document domain from the built-in rule base: the sentence priority rule for core terms means that "sentences containing core terms in the domain terminology library are given priority to be determined as core sentences"; the sentence priority rule guided by business logic keywords means that "sentences containing business logic keywords such as 'must meet', 'adopt', 'requirements', 'agreements', and 'arrangements' are given priority to be determined as core sentences".

[0103] For example, the text of semantic node JS-001 contains four sentences: "[Steel component welding process] adopts submerged arc automatic welding", "Welding material is selected from low alloy steel welding wire", "Welding current is controlled within a reasonable range", and "Welder needs to be ultrasonically tested". The first and fourth sentences contain the core term "steel component welding process" and the logical keywords "adopt" and "need to be tested", and will be prioritized for screening.

[0104] Step S1333: Divide the input semantic node text content into independent sentence units according to punctuation marks, assign a unique sentence identifier to each sentence unit, compare each sentence unit with the loaded bid document domain semantic rules, and count the number of rules that each sentence unit conforms to.

[0105] After receiving semantic node text, the model divides the text into independent sentence units according to punctuation marks such as periods, question marks, and exclamation marks, and assigns a unique identifier in the format of "semantic node identifier-sentence sequence number" to each sentence unit, such as JS-001-01, JS-001-02, etc.

[0106] Each sentence unit is compared with the sentence priority rules for core terms and the sentence priority rules guided by business logic keywords, and the number of rules that each sentence unit conforms to is counted. For example, JS-001-01 "[Steel component welding process] adopts submerged arc automatic welding" conforms to 2 rules, JS-001-04 "The weld needs to be ultrasonically tested" conforms to 1 rule, JS-001-02 "The welding material is low alloy steel welding wire" conforms to 1 rule, and JS-001-03 "The welding current is controlled within a reasonable range" conforms to 0 rules.

[0107] Step S1334: Sort the number of rules met by each sentence unit and select the sentence unit with the most rules met as the candidate core sentence.

[0108] Sentence units are sorted from most to least matched by the number of rules they match, and the sentence unit with the most matched rules is selected as the candidate core sentence. For example, the sentence units of JS-001 are sorted as JS-001-01 (2 rules), JS-001-02 (1 rule), JS-001-04 (1 rule), and JS-001-03 (0 rules), and JS-001-01 is selected as the primary candidate core sentence.

[0109] Step S1335: Check whether the candidate core sentence contains the core domain terminology of the semantic node and whether it can independently express a complete semantic meaning. If it meets the requirements, it is determined to be a sentence that embodies the core meaning of the semantic node.

[0110] The sentence JS-001-01, “[Steel component welding process] adopts submerged arc automatic welding,” contains the core domain term “steel component welding process” and can independently express the complete semantic meaning of “the process method adopted for welding steel components.” Therefore, it is identified as a core sentence.

[0111] If a candidate core sentence does not contain core terms or its semantics are incomplete, the next sentence unit is selected sequentially from the sorted sentence units for examination until a core sentence that meets the requirements is found. If a single sentence cannot fully express the semantics, multiple sentences are selected and combined to form a set of core sentences to ensure the completeness of the core meaning.

[0112] Step S134: The core meaning sentences selected are simplified and expressed using the domain-adapted semantic parsing model to form a dynamic summary fragment corresponding to the semantic node.

[0113] The model simplifies the selected core sentences by removing redundant modifiers (such as adverbs and adjectives) and retaining core terms and key action descriptions; it simplifies complex sentences into simple ones to ensure concise and clear expression; and it unifies the expression style to make the summary fragments conform to the formal expression habits of tender documents.

[0114] For example, the core sentence "[Welding process for steel components] adopts submerged arc automatic welding, low alloy steel welding wire is selected as the welding material, and ultrasonic testing is required for the weld" is simplified to "The welding process for steel components adopts submerged arc automatic welding, low alloy steel welding wire is selected, and ultrasonic testing is required for the weld", forming the corresponding dynamic summary fragment of JS-001.

[0115] If there are multiple core sentences, they should be simplified and combined in a logical order to form a coherent summary segment, ensuring semantic coherence and no repetition.

[0116] Step S135: Add a corresponding semantic node identifier, a position identifier in the semantic chain of the tender document domain, and a list of domain terms to each dynamic summary fragment. Integrate all generated dynamic summary fragments according to the order of their corresponding semantic nodes in the semantic node sequence, associate the corresponding association information, and form a dynamic summary fragment set.

[0117] Add metadata information to each dynamic summary fragment: the semantic node identifier is consistent with the corresponding semantic node (e.g., "JS-001"); the position identifier is the sequence number of the semantic node in the semantic node sequence (e.g., "3", indicating the third node in the sequence); the domain term list is all the domain terms contained in the semantic node (e.g., "{steel component welding process}").

[0118] All dynamic summary fragments are arranged in the order of semantic node sequence, such as in the order of "XQ-001 summary, XQ-002 summary, JS-001 summary, JS-002 summary...", and each summary fragment is associated with its metadata information.

[0119] The integrated summary fragments and metadata information are stored as a dynamic summary fragment collection in JSON format. Each entry in the collection contains four fields: "summary text", "semantic node identifier", "location identifier" and "domain term list", which facilitates the determination of the subsequent association order and semantic connection processing.

[0120] Step S140: Based on the connection relationship of semantic nodes in the semantic chain of the tender document domain, determine the association order of each dynamic summary fragment in the dynamic summary fragment set, and perform semantic connection processing on each dynamic summary fragment according to the association order to obtain a preliminary fused summary text.

[0121] Based on the node connection relationships of the semantic chain in the tender document domain, the association logic of the summary fragments is sorted out, and a coherent preliminary fused summary is formed through connection processing. The specific process is as follows.

[0122] Step S141: Extract the link information between semantic nodes from the structural data of the semantic chain in the tender document domain, and form a connection table containing semantic node identifiers, associated node identifiers and association relationship types.

[0123] Parse the graph structure data of semantic chain in the tender document domain and extract information of all chain connection edges: each connection edge contains the starting semantic node identifier, the ending semantic node identifier, and the type of association (referential relationship, subordinate relationship, supplementary relationship).

[0124] The above information is organized into a connection relationship table, where each record corresponds to a connection edge. For example, record 1 is "starting point: JS-001, ending point: JS-002, association type: supplementary relationship"; record 2 is "starting point: JS-003, ending point: JS-007, association type: subordinate relationship"; record 3 is "starting point: JS-005, ending point: JS-006, association type: referential relationship", etc., which comprehensively reflects the association logic between semantic nodes.

[0125] Step S142: By using the semantic node identifier of the dynamic summary fragment, associate each dynamic summary fragment with the corresponding semantic node entry in the connection relationship table, and determine the associated fragment identifier and association relationship type of each dynamic summary fragment.

[0126] Traverse the dynamic summary fragment set, extract the semantic node identifier of each summary fragment, search for the record starting from the identifier in the connection relationship table, and obtain the corresponding endpoint semantic node identifier (i.e., the associated fragment identifier) ​​and the association relationship type.

[0127] For example, the semantic node identifier of the abstract fragment JS-001 is "JS-001". The endpoint identifier of the corresponding record in the connection relationship table is "JS-002", and the association type is "supplementary relationship". Therefore, the associated fragment identifier of the JS-001 abstract is determined to be "JS-002", and the association relationship type is "supplementary relationship". The associated fragment identifier of the abstract fragment JS-003 is "JS-007", and the association relationship type is "subordinate relationship".

[0128] For records in the connection table that end with a certain identifier, reverse association information is also added to the corresponding summary fragment to ensure the bidirectional nature of the association.

[0129] Step S143: Starting from the dynamic summary fragment corresponding to the initial semantic node of the domain semantic chain of the tender document, arrange the subsequent dynamic summary fragments in sequence according to the order of the associated nodes in the connection relationship table to form the dynamic summary fragment association order.

[0130] The starting semantic node of the domain semantic chain in tender documents is usually the first node of the project requirement class (such as XQ-001), with its corresponding dynamic summary fragment as the starting point of the association sequence.

[0131] Based on the connection table, find the associated node identifier starting from XQ-001 (e.g., XQ-002), and arrange the summary fragment corresponding to XQ-002 after the summary of XQ-001; then find the associated node identifier starting from XQ-002 (e.g., JS-001), and arrange the summary fragment corresponding to JS-001 after the summary of XQ-002; then find the associated node identifier starting from JS-001 (e.g., JS-002), and arrange the summary of JS-002 after the summary of JS-001, and so on.

[0132] If a semantic node has multiple associated nodes (such as JS-003 being associated with both JS-007 and SW-001), the order of arrangement is determined according to the principles of "technical logic takes precedence over business logic" and "main process takes precedence over auxiliary process". For example, the summary of the technical associated node JS-007 is arranged first, and then the summary of the business associated node SW-001 is arranged.

[0133] The above method forms a complete dynamic sequence of abstract fragment associations. The arrangement logic of each fragment in the sequence is consistent with the node connection relationship of the semantic chain in the tender document, ensuring that the association of abstract fragments conforms to the business expression logic of the tender document for industrial steel structure installation.

[0134] Step S144: Based on the association type of adjacent dynamic summary fragments, call the bid document domain connection statement template library, which contains connection statement templates corresponding to different association types, and generate connection statements connecting adjacent dynamic summary fragments based on the connection statement templates.

[0135] Generate the connector statement through the following sub-steps: Step S1441: Based on the default term association type in the semantic chain of the tender document domain, divide the connector statement templates into referential connector statement templates, subordinate connector statement templates, and supplementary connector statement templates, and store them in different directories of the tender document domain connector statement template library.

[0136] The template library for connecting statements in tender documents is categorized by relational type: The referential relationship template directory contains templates such as "For [term], its corresponding [expression] is..." and "The specific implementation requirements for the above [term] are..."; the subordinate relationship template directory contains templates such as "The [local terms] covered by [overall term] must meet..." and "Under the framework of [overall term], the implementation standard for [local terms] is..."; the supplementary relationship template directory contains templates such as "In addition to meeting the above requirements, [term] should also..." and "For [term], the supplementary explanation is as follows...".

[0137] Each template includes terminators (such as "[Terminology]" or "[Overall Terminology]") to facilitate the subsequent substitution of specific terms to generate connecting sentences.

[0138] Step S1442: Read the relationship type of adjacent dynamic summary fragments. If it is a term reference relationship, call the reference relationship connection statement template. If it is a term subordination relationship, call the subordination relationship connection statement template. If it is a term supplement relationship, call the supplement relationship connection statement template.

[0139] Extract the association type of adjacent summary fragments in the connection relationship table: for example, the association relationship between JS-005 summary and JS-006 summary is a referential relationship, so call the referential relationship connection statement template; the association relationship between JS-003 summary and JS-007 summary is a subordinate relationship, so call the subordinate relationship connection statement template; the association relationship between JS-001 summary and JS-002 summary is a complementary relationship, so call the complementary relationship connection statement template.

[0140] Step S1443: Obtain the core domain terms contained in each of the two adjacent dynamic summary segments from the association information, and determine the core domain terms common to the two dynamic summary segments.

[0141] Parse the metadata information of adjacent abstract fragments and extract the domain terminology list: for example, the domain terminology list of JS-005 abstract is {main steel column installation}, the domain terminology list of JS-006 abstract is {main steel column installation}, and the common core domain term is determined to be "main steel column installation"; the domain terminology list of JS-003 abstract is {high-strength bolt connection standard}, the domain terminology list of JS-007 abstract is {high-strength bolt connection standard}, and the common core domain term is "high-strength bolt connection standard".

[0142] Step S1444: Substitute the common core domain terms into the term placeholders in the called connector statement template, replace the general expressions in the connector statement template with specific expressions related to the core terms, and generate preliminary connector statements.

[0143] Substituting "main steel column installation" into the referential relationship connection statement template "For [term], its corresponding [expression content] is...", the preliminary connection statement "For main steel column installation, its corresponding hoisting process requires..." is generated; substituting "high-strength bolt connection standard" into the subordinate relationship connection statement template "[overall term] covers [local term] needs to meet...", combined with the local expression "preload control" in the JS-007 abstract, the preliminary connection statement "the preload control covered by the high-strength bolt connection standard needs to meet..." is generated; substituting "steel component welding process" into the supplementary relationship connection statement template "[term] in addition to meeting the above requirements, should also...", the preliminary connection statement "steel component welding process in addition to meeting the above requirements, should also..." is generated.

[0144] Step S1445: Based on the overall style of the preliminary fusion abstract text, adjust the language style of the preliminary connecting sentences to be consistent with the overall style. Check whether the connecting sentences accurately reflect the relationship between adjacent dynamic abstract segments and whether there is any semantic ambiguity. If so, modify the content of the expression until the semantics are accurate.

[0145] The overall style of the preliminary summary text is formal and concise, in line with the style of a tender document. Therefore, the initial connecting statement "For the installation of the main steel column, the corresponding hoisting process requires..." has been adjusted to "For the installation of the main steel column, the hoisting process requires..." to better conform to formal writing habits. The statement "The preload control covered by the high-strength bolt connection standard must meet..." has been checked and confirmed to accurately reflect the subordinate relationship between the whole and the parts without semantic ambiguity. If a connecting statement such as "In addition to meeting the above requirements, the welding process of steel components should also..." has a problem of ambiguous reference to "the above requirements", it has been revised to "In addition to meeting the requirements of submerged arc welding and material requirements, the welding process of steel components should also..." to clarify the reference and ensure semantic accuracy.

[0146] Step S145: Insert corresponding semantic connecting statements between two adjacent dynamic summary segments according to the association order to form a continuous text sequence. Analyze the length distribution of sentences in the text sequence, break down excessively long sentences into short sentences that conform to reading habits, adjust the word order between sentences, format the adjusted text sequence, remove connecting statements with repeated expressions, and obtain the preliminary fused summary text.

[0147] According to the association order of dynamic summary fragments, each summary fragment is concatenated with its corresponding connecting statement in sequence: for example, XQ-001 summary + connecting statement 1 + XQ-002 summary + connecting statement 2 + JS-001 summary + connecting statement 3 + JS-002 summary... to form a continuous text sequence.

[0148] Analyze the sentence length in the text sequence. If a sentence contains multiple clauses and its total length exceeds a preset threshold (e.g., more than 30 characters), split it into shorter sentences. For example, "The steel component welding process adopts submerged arc automatic welding, low alloy steel welding wire is selected, the weld needs to be ultrasonically tested and the quality level needs to reach level two" is split into "The steel component welding process adopts submerged arc automatic welding, low alloy steel welding wire is selected. The weld needs to be ultrasonically tested and the quality level needs to reach level two".

[0149] Adjust the sentence order to make the expression more logical. For example, change "The weld seam needs to be ultrasonically tested, and the steel component welding process adopts submerged arc automatic welding" to "The steel component welding process adopts submerged arc automatic welding, and the weld seam needs to be ultrasonically tested."

[0150] The text sequence is formatted: punctuation is standardized, and duplicate connecting phrases are removed (e.g., if two consecutive "in addition" phrases appear, only one is retained and its wording is adjusted) to ensure consistent text formatting, concise and coherent expression, ultimately forming a preliminary integrated abstract text.

[0151] Step S150: Receive semantic feedback information from the tender document reader regarding the preliminary fused summary text, adjust the content and order of each dynamic summary fragment based on the semantic feedback information, update the semantic chain of the tender document domain, generate the final tender document summary, and output it to the target receiving end.

[0152] For example, the summary text can be optimized based on reader feedback to ensure that the summary meets actual usage needs. The specific process is as follows.

[0153] Step S151: On the display interface of the preliminary fused abstract text, add a feedback input control. The feedback input control supports the reader of the tender document to input semantic feedback information in text form. The semantic feedback information includes semantic missing feedback, redundant expression feedback, and disordered order feedback.

[0154] In the initial integrated summary text display platform (such as web page or client), add feedback input controls to the text area corresponding to each dynamic summary fragment, including three radio buttons for "semantic missing", "redundant expression", and "disordered sequence" and a text input box.

[0155] If a reader finds that a summary passage is missing core information, they can check "Semantic Missing" and enter the missing content in the text box (e.g., "The welding process for steel components does not mention welding current requirements"). If they find that the description is repetitive or redundant, they can check "Redundant Description" and enter the redundant content (e.g., "'Weld inspection' is mentioned twice"). If they find that the order of the passages is illogical, they can check "Disordered Order" and enter adjustment suggestions (e.g., "The business quotation passage should follow the technical solution passage").

[0156] Step S152: Obtain semantic feedback information input by the reader through the feedback receiving interface, extract keywords from the feedback information, and determine the dynamic summary fragment identifier and feedback type corresponding to the feedback information.

[0157] The feedback receiving interface receives feedback information submitted by readers in real time and stores the feedback information in the feedback database. A keyword extraction algorithm is used to process the feedback text: for example, in the feedback information "The welding process of steel components in the JS-001 abstract does not mention welding current requirements," the keywords "JS-001," "steel component welding process," and "welding current requirements" are extracted, and the corresponding dynamic abstract segment is identified as "JS-001," with the feedback type being "semantic missing." Similarly, in the feedback information "XQ-002 and JS-001 are in reverse order," the keywords "XQ-002," "JS-001," and "inverted order" are extracted, and the corresponding segments are identified as "XQ-002" and "JS-001," with the feedback type being "order disorder."

[0158] Step S153: For semantic missing feedback, supplement the missing core terms and expressions in the corresponding dynamic summary fragment; for expression redundancy feedback, delete the redundant modifiers in the corresponding dynamic summary fragment.

[0159] For the semantic missing feedback "no mention of welding current requirements" identified as JS-001, the text content of the original semantic node JS-001 was retrieved: "The welding process for steel components adopts submerged arc automatic welding, the welding material is low alloy steel welding wire, and the weld needs to be ultrasonically inspected." The phrase "the welding current is controlled within a reasonable range" was added to the corresponding dynamic summary fragment. The supplemented summary fragment is: "The welding process for steel components adopts submerged arc automatic welding, the welding material is low alloy steel welding wire, the welding current is controlled within a reasonable range, and the weld needs to be ultrasonically inspected."

[0160] Regarding the redundant feedback that "the term 'performance guarantee' is mentioned repeatedly in the SW-001 summary," the SW-001 summary was retrieved, which stated "the performance guarantee ratio is a certain percentage, and the performance guarantee must be submitted after signing the contract." The duplicate phrase "performance guarantee" was deleted, and the summary was revised to "the performance guarantee ratio is a certain percentage, and it must be submitted after signing the contract," thus removing the redundant elements.

[0161] Step S154: In response to the disordered order feedback, refer to the connection relationship table of the semantic chain in the tender document domain, rearrange the position of the corresponding dynamic summary fragment in the text sequence, and update the semantic connection statements of adjacent dynamic summary fragments.

[0162] Regarding the feedback that "XQ-002 and JS-001 are in reverse order", referring to the connection relationship table, the associated node of XQ-002 is JS-001, and the business logic should be "project requirements - technical solution". Therefore, the JS-001 summary should be adjusted to follow the XQ-002 summary.

[0163] Regenerate the connecting statements between adjacent segments: The original connecting statement between XQ-002 and JS-001 was "Based on the above project requirements, the corresponding technical solution is...". After adjusting the order, this connecting statement will still be used to ensure that the connecting logic matches the new arrangement order. If the adjustment involves multiple segments (such as the disordered order of JS-002, SW-001, and SG-001), then rearrange them according to the business logic order (technical solution - commercial quotation - construction organization) in the connection relationship table, and update the corresponding connecting statements. For example, adjust the connecting statement between JS-002 and SW-001 to "After the technical solution is determined, the corresponding commercial quotation is as follows...".

[0164] Step S155: Based on the adjusted dynamic summary fragment association order and description content, correct the position and text content of the corresponding semantic nodes in the semantic chain of the tender document domain, and supplement or delete the corresponding link connections.

[0165] For the JS-001 summary of the supplemented content in step S153, find its corresponding semantic node JS-001, add "welding current controlled within a reasonable range" to the text content of the node, and update the domain terminology list of the semantic node (if the supplemented content involves new terms, they need to be added to the terminology list at the same time).

[0166] For the XQ-002 and JS-001 nodes whose order was adjusted in step S154, the arrangement of the two nodes in the semantic chain of the tender document domain is corrected to ensure that the order of the chain nodes is consistent with the order of the adjusted summary fragments. If the adjustment causes the original chain connection edge to become invalid (such as the change of the associated node of a certain node), the invalid connection edge is deleted and a new chain connection edge is added according to the new association relationship. For example, if the associated node of JS-001 changes from JS-003 to JS-002, the connection edge between JS-001 and JS-003 is deleted, and the connection edge between JS-001 and JS-002 is added and the association relationship type is marked.

[0167] Step S156: Reassemble the adjusted dynamic summary fragments and semantic connection statements to form complete text content, check the semantic coherence and completeness of the text content, and obtain the final bid document summary.

[0168] Following the adjusted dynamic summary fragment association order, reassemble each fragment with the updated connecting statements to form complete text content. Check the semantic coherence of the text: Read through the entire text to confirm that the connections between each fragment are natural and without logical breaks (e.g., whether the connection between the technical solution fragment and the business quotation fragment conforms to the "technology-business" logic); check the completeness of expression: confirm that all content mentioned in the feedback regarding semantic deficiencies has been supplemented, all redundant content has been deleted, and no key information has been omitted or repeated.

[0169] For example, the inspection found that the preload requirement of the "high-strength bolt connection standard" has been supplemented, the redundant description of "weld inspection" has been deleted, the order of each segment conforms to the business logic, the text is judged to meet the requirements, and the final bid document summary is obtained.

[0170] Step S157: According to the format requirements of the target receiver, convert the final bid document summary into the corresponding file format, send it to the target receiver through the data transmission interface, and store the final bid document summary and the corresponding updated bid document domain semantic chain in the database.

[0171] If the target recipient is an internal project management system of the enterprise, and the format requirement is PDF, then use a format conversion tool to convert the final bid document summary to PDF format, ensuring that the format conforms to the system's document upload specifications (e.g., page size A4, font SimSun). If the target recipient is an email recipient, and the format requirement is Word format, then convert it to Word format and set a uniform header and footer (including project name and document type).

[0172] The converted file is sent to the target recipient via a data transfer interface (such as an HTTP interface or an FTP interface): if it is a project management system, the PDF file is uploaded through the system's file upload interface; if it is an email recipient, the Word file is sent as an attachment through the email sending interface.

[0173] Simultaneously, the final bid document summary (text format and converted format file) and the updated bid document domain semantic chain (graph structure data) are stored in a designated directory of the database. The directory is named according to "Project Name-Generation Time" (e.g., "Industrial Plant Steel Structure Installation-XXXX") for easy subsequent retrieval and reuse. The database automatically records metadata information such as storage time, file size, and associated bid document sets to ensure data traceability.

[0174] Figure 2 The illustration shows exemplary hardware and software components of a bid summary generation system 100 based on a large language model, which can implement the ideas of this application, according to some embodiments of this application. For example, a processor 120 can be used in the bid summary generation system 100 based on a large language model and to perform the functions in this application.

[0175] The bid summary generation system 100 based on a large language model can be a general-purpose server or a special-purpose server; both can be used to implement the bid summary generation method based on a large language model of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the load.

[0176] For example, a bid summary generation system 100 based on a large language model may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and various forms of storage media 140, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the bid summary generation system 100 based on a large language model may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The methods of this application can be implemented according to these program instructions. The bid summary generation system 100 based on a large language model also includes an I / O interface 150 between the computer and other input / output devices.

[0177] For ease of explanation, only one processor is described in the large language model-based bid summary generation system 100. However, it should be noted that the large language model-based bid summary generation system 100 of this application may also include multiple processors, and therefore the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the large language model-based bid summary generation system 100 performs steps A and B, it should be understood that steps A and B may also be performed jointly by two different processors or individually by one processor. For example, the first processor performs step A, the second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0178] Furthermore, this embodiment of the invention also provides a readable storage medium, wherein computer-executable instructions are preset in the readable storage medium, and when the processor executes the computer-executable instructions, the above-described method for generating bid document summaries based on a large language model is implemented.

[0179] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A method for generating bid document summaries based on a large language model, characterized in that, The method includes: Domain terms are extracted from the input set of tender documents to be processed to form a domain terminology library for tender documents. Based on the terminology associations in each tender document unit, each tender document unit is split into semantic nodes. The semantic nodes are connected through terminology associations to form a domain semantic chain for tender documents. The domain terminology database of the tender document is input into the pre-trained large language model, and the semantic parsing weights of the pre-trained large language model are adjusted to adapt the pre-trained large language model to the semantic parsing requirements of the tender document domain, thus obtaining a domain-adapted semantic parsing model. Each semantic node in the semantic chain of the tender document domain is input into the domain-adapted semantic parsing model. The core expression content of each semantic node is extracted through the domain-adapted semantic parsing model to form a dynamic summary fragment corresponding to each semantic node. All dynamic summary fragments are integrated to obtain a dynamic summary fragment set. Based on the connection relationship of semantic nodes in the semantic chain of the tender document domain, the association order of each dynamic summary fragment in the dynamic summary fragment set is determined, and the semantic connection processing of each dynamic summary fragment is performed according to the association order to obtain the preliminary fused summary text; The system receives semantic feedback from the tender document reader regarding the preliminary fused summary text, adjusts the content and order of each dynamic summary segment based on the semantic feedback, updates the semantic chain of the tender document domain, generates the final tender document summary, and outputs it to the target receiving end.

2. The method for generating bid document summaries based on a large language model according to claim 1, characterized in that, The process involves extracting domain terms from the input set of tender documents to form a domain terminology library for tender documents. Based on the terminology relationships within each tender document unit, each tender document unit is divided into semantic nodes. These semantic nodes are then connected through terminology relationships to form a domain semantic chain for tender documents. This includes: The text of each unit of the tender documents in the set of tender documents to be processed is scanned sentence by sentence to identify words with the domain attributes of tender documents. The words with the domain attributes of tender documents include words describing project requirements, words describing technical solutions, words describing commercial terms, and words describing implementation plans, thereby completing the extraction of domain terms from the set of tender documents to be processed. Frequency statistics are performed on the identified words with the domain of bidding documents. Words whose frequency of occurrence meets the core terminology standard of bidding document domain are retained, and duplicate words are deleted to form a terminology database of bidding document domain. Using terms from the domain terminology library of the tender documents as splitting markers, each tender document unit is divided into multiple text segments. Each text segment contains at least one domain term and its corresponding description. Each text segment serves as a semantic node. By comparing domain terms in different semantic nodes, semantic node combinations containing the same domain terms are identified, and term association relationships between semantic node combinations containing the same domain terms are determined. The term association relationships include term reference relationships, term subordination relationships, and term supplementation relationships. Using semantic nodes as chain nodes and the terminological relationships between semantic nodes as chain connecting edges, the chain nodes are arranged in the order of business description of the tender document units to form the initial tender document domain semantic chain. Check if there are isolated chain nodes without connecting edges in the domain semantic chain of the initial bid document. If so, analyze the potential associations between the domain terms in the isolated chain node and the domain terms in other chain nodes, and add connecting edges based on the potential associations to form a complete domain semantic chain of the bid document.

3. The method for generating bid document summaries based on a large language model according to claim 2, characterized in that, The step of comparing domain terms in different semantic nodes, identifying combinations of semantic nodes containing the same domain terms, and determining the terminology associations between combinations of semantic nodes containing the same domain terms includes: Traverse the text content of each semantic node, filter out words belonging to the domain terminology library of the tender document from the text content, and form a domain terminology set corresponding to each semantic node; Compare the domain term sets of any two semantic nodes item by item, record the semantic node pairs that have the same domain terms, and form a semantic node combination containing the same domain terms. In analyzing semantic node combinations containing terms from the same domain, the grammatical roles of the terms in the text content of the two semantic nodes are determined. If the term in one semantic node is the main description and the term in the other semantic node is the referential description, then a term reference relationship is determined between the two semantic nodes. Analyze the scope of the content to be expressed by the same domain terms in the combination of semantic nodes. If the content to be expressed by the same domain terms in one semantic node is a whole concept and the content to be expressed by the same domain terms in another semantic node is a local concept, then it is determined that there is a term dependency relationship between the two semantic nodes. Analyze the completeness of the expression content corresponding to the same domain term in the semantic node combination. If the expression content of the same domain term in two semantic nodes covers different aspects and together constitutes a complete concept, then it is determined that there is a term supplement relationship between the two semantic nodes. Add a corresponding term association type label to each semantic node combination containing terms from the same domain to form a list of semantic node associations.

4. The method for generating bid document summaries based on a large language model according to claim 1, characterized in that, The terminology database of the bidding document domain is input into a pre-trained large language model. The semantic parsing weights of the pre-trained large language model are adjusted to adapt the pre-trained large language model to the semantic parsing requirements of the bidding document domain, resulting in a domain-adapted semantic parsing model, including: Core terms are selected from the domain terminology database of the tender documents, and corresponding standard expression texts of the tender document domain are matched for each core term to form a domain-adapted corpus pair consisting of core terms and standard expression texts; The domain-adapted corpus pairs are converted according to the input format requirements of the pre-trained large language model and then input in batches into the corpus input module of the pre-trained large language model. Call the parameter viewing interface of the pre-trained large language model to obtain the set of weight parameters related to term recognition and semantic understanding in the semantic parsing layer; Based on the core terms in the domain-adapted corpus, the weight parameter value for identifying domain terms in the bidding documents is increased in the semantic parsing layer, while the weight parameter value for identifying general domain terms is decreased, so that the semantic parsing layer prioritizes the identification of domain terms in the bidding documents. The adjusted semantic parsing weight parameters are encapsulated into an independent domain adaptation layer module, which includes a term weight adjustment submodule, a domain semantic correction submodule, and a representation standardization submodule. The domain adaptation layer module is embedded between the input layer and the semantic parsing layer of the pre-trained large language model, so that the pre-trained large language model processes the input text first through the domain adaptation layer module. The test tender document text is input into a pre-trained large language model embedded in the domain adaptation layer module. The consistency of the output semantic parsing result is compared with the standard parsing result of the tender document domain. If the consistency meets the preset adaptation standard, the domain-adapted semantic parsing model is determined.

5. The method for generating bid document summaries based on a large language model according to claim 4, characterized in that, The step of encapsulating the adjusted semantic parsing weight parameters into an independent domain adaptation layer module includes: The program of the terminology weight adjustment submodule is written into the identification weight parameters of the domain terms in the tender document. When the terminology weight adjustment submodule receives input text, it can enhance the identification strength of the domain terms in the tender document based on the identification weight parameters and output text containing domain term tags. Import the semantic rule library of the bidding document domain. The semantic rule library of the bidding document domain contains common semantic expression logic rules in the bidding document domain. This enables the semantic correction submodule of the domain to compare the semantic expression of the input text with the rules in the semantic rule library of the bidding document domain and correct the expression content that does not conform to the semantic rules of the domain. The system stores standard expression templates for the bidding document field. These templates cover the standardized expression formats for project requirements, technical solutions, and commercial terms. This allows the expression specification submodule to adjust the corrected semantic expressions to conform to the format of the standard expression templates for the bidding document field. After receiving the input text, the constraint domain adaptation layer module first calls the term weight adjustment submodule to process it, and then inputs the processing result into the domain semantic correction submodule. Finally, it inputs the correction result into the expression standardization submodule. After integrating the program code of the terminology weight adjustment submodule, the domain semantic correction submodule, and the expression standardization submodule, input the tender document text fragment containing non-standard expressions into the domain adaptation layer module. Check whether the output text processed by the terminology weight adjustment submodule, the domain semantic correction submodule, and the expression standardization submodule in sequence meets the requirements of tender document domain terminology marking, semantic rules, and expression standardization. If they meet the requirements, the construction of the domain adaptation layer module is completed, and the test of the submodule collaboration effect is completed.

6. The method for generating bid document summaries based on a large language model according to claim 1, characterized in that, Each semantic node in the semantic chain of the tender document domain is input into the domain-adapted semantic parsing model. The core expression content of each semantic node is extracted through the domain-adapted semantic parsing model to form a dynamic summary fragment corresponding to each semantic node. All dynamic summary fragments are integrated to obtain a dynamic summary fragment set, including: According to the arrangement order of the chain nodes in the semantic chain of the tender document domain, the semantic nodes corresponding to each chain node are read in sequence to form a semantic node sequence, thus completing the extraction of the semantic node sequence of the semantic chain of the tender document domain. The text content of each semantic node in the semantic node sequence is sequentially input into the text input port of the domain-adapted semantic parsing model, thus completing the sequential input of semantic nodes into the domain-adapted semantic parsing model. In the domain-adapted semantic parsing model, a core information extraction mode is enabled. This core information extraction mode is used to filter out sentences that embody the core meaning of semantic nodes from the input semantic node text content based on the domain semantic rules of the tender document. The core meaning sentences selected by the domain-adapted semantic parsing model are simplified and expressed to form dynamic summary fragments corresponding to the semantic nodes. Add a corresponding semantic node identifier, its position identifier in the semantic chain of the tender document, and a list of domain terms to each dynamic summary fragment; All generated dynamic summary fragments are arranged according to the order of their corresponding semantic nodes in the semantic node sequence, and associated with the corresponding related information to form a dynamic summary fragment set.

7. The method for generating bid document summaries based on a large language model according to claim 6, characterized in that, The aforementioned core information extraction mode is enabled in the domain-adapted semantic parsing model. This core information extraction mode is used to filter sentences that embody the core meaning of semantic nodes from the input semantic node text content based on the domain semantic rules of the tender document, including: In the function selection interface of the domain-adapted semantic parsing model, check the core information extraction option and set the running parameters of the core information extraction mode. The running parameters include the number of core sentences to be filtered and the proportion of core terms to be retained. From the built-in rule base of the domain-adapted semantic parsing model, target rules related to the identification of core information in the tender document domain are called. These target rules include sentence priority rules for core terms and sentence priority rules for business logic keywords. The input semantic node text content is divided into independent sentence units according to punctuation marks. A unique sentence identifier is assigned to each sentence unit. Each sentence unit is compared with the loaded bid document domain semantic rules, and the number of rules that each sentence unit conforms to is counted. The number of rules met by each sentence unit is sorted, and the sentence unit with the most rules met is selected as the candidate core sentence. Check whether the candidate core sentence contains the core domain terminology of the semantic node and whether it can independently express a complete semantic meaning. If it meets the requirements, it is determined to be a sentence that embodies the core meaning of the semantic node.

8. The method for generating bid document summaries based on a large language model according to claim 1, characterized in that, The process involves determining the association order of dynamic summary fragments in the dynamic summary fragment set based on the connection relationships of semantic nodes in the semantic chain of the tender document domain, and performing semantic connection processing on each dynamic summary fragment according to the association order to obtain a preliminary fused summary text, including: From the structural data of the semantic chain in the tender document domain, extract the link connection information between each semantic node to form a connection table containing semantic node identifiers, associated node identifiers, and association relationship types; By using the semantic node identifier of the dynamic summary fragment, each dynamic summary fragment is associated with the corresponding semantic node entry in the connection relationship table, thereby determining the associated fragment identifier and association relationship type of each dynamic summary fragment; Starting with the dynamic summary fragment corresponding to the initial semantic node of the domain semantic chain of the tender document, the subsequent dynamic summary fragments are arranged in sequence according to the order of the associated nodes in the connection relationship table to form the dynamic summary fragment association order; Based on the relationship type between adjacent dynamic summary fragments, the bid document domain connection statement template library is invoked. The bid document domain connection statement template library contains connection statement templates corresponding to different relationship types. Connection statements connecting adjacent dynamic summary fragments are generated based on the connection statement templates. Following the order of association, corresponding semantic connecting statements are inserted between two adjacent dynamic summary segments to form a continuous text sequence. The length distribution of sentences in the text sequence is analyzed, and excessively long sentences are broken down into short sentences that conform to reading habits. The word order between sentences is adjusted, and the adjusted text sequence is formatted and repetitive connecting statements are removed to obtain a preliminary fused summary text.

9. The method for generating bid document summaries based on a large language model according to claim 8, characterized in that, The step involves calling a domain-specific connector template library for tender documents based on the association type of adjacent dynamic summary fragments. This library contains connector templates corresponding to different association types. Connecting statements linking adjacent dynamic summary fragments are generated based on these templates, including: Based on the default term association type in the semantic chain of the tender document domain, the connector statement templates are divided into referential connector statement templates, subordinate connector statement templates, and supplementary connector statement templates, which are stored in different directories of the tender document domain connector statement template library. Read the relationship type of adjacent dynamic summary fragments. If it is a term reference relationship, call the reference relationship connection statement template. If it is a term subordination relationship, call the subordination relationship connection statement template. If it is a term supplement relationship, call the supplement relationship connection statement template. From the association information of two adjacent dynamic summary segments, obtain the core domain terms contained in each segment and determine the core domain terms common to the two dynamic summary segments; Substitute common core domain terms into the terminators in the called connector template, replace the general expressions in the connector template with specific expressions related to the core terms, and generate preliminary connector statements. Based on the overall style of the preliminary integrated abstract text, the language style of the preliminary connecting sentences is adjusted to be consistent with the overall style. The connecting sentences are checked to see if they accurately reflect the relationship between adjacent dynamic abstract segments and whether there is any semantic ambiguity. If so, the content is modified until the semantics are accurate.

10. A tender document summary generation system based on a large language model, characterized in that, The method includes a processor and a memory, the memory being connected to the processor. The memory is used to store programs, instructions, or code, and the processor is used to execute the programs, instructions, or code in the memory to implement the bid document summary generation method based on a large language model as described in any one of claims 1-9.

Images