Electronic bidding document generation method and system based on artificial intelligence

Abstract

The invention provides an electronic bidding document generation method and system based on artificial intelligence, and the method comprises the steps: firstly obtaining a bidding demand text issued by a bidding party, a historical bidding document material set accumulated by a bidding party, and bidding demand change records of similar projects in an industry; and constructing a bid inviting scene semantic map containing demand nodes, multi-modal material nodes, trend association nodes and inter-node dynamic semantic association strength information. Generating an electronic bidding document dynamic framework according to the bidding scene semantic map, and calling a cross-modal artificial intelligence content generation model to generate a multi-modal module content first draft of each function module unit; and then semantic consistency adjustment and connection optimization processing are carried out on the first draft of the multi-modal module content to obtain the integrated multi-modal electronic bidding document content. And finally, in combination with the presentation specification information and the industry bidding preference data, carrying out structural typesetting optimization and content adaptation adjustment processing on the integrated content to obtain a final multi-modal electronic bidding document, so that the quality and pertinence of the electronic bidding document are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an electronic tender generation method and system based on artificial intelligence. BACKGROUND

[0002] In today's business competitive environment, bidding activities are increasingly frequent and complex, and the quality and pertinence of electronic tenders, as key documents in the bidding process, directly affect the winning probability of the bidder. The traditional method of generating electronic tenders mainly relies on manual writing and editing, and the bidder needs to invest a lot of manpower, material resources and time to collect tender demand information, organize historical tender materials, and prepare tenders according to experience and professional knowledge.

[0003] However, the above method has many limitations. On the one hand, manual processing is difficult to fully and accurately grasp the details and potential requirements of the tender demand, especially for some complex and multi-scenario tender projects, which is prone to inaccurate understanding of the demand, omission of key information, and other problems, resulting in mismatch between the tender content and the tender demand. On the other hand, the utilization efficiency of historical tender materials is low, and it is difficult for manual processing to quickly filter out the relevant content from a large amount of historical materials and effectively integrate and innovate, so that the tender content lacks pertinence and innovation. In addition, with the continuous development and change of the industry, the tender demand also shows a trend of dynamic change, and the traditional method cannot timely capture these changes and reflect them in the tender, resulting in lag of the tender content and difficulty in meeting the market demand. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the present application embodiment provides an electronic tender generation method based on artificial intelligence, which comprises: obtaining the tender demand text published by the tenderer, the historical tender material set accumulated by the bidder, and the tender demand change record of the same type of project in the industry, constructing a dynamically updated tender scene semantic graph based on the scene description information of the tender demand text, the multi-modal material information of the historical tender material set, and the trend characteristics of the tender demand change record, wherein the tender scene semantic graph comprises demand nodes, multi-modal material nodes, trend association nodes, and dynamic semantic association strength information between the nodes; According to the hierarchical relationship of the demand nodes and the evolution characteristics of the trend association nodes in the tender scene semantic graph, the overall structure of the electronic tender is adaptively disassembled, and an electronic tender dynamic framework comprising a plurality of functional module units and dynamic connection rules between the functional module units is generated, each functional module unit corresponding to a group of associated demand nodes, associated multi-modal material nodes and associated trend association nodes in the tender scene semantic graph; The pre-trained cross-modal AI content generation model is invoked, and the dynamic semantic association strength information of the electronic tender document dynamic framework, the semantic graph of the bidding scenario, and the multimodal material parsing rules are input. Cross-modal material association generation processing is performed on each functional module unit to obtain the first draft of the multimodal module content corresponding to each functional module unit. Based on the dynamic semantic association strength information of the semantic graph of the bidding scenario and the dynamic connection rules between functional module units, the initial draft of the multimodal module content of each functional module unit is subjected to semantic consistency adjustment and connection optimization processing between functional module units, resulting in the integrated multimodal electronic tender document content. Obtain presentation standard information and industry bidding preference data for electronic tender application scenarios. Combine the presentation standard information and industry bidding preference data to perform structural layout optimization and content adaptation adjustment on the integrated multimodal electronic tender content to obtain the final multimodal electronic tender.

[0005] Furthermore, embodiments of the present invention also provide an artificial intelligence-based electronic tender document generation system, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to perform the above-described AI-based electronic tender generation method by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, a processor of a computer device reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the computer device to execute the above-described artificial intelligence-based electronic tender generation method.

[0007] Based on the above aspects, by constructing a dynamically updated bidding scene semantic graph, multi-source information such as bidding demand text, historical bid document material set, and bidding demand change records of similar projects in the industry is integrated, and the complex characteristics and dynamic change trend of the bidding scene are described with demand nodes, multi-modal material nodes, trend association nodes, and dynamic semantic association strength information between nodes. According to the bidding scene semantic graph, a dynamic framework of the electronic bid document is generated, the overall structure of the electronic bid document is adaptively disassembled, the bid document structure can closely fit the changes of bidding demand, and the flexibility and adaptability of the bid document are improved. The multi-modal module content draft is generated by calling the cross-modal artificial intelligence content generation model, which fully utilizes the powerful generation capability of artificial intelligence, realizes efficient association and generation of cross-modal materials, and greatly improves the generation efficiency and quality of bid document content. The multi-modal module content draft is adjusted for semantic consistency and optimized for connection between functional module units. Finally, combined with the presentation specification information of the electronic bid document application scene and the industry bidding preference data, the integrated multi-modal electronic bid document content is optimized for structure and layout and adjusted for content adaptation, so that the generated electronic bid document not only meets the industry standards, but also better meets the preferences of the bidding party, significantly improving the quality and pertinence of the electronic bid document. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 is the execution flow diagram of the electronic bid document generation method based on artificial intelligence provided by an embodiment of the present application.

[0009] Figure 2 is a schematic diagram of exemplary hardware and software components of the electronic bid document generation system based on artificial intelligence provided by an embodiment of the present application. DETAILED DESCRIPTION

[0010] The present application will be described in detail below with reference to the accompanying drawings, Figure 1 is a flow diagram of the electronic bid document generation method based on artificial intelligence provided by an embodiment of the present application, and the electronic bid document generation method based on artificial intelligence will be described in detail below.

[0011] Step S110: Obtain the bidding demand text published by the bidding party, the historical bid document material set accumulated by the bidding party, and the bidding demand change records of similar projects in the industry. Based on the scene description information of the bidding demand text, the multi-modal material information of the historical bid document material set, and the trend characteristics of the bidding demand change records, a dynamically updated bidding scene semantic graph is constructed, which includes demand nodes, multi-modal material nodes, trend association nodes, and dynamic semantic association strength information between nodes.

[0012] For example, in the context of a construction project bidding scenario, the bidding party will issue a detailed bidding requirement text, which covers various aspects of the project, such as building size, functional requirements, quality standards, and construction period requirements. Bidders have accumulated a rich collection of historical bid materials over time, which have multi-modal characteristics, including text-based construction plans, technical specifications, chart-based project schedule plans, cost budgets, and formula-based cost calculations, mechanical analysis, etc. At the same time, there are also records of changes in bidding requirements for similar projects in the industry, which reflect the changing trends in environmental protection requirements, intelligentization level, and application of new technologies in the construction industry over time.

[0013] Based on the above-mentioned scenario description information of the bidding requirement text, the multi-modal material information of the historical bid material collection, and the trend characteristics of the bidding requirement change records, the bidding scenario semantic graph is constructed. The requirement nodes in the bidding scenario semantic graph correspond to the specific requirements in the bidding requirement text, the multi-modal material nodes represent various materials in the historical bid material collection, and the trend association nodes reflect the time-varying trends of the bidding requirements. The dynamic semantic association strength information between nodes represents the closeness of their semantic connection, and will be dynamically adjusted as the bidding requirements change.

[0014] Step S111: Extract the scenario description information in the bidding requirement text, perform semantic segmentation processing on the scenario description information, divide the continuous text into multiple semantically independent requirement expression units, extract the core requirement vocabulary and requirement constraints from each requirement expression unit, and generate a requirement basis information list.

[0015] For the bidding requirement text of a construction project, it contains various detailed descriptions about the project, such as the geographical location of the building, the surrounding environment, the design style, the functional layout, and other scenario information. Through semantic segmentation processing, the above continuous text content is divided into individual semantically independent requirement expression units. For example, the description "the building should use environmentally friendly and energy-saving materials, and meet the relevant national standards" is divided into an independent requirement expression unit.

[0016] Extract the core requirement vocabulary from each requirement expression unit, which is "environmentally friendly and energy-saving materials" in the above example, and extract the requirement constraints, i.e. "should use" and "meet the relevant national standards". Organize the core requirement vocabulary and requirement constraints in all requirement expression units to generate a requirement basis information list.

[0017] Step S112: Perform multi-modal classification processing on the historical bid material set to divide the historical bid materials into three categories: text materials, chart materials, and formula materials. Extract the core theme information of each type of historical bid material. The core theme information of text materials is theme vocabulary and key paragraphs. The core theme information of chart materials is chart title, coordinate axis meaning, and data trend description. The core theme information of formula materials is formula purpose and parameter definition.

[0018] In the historical bid material set of construction engineering, there are various types of materials. The above materials are classified by multi-modal classification processing into text materials, chart materials, and formula materials.

[0019] For text materials, they may include construction organization design, technical scheme, etc. Extract theme vocabulary such as "construction management" and "technical innovation", and key paragraphs, which usually contain important technical measures and management methods. Chart materials may include engineering progress chart, cost budget table, etc. Extract chart title such as "engineering progress plan chart", coordinate axis meaning such as time on the horizontal axis and engineering progress on the vertical axis, and data trend description such as the trend of engineering progress over time. Formula materials may involve engineering cost calculation and structural mechanics analysis formulas. Extract formula purpose such as for engineering cost calculation, and parameter definition, i.e. the specific meaning of each parameter in the formula.

[0020] Step S113: Each type of historical bid material is treated as a multi-modal material node, and each multi-modal material node contains material type identification, core theme information, and material storage path.

[0021] After the classification and core theme information extraction of historical bid materials, each type of material is converted into a multi-modal material node in the graph. Each multi-modal material node has a clear material type identification, such as "text material", "chart material", and "formula material". At the same time, it contains the previously extracted core theme information to describe the main content and purpose of the material. In addition, the material storage path is also recorded, which points to the specific location of the material in the storage system.

[0022] Step S114: Analyze the bidding requirement change records of similar projects in the industry, extract the time series information and demand change content in the bidding requirement change records, perform trend feature recognition on the demand change content, determine the demand focus change direction and change frequency of similar projects at different time stages, and take the demand focus change direction at each time stage as a trend association node. Each trend association node contains time stage identification, demand change direction description, and change frequency statistical information.

[0023] In the construction industry, the bidding requirements of similar projects change over time. By analyzing the bidding requirement change records of similar projects in the industry, the time series information is extracted, and the records are divided into different time stages according to the time sequence, such as the past year, the past two years, etc. At the same time, the demand change content is extracted, such as "increasing the application of intelligent equipment" and "improving environmental protection standards".

[0024] The trend feature of the demand change content is identified, and the demand focus change direction of different time stages is determined. For example, in a certain time period, the application of intelligent equipment becomes the demand focus. The frequency of demand change in each time stage is counted, and the demand focus change direction of each time stage is taken as a trend correlation node. Each trend correlation node contains time stage identification, such as "past year", demand change direction description, such as "increasing the application of intelligent equipment", and change frequency statistical information, reflecting the frequency of the demand change.

[0025] Step S1141: Perform text structuring processing on the bidding requirement change records of similar projects in the industry, eliminate redundant descriptions and repetitive information in the bidding requirement change records, and retain the core content containing time identification, demand change items and change reasons, to generate a structured demand change record list.

[0026] The bidding requirement change records of similar projects in the industry may contain a large amount of text information, which contains some redundant descriptions and repetitive content. The text structuring processing is performed on the above records, and the redundant and repetitive information is first identified and eliminated. For example, some insignificant adjectives, repeated similar descriptions, etc.

[0027] The core content of the record is retained, including time identification, which clearly indicates the time of the demand change; demand change items, which are the specific demand change content; and change reasons, which explain why the above demand change occurs. The above core content is sorted to generate a structured demand change record list.

[0028] Step S1142: Extract time series information from the structured demand change record list, and divide the records in the structured demand change record list into multiple continuous time stages according to the time identification, and determine the length of each time stage according to the bidding period of similar projects, and assign a unique time stage identification to each time stage.

[0029] The time series information is extracted from the structured demand change record list generated above, which reflects the time sequence of the demand change. According to the time identification, the records in the list are divided into multiple continuous time stages. The length of the time stage is determined according to the bidding period of similar projects, because the bidding period can reflect a relatively stable time span of demand change in the industry.

[0030] Assign a unique time stage identifier to each time period to facilitate the differentiation and analysis of demand changes across different time periods. For example, assign the identifier "Stage A" to the time period of the past year, and assign the identifier "Stage B" to the time period between the past two years and one year, and so on.

[0031] Step S1143: For each time period, summarize the content of all requirement change items within that time period, perform topic clustering on the content of the requirement change items, and group the requirement change items that describe the same type of requirement change into a change topic. Each change topic corresponds to a type of requirement change direction.

[0032] For each defined time period, summarize the contents of all requirement change items within that period. For example, during a certain time period, requirement change items might include "increasing the use of solar energy equipment," "improving the building's intelligent control level," and "adopting more efficient energy-saving technologies," etc.

[0033] The content of the aforementioned requirement change items was subjected to topic clustering, grouping items expressing similar requirement changes into a single change topic. In the example above, "increasing the use of solar energy equipment" and "adopting more efficient energy-saving technologies" can be categorized under the change topic of "energy conservation and environmental protection," while "improving the level of intelligent control in buildings" can be categorized under the change topic of "intelligent applications." Each change topic corresponds to a type of requirement change direction, reflecting the industry's demand change trends in certain aspects during that time period.

[0034] Step S1144: Count the number of times each change topic appears in this time period, calculate the proportion of the number of occurrences to the total number of change items in this time period, determine the change topic with the highest proportion as the key change direction of the requirements in this time period, and generate a description of the change direction of the requirements.

[0035] Count the number of times each change topic appears within that time period. For example, in the requirement change items for a certain time period, the topic "energy conservation and environmental protection" appeared several times, and the topic "intelligent application" also appeared a certain number of times.

[0036] Calculate the proportion of each change topic's occurrences relative to the total number of change items within that time period. Identify the change topic with the highest proportion as the key area of ​​demand change for that time period. For example, if the "energy conservation and environmental protection" topic has the highest occurrence rate, then the key area of ​​demand change for that time period is an increase in demand related to energy conservation and environmental protection. Based on this key area of ​​demand change, generate a detailed description of the demand change direction, such as "During this time period, the construction industry's demand for energy-saving and environmentally friendly technologies and equipment has increased significantly."

[0037] Step S1145: Calculate the ratio of the total number of demand change entries in each time phase to the length of the time phase, to obtain the demand change frequency of the time phase. Integrate the time phase identifier, demand change direction description, and change frequency statistical information to generate the complete information of each trend association node.

[0038] The ratio of the total number of demand change entries in each time phase to the length of the time phase is calculated to obtain the demand change frequency of the time phase. The ratio reflects the frequency of demand changes in the time phase.

[0039] Integrate the time phase identifier, demand change direction description, and change frequency statistical information to generate the complete information of each trend association node. For example, for "Phase A", the demand change direction description is "increasing intelligent device application", and the change frequency is a relatively high value. Combine the above information to form the complete description of the trend association node.

[0040] Step S115: Calculate the semantic similarity value between the demand node and the multi-modal material node, determine the initial semantic association strength information between the demand node and the multi-modal material node according to the semantic similarity value, calculate the demand matching degree value between the demand node and the trend association node, and determine the initial semantic association strength information between the demand node and the trend association node according to the demand matching degree value; wherein the value of the initial semantic association strength information is calculated by a preset linear or nonlinear mapping function from the corresponding semantic similarity value or demand matching degree value.

[0041] In the process of constructing the bidding scene semantic graph, it is necessary to determine the initial semantic association strength information between the demand node and the multi-modal material node and the demand node and the trend association node.

[0042] For the demand node and the multi-modal material node, calculate the semantic similarity value between them. For example, the demand node is "adopting environmentally friendly and energy-saving materials", and one of the text materials in the multi-modal material node is an introduction to a new type of environmentally friendly and energy-saving material. The semantic similarity between the two is calculated by semantic analysis method.

[0043] According to the calculated semantic similarity value, the initial semantic association strength information between the demand node and the multi-modal material node is determined by a preset linear or nonlinear mapping function. If the semantic similarity is high, the value of the initial semantic association strength information will also be relatively high, indicating that the semantic connection between them is close.

[0044] For the demand node and the trend association node, the demand matching degree value between them is calculated. For example, the demand node is "improve the intelligent level of the building", and the demand change direction reflected by the trend association node is "increase the application of intelligent equipment". The demand matching degree of the two is calculated.

[0045] According to the demand matching degree value, the initial semantic association strength information between the demand node and the trend association node is determined by a preset linear or nonlinear mapping function. The higher the demand matching degree is, the higher the value of the initial semantic association strength information is, indicating that the demand node and the trend association node are more closely related in semantics.

[0046] Step S116: Construct a dynamic association update rule, adjust the initial semantic association strength information between the demand node, the multi-modal material node and the trend association node according to the latest trend characteristics of the bid demand change record, generate dynamic semantic association strength information, input all demand nodes, all multi-modal material nodes, all trend association nodes and dynamic semantic association strength information into a graph construction tool, and generate a bid scene semantic graph taking nodes as cores and dynamic semantic association strength information as connecting edges. The update frequency of the connecting edges in the bid scene semantic graph is consistent with the change frequency of the bid demand change record.

[0047] The dynamic association update rule is constructed for adjusting the initial semantic association strength information between the nodes according to the latest trend characteristics of the bid demand change record. For example, when a new technology trend appears in the industry, such as the widespread application of a new type of building material, the bid demand may change accordingly.

[0048] According to the latest trend characteristics, the initial semantic association strength information between the demand node, the multi-modal material node and the trend association node is adjusted. If the matching degree between a demand node and a new trend association node increases, the semantic association strength information value between them will be increased accordingly.

[0049] Through the above adjustment, the dynamic semantic association strength information is generated. All demand nodes, multi-modal material nodes, trend association nodes and dynamic semantic association strength information are input into a graph construction tool. The graph construction tool takes these nodes as cores and dynamic semantic association strength information as connecting edges to generate a bid scene semantic graph. The update frequency of the connecting edges in the bid scene semantic graph is consistent with the change frequency of the bid demand change record.

[0050] Step S1161: Set an association update period, and the length of the association update period is determined according to the change frequency of the bid demand change record.

[0051] An association update period is set, and the length of the association update period is determined according to the change frequency of the bidding demand change record. If the bidding demand changes frequently, the association update period is relatively short, so as to timely adjust the semantic association strength information between the nodes. If the demand changes relatively slowly, the association update period can be appropriately lengthened. For example, in the active period of technological innovation in the construction industry, the bidding demand changes fast, and the association update period can be set to a relatively short period of time. In the relatively stable period of the industry, the update period can be appropriately lengthened.

[0052] Step S1162: In each association update period, the latest trend characteristics in the bidding demand change record are extracted, the latest demand focus change direction and the latest change frequency are determined, and the latest demand focus change direction and the demand similarity value of each demand node are calculated.

[0053] In each association update period, the latest trend characteristics in the bidding demand change record are extracted, and the above characteristics reflect the latest change of the bidding demand in the current industry, such as the appearance of new environmental protection standards, new intelligent technology applications, etc.

[0054] According to the above latest trend characteristics, the latest demand focus change direction is determined, for example, “promote new prefabricated building technology”. At the same time, the latest change frequency is counted to know the frequency of demand change.

[0055] The latest demand focus change direction and the demand similarity value of each demand node are calculated. For example, the demand nodes are “traditional building construction method”, “new building material application”, etc., and the demand similarity of “promote new prefabricated building technology” and these demand nodes is calculated. If the demand node is similar in semantics to the latest demand focus change direction, the demand similarity value will be higher.

[0056] Step S1163: According to the calculated demand similarity value, the initial semantic association strength information between the demand node and the trend association node is adjusted. When the demand similarity value is greater than a preset first threshold, the initial semantic association strength information value of the corresponding demand node and the trend association node is increased, and when the demand similarity value is less than a preset second threshold, the initial semantic association strength information value of the corresponding demand node and the trend association node is decreased.

[0057] According to the calculated demand similarity value, the initial semantic association strength information between the demand node and the trend association node is adjusted. When the demand similarity value is greater than a preset first threshold, the initial semantic association strength information value of the corresponding demand node and the trend association node is increased, and when the demand similarity value is less than a preset second threshold, the initial semantic association strength information value of the corresponding demand node and the trend association node is decreased.

[0058] When the demand similarity value is less than the preset second threshold value, it indicates that the matching degree of the demand node and the latest demand focus change direction is low, and the initial semantic association strength information value of the demand node and the trend association node is reduced, weakening the semantic association between them. For example, if the demand node "traditional construction method" and the demand similarity value of "promote new type of fabricated building technology" are less than the second threshold value, the semantic association strength information value of the demand node and the related trend association node will be reduced.

[0059] Step S1164: Extract the multi-modal material demand under the latest trend characteristics, determine the multi-modal material type suitable for the latest demand and the core theme suitable for the latest demand, and calculate the theme matching degree value of the core theme information of each multi-modal material node and the latest demand.

[0060] Extract the multi-modal material demand under the latest trend characteristics, and determine the multi-modal material type and core theme suitable for it according to the latest demand focus change direction. For example, when the latest demand focus change direction is "promote new type of fabricated building technology", the suitable multi-modal material types may include technical description in text form, construction process in chart form, cost calculation in formula form, etc., and the core theme is "application of new type of fabricated building technology".

[0061] Calculate the theme matching degree value of the core theme information of each multi-modal material node and the latest demand. If the core theme of the multi-modal material node is similar to "application of new type of fabricated building technology", the theme matching degree value will be relatively high. For example, the core theme of a text material node is "introduction of a certain type of fabricated building technology", and its theme matching degree with the latest demand will be relatively high.

[0062] Step S1165: Adjust the initial semantic association strength information between the demand node and the multi-modal material node according to the calculated theme matching degree value. When the theme matching degree value is greater than a preset third threshold value, the initial semantic association strength information value of the multi-modal material node and the demand node is increased, and when the theme matching degree value is less than a preset fourth threshold value, the initial semantic association strength information value of the multi-modal material node and the demand node is reduced.

[0063] According to the calculated theme matching degree value, adjust the initial semantic association strength information between the demand node and the multi-modal material node. When the theme matching degree value is greater than the preset third threshold value, it indicates that the core theme of the multi-modal material node has a high matching degree with the latest demand, and the initial semantic association strength information value of the multi-modal material node and the demand node is increased, strengthening the semantic association between them.

[0064] When the theme matching degree value is less than the fourth preset threshold value, it indicates that the core theme of the multi-modal material node matches the latest demand at a low degree, and the initial semantic association strength information value of the multi-modal material node and the demand node is reduced, weakening the semantic connection between them. For example, if the core theme of a chart material node matches the theme of "promoting new assembly type building technology" at a degree value less than the fourth threshold value, the semantic association strength information value of the multi-modal material node and the related demand node will be reduced.

[0065] Step S1166: aggregate all adjusted initial semantic association strength information to generate dynamic semantic association strength information, record the time of each adjustment, the basis of each adjustment, and the change of the initial semantic association strength information value before and after adjustment, and generate a dynamic association update log.

[0066] Aggregate all adjusted initial semantic association strength information to form dynamic semantic association strength information, which reflects the latest semantic association tightness between nodes.

[0067] Record the time of each adjustment, the basis of adjustment, and the change of the initial semantic association strength information value before and after adjustment. For example, record the adjustment time as a specific date, the adjustment basis as the appearance of a new industry technology trend, the initial semantic association strength information value of two nodes before adjustment as a value, and the value after adjustment as another value. Organize the above records into a dynamic association update log, which can be used for subsequent analysis and backtracking to know the dynamic change process of the bidding scene semantic graph.

[0068] Step S120: according to the hierarchical relationship of the demand node in the bidding scene semantic graph and the evolution characteristics of the trend association node, perform adaptive decomposition processing on the overall structure of the electronic tender to generate an electronic tender dynamic framework containing multiple functional module units and dynamic connection rules between the functional module units, and each functional module unit corresponds to a group of associated demand nodes, associated multi-modal material nodes and associated trend association nodes in the bidding scene semantic graph.

[0069] After the bidding scene semantic graph is constructed, the overall structure of the electronic tender is adaptively decomposed according to the hierarchical relationship of the demand node in the graph and the evolution characteristics of the trend association node. The hierarchical relationship of the demand node reflects the primary and secondary relationship of the bidding demand, and the evolution characteristics of the trend association node reflect the change trend of the demand over time.

[0070] Through the above disassembly processing, an electronic tender dynamic framework is generated, which includes a plurality of functional module units and dynamic connection rules between the functional module units. Each functional module unit corresponds to a group of associated demand nodes, associated multi-modal material nodes and associated trend association nodes in the tender scenario semantic graph. For example, in the construction engineering tender, a functional module unit may correspond to the demand nodes, multi-modal material nodes and trend association nodes related to "architectural design", including design scheme text, design drawing charts and design concept trend over time, etc.

[0071] Step S121: Analyze the hierarchical relationship of the demand nodes in the tender scenario semantic graph, and divide the demand nodes with subordinate relationship into different levels by a semantic hierarchical clustering algorithm. The demand node at the highest level is the core demand node, the demand node at the next level is the secondary demand node, and so on, to generate a demand node hierarchical structure.

[0072] The hierarchical relationship of the demand nodes in the tender scenario semantic graph is analyzed, and a semantic hierarchical clustering algorithm is used to divide the demand nodes with subordinate relationship. In the construction engineering tender scenario, the demand nodes can include "overall function requirements of the building", "functional layout of each floor", "room interior decoration standards", etc.

[0073] The above demand nodes are divided into different levels by a semantic hierarchical clustering algorithm. Among them, "overall function requirements of the building" is at the highest level, which is the core demand node, and it is the overall goal of the whole project; "functional layout of each floor" is at the next level, which is the secondary demand node, and it is the refinement of the core demand node at the floor level; "room interior decoration standards" is at a further lower level, which further refines the requirements of the floor functional layout.

[0074] The above divided levels are arranged to generate a demand node hierarchical structure, which shows the hierarchical relationship between the demand nodes.

[0075] Step S1211: Extract the core demand vocabulary of all demand nodes and the demand constraint conditions of all demand nodes in the tender scenario semantic graph, and splice the core demand vocabulary of each demand node with the demand constraint conditions of the demand node into a demand feature text.

[0076] The core demand vocabulary and the demand constraint conditions of all demand nodes are extracted from the tender scenario semantic graph. For example, the demand node "the building should adopt environmentally friendly and energy-saving materials and meet the relevant national standards", the core demand vocabulary is "environmentally friendly and energy-saving materials", and the demand constraint conditions are "should adopt" and "meet the relevant national standards".

[0077] The core requirement vocabulary of each requirement node is spliced with the requirement constraint condition to form a requirement feature text. For the above example, the requirement feature text is "the building should use environmentally friendly and energy-saving materials and meet the relevant national standards". By the above manner, the key information of each requirement node is integrated into a complete text description, which is convenient for subsequent coding processing.

[0078] Step S1212: calling a pre-trained semantic coding model, performing coding processing on the requirement feature text of each requirement node to generate a requirement node semantic vector, and the dimensions of all requirement node semantic vectors remain consistent.

[0079] The pre-trained semantic coding model is called to perform coding processing on the requirement feature text of each requirement node. The semantic coding model converts text information into vector form for subsequent similarity calculation.

[0080] In the coding process, the dimensions of all requirement node semantic vectors remain consistent. For example, regardless of the length and content of the requirement feature text, the coded semantic vector has the same dimension. This allows the semantic vectors of different requirement nodes to be compared and analyzed in the same dimensional space.

[0081] Step S1213: calculating the cosine similarity between any two requirement node semantic vectors, constructing a requirement node association matrix according to the cosine similarity, and the element value in the requirement node association matrix is the semantic similarity of the corresponding two requirement nodes.

[0082] The cosine similarity between any two requirement node semantic vectors is calculated. The cosine similarity is used to measure the similarity in direction between two vectors, and the closer the value is to 1, the more similar the semantics of the two requirement nodes are.

[0083] According to the calculated cosine similarity, a requirement node association matrix is constructed. The element value in the matrix is the semantic similarity of the corresponding two requirement nodes. For example, a certain element in the matrix represents the semantic similarity of requirement node A and requirement node B. Through the association matrix, the semantic association degree between each requirement node can be intuitively obtained.

[0084] Step S1214: performing clustering processing on the requirement nodes using a hierarchical clustering algorithm, taking the requirement node association matrix as input, and grouping the requirement nodes with a semantic similarity higher than a clustering distance threshold into the same cluster, and each cluster corresponds to a requirement level.

[0085] The hierarchical clustering algorithm is used to cluster the requirement nodes, taking the requirement node association matrix as input. The hierarchical clustering algorithm groups the requirement nodes with a semantic similarity higher than a clustering distance threshold into the same cluster according to the semantic similarity between the requirement nodes.

[0086] Each cluster corresponds to a demand level. For example, demand nodes with high semantic similarity are clustered into the same cluster, which may belong to the same level of demand. By the above clustering method, the demand nodes are divided into different hierarchical structures.

[0087] Step S1215: Perform importance evaluation on the demand nodes in each cluster, calculate the sum of dynamic semantic association strength information of each demand node and other demand nodes outside the cluster, and the demand node with the maximum sum of dynamic semantic association strength information is the core node of the cluster, that is, the core demand node of the corresponding demand level, and the remaining demand nodes in the cluster are the secondary demand nodes of the corresponding demand level.

[0088] The importance of each demand node in the cluster is evaluated. The sum of dynamic semantic association strength information of each demand node and other demand nodes outside the cluster is calculated, which reflects the semantic contact degree of the demand node and other level demand nodes.

[0089] The demand node with the maximum sum of dynamic semantic association strength information is determined as the core node of the cluster, that is, the core demand node of the corresponding demand level. The remaining demand nodes in the cluster are the secondary demand nodes of the corresponding demand level. For example, in a cluster, a demand node has the maximum sum of semantic association strength information with other level demand nodes, so it is the core demand node of the level, and other nodes are the secondary demand nodes around it.

[0090] Step S1216: Arrange each demand level from high to low according to the hierarchical relationship of the cluster, generate a demand node hierarchical structure, and record the core demand node of each demand level, the secondary demand node of each demand level, and the subordinate relationship between the core demand node and the secondary demand node.

[0091] Arrange each demand level from high to low according to the hierarchical relationship of the cluster. For example, arrange the cluster in the highest level at the front, and arrange the clusters of other levels in turn.

[0092] Generate a demand node hierarchical structure and record the core demand node, the secondary demand node of each demand level, and the subordinate relationship between the core demand node and the secondary demand node. For example, record the subordinate relationship between the core demand node "overall function requirement of building" and the secondary demand node "function layout of each floor", and clarify the position and relationship of each demand node in the hierarchical structure.

[0093] Step S122: Combine the evolution characteristics of the trend association nodes to determine the demand evolution direction of each core demand node at different time stages, and match the corresponding trend association nodes to each core demand node according to the demand evolution direction to generate target association pairs.

[0094] In combination with the evolution characteristics of the trend-related nodes, the evolution direction of each core demand node at different time stages is analyzed. In the construction engineering bidding, the core demand node "building intelligent degree" may change with the development of industry technology at different time stages, such as evolving from simple automation control to higher artificial intelligence control.

[0095] According to the demand evolution direction, the corresponding trend-related node is matched for each core demand node. For example, if the demand evolution direction of the core demand node "building intelligent degree" is to develop towards higher intelligent technology, the trend-related node related thereto is matched, such as "application trend of new intelligent building technology".

[0096] The core demand node and the matched trend-related node are combined into a target-related pair, such as ("building intelligent degree", "application trend of new intelligent building technology").

[0097] Step S123: Based on each target-related pair, the multi-modal material nodes in the bidding scene semantic graph that have a dynamic semantic association strength information value greater than or equal to a preset association strength threshold with the core demand node and the trend-related node are screened, and the core demand node, the matched trend-related node and the screened multi-modal material nodes are combined into an association node group.

[0098] Based on each target-related pair, the multi-modal material nodes in the bidding scene semantic graph are screened. The screening condition is that the multi-modal material nodes have a dynamic semantic association strength information value greater than or equal to a preset association strength threshold with the core demand node and the trend-related node.

[0099] For example, for the target-related pair ("building intelligent degree", "application trend of new intelligent building technology"), multi-modal material nodes that have a semantic association strength information value meeting the threshold requirement with the two nodes are searched in the graph, such as introduction text about new intelligent equipment, design chart of intelligent system, etc.

[0100] The core demand node, the matched trend-related node and the screened multi-modal material nodes are combined into an association node group. The association node group contains a group of nodes related to specific demand and trend, and provides a specific node set for generating a functional module unit.

[0101] Step S124: Integrate the demand content, corresponding trend information and corresponding multi-modal material information of each associated node group to generate a functional module unit. Each functional module unit contains a functional module unit theme name, a functional module unit core demand expression, a functional module unit trend adaptation description and a functional module unit multi-modal material index list. The functional module unit multi-modal material index list records the identification and storage path of all multi-modal material nodes corresponding to the functional module unit.

[0102] Integrate the demand content, trend information and multi-modal material information of each associated node group to generate a functional module unit. The functional module unit has a clear structure and information content.

[0103] The functional module unit theme name is determined according to the core demand node and the trend associated node of the associated node group, such as "building intelligent degree improvement and new technology application". The functional module unit core demand expression describes the specific content of the core demand node in detail, such as "improve the intelligent control level of the building, realize the automatic operation and data interaction of the equipment".

[0104] The functional module unit trend adaptation description explains how the functional module unit adapts to the demand change trend represented by the trend associated node, such as "adopt new intelligent building technology to meet the industry's demand for continuously improving the intelligent degree of buildings".

[0105] The functional module unit multi-modal material index list records the identification and storage path of all multi-modal material nodes corresponding to the functional module unit, which facilitates subsequent search and use of related materials. For example, the list records the identification and storage location of the text about new intelligent equipment introduction, and the identification and storage path of the intelligent system design chart.

[0106] Step S125: Analyze the semantic association relationship between adjacent functional module units, generate dynamic connection rules between functional module units according to the number of overlapping nodes of the associated node group and the dynamic semantic association strength information of the associated node group. The dynamic connection rules between functional module units include the expression logic of the connection content, the modal conversion mode of the connection content and the information supplement requirements of the connection content.

[0107] Analyze the semantic association relationship between adjacent functional module units, and determine the degree of connection between them through the number of overlapping nodes and the dynamic semantic association strength information of the associated node group. For example, adjacent functional module units "building intelligent degree improvement and new technology application" and "building energy saving and environmental protection technology upgrading" have some overlapping multi-modal material nodes in their associated node groups, and the dynamic semantic association strength information between these nodes is high, indicating that the semantic association between them is relatively close.

[0108] According to the analysis result, a dynamic connection rule between functional module units is generated. The expression logic of the connection content determines the way of content transition between adjacent functional module units, such as summarizing the core content of the previous functional module unit first, and then introducing the related content of the next functional module unit.

[0109] The modal conversion mode of the connection content determines how the presentation form of the multi-modal material is converted between different functional module units. For example, the transition from the introduction of intelligent technology in text form to the display of energy-saving and environmental protection data in chart form.

[0110] The information supplement requirement of the connection content determines which information needs to be supplemented in the connection process. For example, when transitioning from the intelligent module to the energy-saving and environmental protection module, the technical association information between the two is supplemented.

[0111] Step S1251: Extract the associated node groups corresponding to the adjacent two functional module units, count the number of overlapping demand nodes, overlapping trend association nodes and overlapping multi-modal material nodes in the two associated node groups, calculate the proportion of the number of overlapping nodes to the total number of nodes in the two associated node groups, and take the proportion as the node overlap degree.

[0112] The associated node groups corresponding to the adjacent two functional module units are extracted, and the number of overlapping demand nodes, trend association nodes and multi-modal material nodes in each of them is counted. For example, there are some overlapping demand nodes, trend association nodes and multi-modal material nodes in the associated node groups of adjacent functional module units A and B.

[0113] The proportion of the number of overlapping nodes to the total number of nodes in the two associated node groups is calculated, and the proportion is taken as the node overlap degree. The node overlap degree reflects the degree of content overlap between adjacent functional module units. The higher the proportion, the closer the association between the two functional module units.

[0114] Step S1252: Extract the dynamic semantic association strength information between the overlapping nodes in the two associated node groups, and calculate the average value of the dynamic semantic association strength information, which is taken as the average association strength.

[0115] The dynamic semantic association strength information between the overlapping nodes in the two associated node groups is extracted, which represents the closeness of the semantic connection between the overlapping nodes.

[0116] The average value of these dynamic semantic association strength information is calculated, and the average value is taken as the average association strength. The average association strength further measures the closeness of the semantic association between adjacent functional module units. The higher the average value, the closer the semantic connection between them.

[0117] Step S1253: determining the association level of the adjacent functional module units according to the node overlap value and the average value of the association strength; wherein, the adjacent functional module units with the node overlap value greater than or equal to the first overlap threshold value and the average value of the association strength greater than or equal to the first strength threshold value are of the first association level; the adjacent functional module units with the node overlap value less than the first overlap threshold value and greater than or equal to the second overlap threshold value, and the average value of the association strength less than the first strength threshold value and greater than or equal to the second strength threshold value are of the second association level; the adjacent functional module units with the node overlap value less than the second overlap threshold value and the average value of the association strength less than the second strength threshold value are of the third association level; different association levels correspond to different connection strategies.

[0118] The association level of the adjacent functional module units is determined according to the node overlap value and the average value of the association strength. The first overlap threshold value and the first strength threshold value, the second overlap threshold value and the second strength threshold value are set. If the node overlap value is greater than or equal to the first overlap threshold value and the average value of the association strength is greater than or equal to the first strength threshold value, the adjacent functional module units are of the first association level, indicating that the association between them is very close. If the node overlap value is less than the first overlap threshold value and greater than or equal to the second overlap threshold value, and the average value of the association strength is less than the first strength threshold value and greater than or equal to the second strength threshold value, the adjacent functional module units are of the second association level, and the association degree is moderate. If the node overlap value is less than the second overlap threshold value and the average value of the association strength is less than the second strength threshold value, the adjacent functional module units are of the third association level, and the association degree is weak. Different association levels correspond to different connection strategies, so as to reasonably connect the contents according to the association closeness of the adjacent functional module units.

[0119] Step S1254: for the adjacent functional module units of the first association level, the expression logic of the connection content is that the latter functional module unit supplements the new demand content after the trend evolution on the basis of accepting the core content of the former functional module unit; the modal conversion mode of the connection content is to keep the main modal types of the former and latter functional module units consistent, and only supplement the auxiliary modal that adapts to the new demand; the information supplement requirement of the connection content is to clearly mark the association points and difference points of the contents of the former and latter functional module units.

[0120] For the adjacent functional module units of the first association level, the expression logic of the connection content is generated. The latter functional module unit supplements the new demand content after the trend evolution on the basis of accepting the core content of the former functional module unit. For example, the former functional module unit introduces the basic technology of current building intelligence, and the latter functional module unit supplements the application and development trend of new intelligent technology on the basis of accepting these contents.

[0121] The mode conversion method of the connecting content is to keep the main modal type of the two adjacent functional module units consistent, and only supplement the auxiliary modal of the new demand. If the previous functional module unit introduces the intelligent technology in the form of text, the subsequent functional module unit also mainly uses text, supplemented by some new technology application cases in the form of charts as auxiliary modal.

[0122] The information supplement requirement of the connecting content is to clearly mark the connection points and difference points of the content of the two adjacent functional module units. For example, clearly point out the association and difference between the new intelligent technology and the existing technology, so that the reader can clearly understand the transition and development of the content.

[0123] Step S1255: For the adjacent functional module units of the second association level, the expression logic of the connecting content is to respectively elaborate the demand content of different dimensions of the two adjacent functional module units, and establish the association through the common node; the mode conversion method of the connecting content is to adjust the modal type according to the theme of the functional module unit; and the information supplement requirement of the connecting content is to mark the content association corresponding to the common node.

[0124] For the adjacent functional module units of the second association level, the expression logic of the connecting content is to respectively elaborate the demand content of different dimensions of the two adjacent functional module units, and establish the association through the common node. For example, one functional module unit introduces building intelligence from the technical point of view, and the other functional module unit introduces intelligent construction from the cost point of view, and the two are associated through the common intelligent equipment node.

[0125] The mode conversion method of the connecting content is to adjust the modal type according to the theme of the functional module unit. If the previous functional module unit shows the performance index of the intelligent technology in the form of chart, the subsequent functional module unit analyzes the cost composition of the intelligent construction in the form of text.

[0126] The information supplement requirement of the connecting content is to mark the content association corresponding to the common node. Clearly point out the association of the common intelligent equipment in the two dimensions of technology and cost, and help the reader understand the connection between the contents of different dimensions.

[0127] Step S1256: For the adjacent functional module units of the third association level, the expression logic of the connecting content is to connect the two adjacent functional module units through the newly added transitional demand content; the mode conversion method of the connecting content is to reselect the modal type according to the theme of the subsequent functional module unit; and the information supplement requirement of the connecting content is to introduce the transition content and the association logic of the two adjacent functional module units in detail, integrate the connection strategies of different association levels, and generate the dynamic connection rule between the functional module units.

[0128] For adjacent functional module units of the third correlation level, the expression logic of the bridging content is to connect the two functional module units before and after by adding transitional demand content. For example, when transitioning from the building intelligentization function module to the building energy saving and environmental protection function module, transitional content about the application potential of intelligent technology in energy saving and environmental protection is added.

[0129] The modal conversion mode of the bridging content is to reselect the modal type according to the theme of the next functional module unit. If the next functional module unit displays energy saving and environmental protection data in the form of a chart, a transition mode suitable for chart display is adopted when bridging.

[0130] The information supplement requirement of the bridging content is to introduce the transitional content and the correlation logic of the two functional module units in detail. For example, it is explained in detail how intelligent technology realizes energy saving and environmental protection effect through energy saving control algorithm, and different correlation level bridging strategies are integrated to form complete dynamic bridging rules between functional module units.

[0131] Step S126: Score and sort all functional module units according to the importance of the core demand nodes, combine the dynamic bridging rules between functional module units, and generate an electronic tender dynamic framework.

[0132] Score and sort all functional module units according to the importance of the core demand nodes. The importance score can be determined according to the position of the core demand node in the bidding demand, the influence degree on the project, and other factors.

[0133] Combine the sorted functional module units according to the dynamic bridging rules between functional module units. According to the dynamic bridging rules, the content bridging and modal conversion between adjacent functional module units are reasonably arranged.

[0134] Through the above-mentioned manner, the electronic tender dynamic framework is generated, which clearly defines the overall structure of the electronic tender, including the order of each functional module unit and the bridging mode between them.

[0135] Step S130: Call the pre-trained cross-modal artificial intelligence content generation model, input the electronic tender dynamic framework, the dynamic semantic correlation strength information of the bidding scene semantic graph, and the multi-modal material analysis rules, perform cross-modal material correlation generation processing on each functional module unit, and obtain the multi-modal module content preliminary draft corresponding to each functional module unit.

[0136] Call the pre-trained cross-modal artificial intelligence content generation model, input the electronic tender dynamic framework, the dynamic semantic correlation strength information of the bidding scene semantic graph, and the multi-modal material analysis rules. The cross-modal artificial intelligence content generation model has the ability to process different modal information, and can perform cross-modal material correlation generation processing on each functional module unit according to the input information.

[0137] In the construction project bidding scenario, for each functional module unit, such as the "architectural design" functional module unit, the model combines the demand content, trend information, and multi-modal material information corresponding to the functional module unit, determines the reference weight of the material according to the dynamic semantic correlation strength information, analyzes the material according to the multi-modal material analysis rule, and finally generates the multi-modal module content preliminary draft corresponding to the functional module unit.

[0138] Step S131: For each functional module unit, extract the functional module unit topic name of the functional module unit, the functional module unit core demand expression of the functional module unit, the functional module unit trend adaptation description of the functional module unit, and the functional module unit multi-modal material index list of the functional module unit from the electronic tender dynamic framework.

[0139] For each functional module unit, extract the key information from the electronic tender dynamic framework. The functional module unit topic name clearly defines the core topic of the functional module unit, such as "improve the intelligent degree of building".

[0140] The functional module unit core demand expression describes the core demand of the functional module unit in detail, such as "realize the intelligent control and management of building equipment". The functional module unit trend adaptation description explains how the functional module unit adapts to the changing trend of industry demand, such as "adopt new intelligent technology to meet future development needs".

[0141] The functional module unit multi-modal material index list records the identification and storage path of all multi-modal material nodes corresponding to the functional module unit, which facilitates the retrieval of related materials. For example, it records the identification and storage location of the intelligent device introduction text, as well as the identification and storage path of the intelligent system design chart.

[0142] Step S132: According to the material identification in the functional module unit multi-modal material index list and the material storage path in the functional module unit multi-modal material index list, obtain the corresponding multi-modal material from the historical tender material set, and divide the multi-modal material into a text material subset, a chart material subset, and a formula material subset according to the material type identification.

[0143] According to the material identifier and storage path in the functional module unit multi-modal material index list, the corresponding multi-modal material is obtained from the historical bid material set. In construction engineering, these materials may include construction plans in text form, engineering progress plans in chart form, cost calculation formulas, etc. According to the material type identifier, the obtained multi-modal material is divided into a text material subset, a chart material subset, and a formula material subset. For example, all text-form materials are classified into the text material subset, all chart-form materials are classified into the chart material subset, and all formula-form materials are classified into the formula material subset.

[0144] Step S133: Generate multi-modal material analysis rules, text material analysis rules for extracting key arguments in text and argument logic in text, chart material analysis rules for extracting data relationships in charts and trend conclusions in charts, formula material analysis rules for extracting application scenarios of formulas and calculation logic of formulas, and perform analysis processing on the text material subset, chart material subset, and formula material subset according to the multi-modal material analysis rules to obtain material analysis results.

[0145] Generate multi-modal material analysis rules, and develop different analysis methods for different types of materials. For text materials, the analysis rule is to extract the key arguments and argument logic therein. For example, in the construction plan text, extract the key arguments about the construction process and the logic of demonstrating the feasibility of these processes.

[0146] For chart materials, the analysis rule is to extract the data relationships in the charts and the trend conclusions. For example, in the engineering progress plan chart, extract the data relationships of the engineering progress in each stage and the trend conclusion of the engineering progress over time.

[0147] For formula materials, the analysis rule is to extract the application scenarios of formulas and the calculation logic. For example, in the cost calculation formula, extract the application scenarios of the formula for which project cost calculation is applicable, and the calculation logic of each parameter in the formula. According to the above analysis rules, the text material subset, chart material subset, and formula material subset are analyzed to obtain material analysis results, which contain the key information of each type of material.

[0148] For example, step S1331: For the text material subset, a text structured analysis algorithm is used to perform paragraph division on each text material, identify the center sentence of each paragraph, extract key argument words from the center sentence, analyze the logical relationship between adjacent paragraphs, determine the argument logic framework of the text material, integrate the key argument words and the argument logic framework, and generate the text material analysis result.

[0149] For the text material subset, a text structured analysis algorithm is used for processing. First, paragraph division is performed on each text material, and the text is divided into different paragraphs according to content logic.

[0150] Identify the central sentence of each paragraph, which usually summarizes the main content of the paragraph. Extract key argument words from the central sentence, such as "new construction technology" "efficient construction method" in the text about construction technology.

[0151] Analyze the logical relationship between adjacent paragraphs to determine the argument logical framework of the text material. For example, there may be cause-and-effect relationship, progressive relationship, etc. between paragraphs. Integrate the key argument words with the argument logical framework to generate the text material analysis result, which shows the core argument and argument logic of the text material.

[0152] Step S1332: For the subset of chart materials, use image recognition algorithm to identify chart type, extract chart axis label, chart data node value and chart title, analyze data node trend by data fitting algorithm, determine data relationship such as increase and decrease relationship, proportion relationship, etc. Integrate chart type, axis label, data node value, data relationship and trend conclusion to generate chart material analysis result.

[0153] For the subset of chart materials, use image recognition algorithm to identify chart type, such as column chart, line chart, pie chart, etc. Extract the coordinate axis label of the chart to clarify the meaning and range of the data in the chart.

[0154] Extract the data node values of the chart, which are the specific data displayed by the chart. At the same time, get the title of the chart, which summarizes the main content of the chart.

[0155] Analyze the trend of data nodes by data fitting algorithm to determine the relationship between data such as increase and decrease, proportion, etc. For example, analyze the increase and decrease trend of data over time in line chart, and determine the proportion relationship of each part of data in pie chart. Integrate chart type, axis label, data node value, data relationship and trend conclusion to generate chart material analysis result, which describes the information and data characteristics contained in the chart in detail.

[0156] Step S1333: For the subset of formula materials, use formula recognition and analysis algorithm to extract variable symbols in the formula, operator symbols in the formula and parameter definitions in the formula, combine with the context text of the formula material to determine the application scenario of the formula, analyze the calculation logic between variables in the formula by logical reasoning, clarify the input parameters of the formula, the calculation steps of the formula and the meaning of the output results of the formula, integrate variable symbols, parameter definitions, application scenarios and calculation logic to generate formula material analysis result.

[0157] For the formula material subset, a formula recognition and analysis algorithm is used for processing. The variable symbols, operator symbols and parameter definitions in the formula are extracted, and the basic components of the formula are determined.

[0158] In combination with the context text of the formula, the application scenario of the formula is determined. For example, in the text related to cost calculation, it is determined that the formula is applicable to which type of project cost calculation.

[0159] Through logical reasoning analysis of the calculation logic between variables in the formula, the input parameters, calculation steps and output result meanings of the formula are determined. For example, it is determined which parameters in the formula are input values, how to obtain the output result through operation, and the actual meaning represented by the output result.

[0160] The variable symbols, parameter definitions, application scenarios and calculation logic are integrated to generate formula material analysis results, which explain in detail the use method and application range of the formula.

[0161] Step S1334: Perform consistency checking on the text material analysis results, chart material analysis results and formula material analysis results, check whether the expressions about the same demand content in the text material analysis results, chart material analysis results and formula material analysis results are consistent, if there is inconsistency, modify according to the material analysis results with higher dynamic semantic association strength information, and finally generate material analysis results.

[0162] The consistency of the text material analysis results, chart material analysis results and formula material analysis results is checked. Check whether the expressions about the same demand content in these results are consistent. For example, in the content related to building cost calculation, check whether the cost calculation methods and data in the text material analysis results, chart material analysis results and formula material analysis results are consistent.

[0163] If there is inconsistency, modify according to the material analysis results with higher dynamic semantic association strength information. If a certain text material has higher dynamic semantic association strength information with the demand node, modify according to the expression in the text material analysis results.

[0164] Step S134: Input the functional module unit theme name, functional module unit core demand expression, functional module unit trend adaptation description, material analysis results and dynamic semantic association strength information corresponding to the functional module unit into the cross-modal artificial intelligence content generation model. The cross-modal artificial intelligence content generation model determines the reference weight of each part of the material analysis results according to the dynamic semantic association strength information. The higher the dynamic semantic association strength information of the material analysis content, the greater the reference weight.

[0165] The functional module unit theme name, the functional module unit core requirement expression, the functional module unit trend adaptation description, the material analysis result, and the dynamic semantic correlation strength information corresponding to the functional module unit are input into the cross-modal artificial intelligence content generation model.

[0166] The cross-modal artificial intelligence content generation model determines the reference weight of each part of the material analysis result according to the dynamic semantic correlation strength information. The higher the dynamic semantic correlation strength information of the material analysis content, the greater the reference weight. For example, if the dynamic semantic correlation strength information of a certain text material analysis result and the core requirements and trend adaptation of the functional module unit is high, the reference weight of the text material analysis result in the content generation process will be larger.

[0167] Step S135: The cross-modal artificial intelligence content generation model fuses the material analysis result with the functional module unit core requirement expression and the functional module unit trend adaptation description according to the reference weight, generates a text description that meets the requirement logic for text content, generates a text description corresponding to the graph and a data interpretation corresponding to the graph for graph content, and generates an application scenario description of the formula and a calculation step explanation of the formula for formula content.

[0168] The cross-modal artificial intelligence content generation model fuses the material analysis result with the functional module unit core requirement expression and the functional module unit trend adaptation description according to the reference weight.

[0169] For text content, the model generates a text description that meets the requirements according to the requirement logic. For example, combining the construction scheme text material analysis result and the functional module unit core requirement expression, a detailed text description about the construction process and technical measures is generated.

[0170] For graph content, the model generates a text description corresponding to the graph and a data interpretation. For example, for the engineering progress plan graph, a text description about the engineering progress arrangement of each stage and the reason for the progress change is generated, as well as a detailed interpretation of the data of the graph.

[0171] For formula content, the model generates an application scenario description of the formula and a calculation step explanation. For example, for the cost calculation formula, it is explained that the formula is applicable to which project cost calculation, and the calculation steps and meanings of each parameter in the formula.

[0172] Step S136: The generated text description, the generated graph text description, and the generated formula application description are integrated to generate a multi-modal module content preliminary draft corresponding to the functional module unit, and each multi-modal module content preliminary draft includes a multi-modal module content preliminary draft text part, a multi-modal module content preliminary draft graph and corresponding description part, and a multi-modal module content preliminary draft formula and corresponding explanation part.

[0173] The generated text description, chart text explanation, and formula application explanation are integrated to generate a preliminary multi-modal module content corresponding to the functional module unit, which includes three parts: a preliminary multi-modal module content text part, which contains the text description of the functional module unit; a preliminary multi-modal module content chart and corresponding explanation part, which contains the chart and the text explanation and data interpretation of the chart; and a preliminary multi-modal module content formula and corresponding explanation part, which contains the formula and the application scenario description and calculation step explanation of the formula.

[0174] Step S140: Based on the dynamic semantic association strength information of the bidding scenario semantic graph and the dynamic connection rules between the functional module units, the semantic consistency adjustment and connection optimization processing between the functional module units are performed on the preliminary multi-modal module content of each functional module unit to obtain the integrated multi-modal electronic tender content.

[0175] Based on the dynamic semantic association strength information of the bidding scenario semantic graph and the dynamic connection rules between the functional module units, the preliminary multi-modal module content of each functional module unit is processed. First, semantic consistency adjustment is performed to ensure that the content of each functional module unit is consistent with the core demand expression and trend adaptation description in semantics; then, connection optimization processing between functional module units is performed to make the content transition between adjacent functional module units natural and logically coherent.

[0176] In the construction engineering bidding, through the above processing, the contents of each functional module unit are integrated to form a complete and coherent multi-modal electronic tender content.

[0177] For example, step S141: extract the multi-modal module content text part in the preliminary multi-modal module content of each functional module unit, the multi-modal module content chart explanation part in the preliminary multi-modal module content, and the multi-modal module content formula explanation part in the preliminary multi-modal module content, respectively, and perform semantic comparison with the functional module unit core demand expression corresponding to the functional module unit and the functional module unit trend adaptation description corresponding to the functional module unit, calculate the semantic similarity of the text expression and the demand expression, the logical matching degree of the chart explanation and the trend adaptation, and the adaptation degree of the formula explanation and the demand constraint.

[0178] The text part, chart explanation part, and formula explanation part are extracted from the preliminary multi-modal module content of each functional module unit. The above parts are respectively compared with the functional module unit core demand expression and the functional module unit trend adaptation description corresponding to the functional module unit.

[0179] The semantic similarity of the text expression and the requirement expression is calculated to measure the semantic similarity between the multi-modal module content draft text part and the core requirement expression. For example, check whether the text description of the construction scheme accurately reflects the requirements for construction technology and quality in the core requirement.

[0180] The logical matching degree of the chart explanation and the trend adaptation is calculated to evaluate whether the chart explanation part is logically consistent with the functional module unit trend adaptation explanation. For example, check whether the explanation of the engineering progress plan chart is consistent with the requirements of the industry technology development trend for engineering progress.

[0181] The adaptation degree of the formula explanation and the requirement constraint is calculated to determine whether the formula explanation part meets the constraint conditions in the core requirement. For example, check whether the explanation of the cost calculation formula meets the demand constraints of cost control.

[0182] Step S142: According to the dynamic semantic association strength information value corresponding to the functional module unit, dynamically set the qualified threshold of semantic similarity, the qualified threshold of logical matching degree, and the qualified threshold of adaptation degree; if the comparison result value of any one part of the content is lower than the corresponding qualified threshold, adjust the part of the content, the multi-modal module content draft text part is adjusted to supplement the key expression related to the demand, the multi-modal module content draft chart explanation part is adjusted to strengthen the conclusion related to the trend, and the multi-modal module content draft formula explanation part is adjusted to clarify the application conditions of the demand constraints, until the comparison result value of all parts of the content reaches the corresponding qualified threshold.

[0183] According to the dynamic semantic association strength information value corresponding to the functional module unit, dynamically set the qualified threshold of semantic similarity, the qualified threshold of logical matching degree, and the qualified threshold of adaptation degree. The higher the dynamic semantic association strength information value, the closer the association between the functional module unit and the bidding requirement, and the higher the corresponding qualified threshold.

[0184] If the comparison result value of any one part of the content is lower than the corresponding qualified threshold, adjust the part of the content. For the multi-modal module content draft text part, supplement the key expression related to the demand to make the text description more accurately reflect the core requirement. For example, supplement the key expression about the application of new construction technology in the construction scheme text.

[0185] For the multi-modal module content draft chart explanation part, strengthen the conclusion related to the trend to make the chart explanation better meet the requirements of the industry trend. For example, emphasize the positive impact of new technology application on engineering progress in the engineering progress plan chart explanation.

[0186] For the multi-modal module content draft formula explanation part, the application conditions of the demand constraints are clearly defined to ensure that the formula explanation meets the core demand constraints. For example, in the cost calculation formula explanation, the specific application conditions of cost control are clearly defined.

[0187] Adjustments are made continuously until the comparison results of all parts of the content reach the corresponding qualified threshold, ensuring that the content of each functional module unit is semantically consistent with the core demand and trend.

[0188] Step S143: According to the dynamic connection rules between functional module units, the multi-modal module content draft of adjacent functional module units is processed. If the adjacent functional module units are of the first correlation level, a summary paragraph of the core content of the previous functional module unit is added at the beginning of the text part of the multi-modal module content draft of the next functional module unit. The chart part of the multi-modal module content draft of the previous functional module unit and the chart part of the multi-modal module content draft of the next functional module unit maintain the same data dimension presentation. The formula part of the multi-modal module content draft of the previous functional module unit and the formula part of the multi-modal module content draft of the next functional module unit follow the same parameter definition logic.

[0189] According to the dynamic connection rules between functional module units, the multi-modal module content draft of adjacent functional module units is processed. If the adjacent functional module units are of the first correlation level, it means that they are closely related.

[0190] A summary paragraph of the core content of the previous functional module unit is added at the beginning of the text part of the multi-modal module content draft of the next functional module unit, making the content transition natural. For example, when transitioning from the "Building Intelligent Design" functional module to the "Intelligent Equipment Installation" functional module, the core content of "Building Intelligent Design" is summarized at the beginning of the text part of the "Intelligent Equipment Installation" module.

[0191] The chart part of the multi-modal module content draft of the previous functional module unit and the chart part of the multi-modal module content draft of the next functional module unit maintain the same data dimension presentation, making it easy for readers to compare and understand the data. For example, the previous chart shows the design parameters of the intelligent system, and the next chart shows the running parameters after the equipment is installed, maintaining the same data dimension.

[0192] The formula part of the multi-modal module content draft of the previous functional module unit and the formula part of the multi-modal module content draft of the next functional module unit follow the same parameter definition logic, ensuring the coherence of the formula. For example, in cost calculation, the formulas of the previous and next functional modules use the same parameter definition.

[0193] Step S144: If the adjacent functional module units are of the second correlation level, add transition text containing common node content between the two functional module unit multi-modal module content drafts, and the two functional module unit multi-modal module content draft chart parts present dimensions according to functional module unit theme adjustment data, but keep the data source description corresponding to the common node, and the two functional module unit multi-modal module content draft formula parts clearly mark the correlation parameters and difference parameters.

[0194] If the adjacent functional module units are of the second correlation level, the correlation degree is moderate, and appropriate connection processing is needed. Add transition text containing common node content between the two functional module unit multi-modal module content drafts. For example, when the adjacent functional module units are "building energy-saving design" and "green building material application" respectively, the common node may be "energy saving and environmental protection", and the transition text can explain the logical connection from energy-saving design concept to green building material application around this core, such as explaining how energy-saving design guides the selection and use of green building materials.

[0195] For the chart part of the two functional module unit multi-modal module content drafts, the data presentation dimensions are adjusted according to the functional module unit theme. For example, the chart of the "building energy-saving design" module may focus on displaying the simulation data of energy consumption and the comparison of the effects of energy-saving measures, while the chart of the "green building material application" module may focus on presenting the performance parameters and cost-effectiveness of different green building materials. However, the data source description corresponding to the common node should be kept, such as in the chart showing the influence of green building materials on energy-saving effect, the association with the energy consumption data in energy-saving design should be clearly marked, so that the reader can clearly see the data connection between the two modules.

[0196] In the formula part of the two functional module unit multi-modal module content drafts, the correlation parameters and difference parameters are clearly marked. For example, the formula for calculating energy consumption may be used in "building energy-saving design", while the formula for calculating material performance and cost may be used in "green building material application". If there are common parameters in the two formulas, such as building area, it is necessary to clearly mark whether the meaning and function of the parameter in different formulas are the same; for different parameters, their specific meanings in the respective formulas should also be clearly explained to avoid confusion.

[0197] Step S145: If the adjacent functional module units are of the third correlation level, add detailed transition chapters between the two functional module unit multi-modal module content drafts, the transition chapters contain demand transition logic explanation, modality conversion reason and information supplement content, the two functional module unit multi-modal module content draft chart parts redesign data presentation mode and attach correlation comparison explanation, the two functional module unit multi-modal module content draft formula parts complete parameter definition and logical correlation introduction.

[0198] When the adjacent functional module units are of the third correlation level, it indicates that the correlation between them is weak, and more detailed connection processing is needed. Detailed transition chapters are added between the preliminary drafts of the multi-modal module content of the two functional module units. The transition chapter first contains the transition logic explanation, for example, when the adjacent functional module units are "building appearance design" and "building fire protection system design", the transition chapter explains why the fire protection system design needs to be considered after the appearance design is completed, and explains the connection between the two in terms of the overall function and safety requirements of the building.

[0199] The transition chapter also explains the modal conversion reason, because the presentation of the multi-modal materials of the two functional modules may differ greatly. For example, "building appearance design" may mainly use visual modalities such as pictures and effect drawings, while "building fire protection system design" may rely more on text explanations and technical drawings. The transition chapter explains why the modalities are converted from visual to text and drawings, and the necessity of such conversion. At the same time, the transition chapter contains information supplement content to supplement the missing information connection between the two functional modules, such as introducing the impact of appearance design on the layout of fire access and the setting of fire facilities, etc.

[0200] For the preliminary draft chart section of the multi-modal module content of the two functional module units, the data presentation method is redesigned and associated comparison explanations are added. For example, the chart of "building appearance design" may be a three-dimensional effect drawing of the building, while the chart of "building fire protection system design" may be a fire pipe layout drawing. When redesigning the data presentation method, the two can be associated, such as marking the relationship between the door and window positions in the building appearance on the fire pipe layout drawing, and adding associated comparison explanations to enable the reader to understand the spatial and functional connection between the two.

[0201] In the formula section of the preliminary draft of the multi-modal module content of the two functional module units, the parameter definitions and logical correlations are introduced in detail. For example, there may be formulas for area calculation in "building appearance design", and formulas for fire water quantity calculation in "building fire protection system design". The parameter definitions in each formula are introduced in detail, including the meaning of the parameters, the value range, etc., and the logical correlation between the two formulas is explained, such as the influence of building area on fire water quantity calculation, etc.

[0202] Step S146: Repeat the semantic consistency adjustment step and the functional module unit connection optimization step until the preliminary drafts of the multi-modal module content of all functional module units meet the semantic requirements and the preliminary drafts of the multi-modal module content of adjacent functional module units are coherent, integrate all adjusted preliminary drafts of the multi-modal module content in the order of the electronic tender dynamic framework to obtain the integrated multi-modal electronic tender content.

[0203] The semantic consistency adjustment step and the inter-functional module unit connection optimization step are repeatedly performed. In terms of semantic consistency adjustment, the multi-modal module content draft of each functional module unit is continuously checked to ensure that the text expression is consistent with the requirement expression, the chart explanation is consistent with the trend, and the formula explanation is consistent with the requirement constraint in terms of semantics. If it is found that the content of a certain functional module unit is not semantically consistent, timely adjustment is made, such as supplementing key expressions, strengthening trend-related conclusions, and clearly defining requirement constraint application conditions.

[0204] In terms of inter-functional module unit connection optimization, according to the correlation level of adjacent functional module units, the corresponding connection rules are followed. The content transition between adjacent functional module units is continuously checked for naturalness and logical coherence. If it is found that the connection has problems, such as unclear transition text, unrelated chart data presentation, and unclear formula parameter definition, timely correction and improvement are made.

[0205] The above steps are continuously repeated until the multi-modal module content draft of all functional module units meets the semantic requirements, i.e., the content of each functional module unit accurately reflects the core requirements and trend adaptation; at the same time, the multi-modal module content draft of adjacent functional module units is coherent and smooth, and the reader can smoothly transition from one functional module to the next.

[0206] Finally, all the adjusted multi-modal module content drafts are integrated according to the order in the electronic tender dynamic framework. The electronic tender dynamic framework has clearly defined the order and connection method of each functional module unit. According to the electronic tender dynamic framework, the contents of each functional module unit are arranged and combined in sequence to form a complete and integrated multi-modal electronic tender content. This content contains rich text description, chart display, and formula explanation, and comprehensively covers all aspects of the bidding project.

[0207] Step S150: Obtain the presentation specification information of the electronic tender application scenario and the industry bidding preference data, combine the presentation specification information and the industry bidding preference data, and perform structure layout optimization and content adaptation adjustment processing on the integrated multi-modal electronic tender content to obtain the final multi-modal electronic tender.

[0208] Obtain the presentation specification information of the electronic tender application scenario, which specifies the requirements of the electronic tender in terms of format, font, layout, etc. For example, it may require the electronic tender to use a specific font and font size, the page layout to comply with certain standards, and the insertion position and size of charts and pictures to be clearly specified. At the same time, collect industry bidding preference data, which reflects the industry's preferences for the content focus, expression style, etc. of the bidding document. For example, some industries pay more attention to the innovation and feasibility of technical solutions, while others focus more on cost control and project schedule arrangement.

[0209] With the presentation specification information and industry bidding preference data, the integrated multi-modal electronic tender content is processed. First, structural layout optimization is performed, and the format and layout of the electronic tender are adjusted according to the presentation specification information. The text content is divided into appropriate paragraphs and chapters, and a uniform font, font size and line spacing are set to make the text layout neat and easy to read. For charts and pictures, they are inserted and adjusted according to the specified position and size to ensure that they cooperate with the text content and enhance the visual effect.

[0210] Then, content adaptation adjustment processing is performed, and the content of the electronic tender is optimized according to the industry bidding preference data. If the industry pays more attention to the innovation of technical solutions, the innovation points and advantages of the technical solution part in the electronic tender are highlighted; if the industry focuses on cost control, the cost budget and control measures are elaborated. Through the above adjustment, the content of the electronic tender is more in line with the industry demand and preference, which can improve the success rate of bidding.

[0211] After structural layout optimization and content adaptation adjustment processing, the final multi-modal electronic tender is obtained. This tender not only meets the presentation specification of the application scenario in format, but also meets the industry bidding preference in content, which can improve the success rate of bidding.

[0212] Based on the same inventive concept, please refer to Figure 2 , a structural schematic block diagram of an electronic tender generation system 100 based on artificial intelligence for executing the above-mentioned method of inspecting video stream is shown, which can include a communication unit 110, a machine-readable storage medium 120 and a processor 130.

[0213] In this embodiment, the machine-readable storage medium 120 and the processor 130 are located in the electronic tender generation system 100 based on artificial intelligence and are separately arranged. However, it should be understood that the machine-readable storage medium 120 can also be independent of the electronic tender generation system 100 based on artificial intelligence, and can be accessed by the processor 130 through a bus interface. Alternatively, the machine-readable storage medium 120 can also be integrated into the processor 130, and can communicate and interact with external systems through the communication unit 110.

[0214] The processor 130 is the control center of the artificial intelligence-based electronic tender generation system 100, connects various parts of the artificial intelligence-based electronic tender generation system 100 through various interfaces and lines, performs various functions of the artificial intelligence-based electronic tender generation system 100 and processes data by running or executing software programs and / or modules stored in the machine readable storage medium 120 and calling data stored in the machine readable storage medium 120, thereby monitoring the artificial intelligence-based electronic tender generation system 100 as a whole. Optionally, the processor 130 can include one or more processing cores; for example, the processor 130 can integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface, and application program, etc., and the modem processor mainly processes wireless communication. It can be understood that the above-mentioned modem processor can also not be integrated into the processor. Among them, the machine readable storage medium 120 is used to store machine executable instructions for executing the scheme of the present application, and the processor 130 is used to execute the machine executable instructions stored in the machine readable storage medium 120 to realize the method of the above-mentioned method embodiment.

[0215] It should be noted that, in order to simplify the expression of the present disclosure and to help the understanding of one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, various features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A method for generating electronic tender documents based on artificial intelligence, characterized in that, The method includes: The system acquires the tender requirements text released by the tendering party, the historical tender document material collection accumulated by the bidders, and the record of changes in tender requirements of similar projects in the industry. Based on the scenario description information of the tender requirements text, the multimodal material information of the historical tender document material collection, and the trend characteristics of the record of changes in tender requirements, a dynamically updated tender scenario semantic graph is constructed. The tender scenario semantic graph includes requirement nodes, multimodal material nodes, trend-related nodes, and information on the dynamic semantic association strength between nodes. Based on the hierarchical relationship of demand nodes and the evolution characteristics of trend-related nodes in the semantic graph of the bidding scenario, the overall structure of the electronic tender is adaptively decomposed to generate a dynamic framework of the electronic tender containing multiple functional module units and dynamic connection rules between functional module units. Each functional module unit corresponds to a set of related demand nodes, related multimodal material nodes and related trend-related nodes in the semantic graph of the bidding scenario. The pre-trained cross-modal AI content generation model is invoked, and the dynamic semantic association strength information of the electronic tender document dynamic framework, the semantic graph of the bidding scenario, and the multimodal material parsing rules are input. Cross-modal material association generation processing is performed on each functional module unit to obtain the first draft of the multimodal module content corresponding to each functional module unit. Based on the dynamic semantic association strength information of the semantic graph of the bidding scenario and the dynamic connection rules between functional module units, the initial draft of the multimodal module content of each functional module unit is subjected to semantic consistency adjustment and connection optimization processing between functional module units, resulting in the integrated multimodal electronic tender document content. Obtain presentation standard information and industry bidding preference data for electronic tender application scenarios. Combine the presentation standard information and industry bidding preference data to perform structural layout optimization and content adaptation adjustment on the integrated multimodal electronic tender content to obtain the final multimodal electronic tender.

2. The method for generating electronic tender documents based on artificial intelligence according to claim 1, characterized in that, The process involves acquiring the tender requirements text published by the tendering party, the historical tender document materials accumulated by the bidders, and the record of changes in tender requirements for similar projects in the industry. Based on the scenario description information in the tender requirements text, the multimodal material information in the historical tender document materials, and the trend characteristics of the record of changes in tender requirements, a dynamically updated semantic graph of the tendering scenario is constructed, including: Extract scenario description information from the tender requirement text, perform semantic segmentation on the scenario description information, divide the continuous text into multiple semantically independent requirement expression units, extract core requirement words and requirement constraints from each requirement expression unit, and generate a list of basic requirement information. Multimodal classification processing was performed on the historical bid document materials, dividing them into three categories: text materials, chart materials, and formula materials. The core theme information of each category of historical bid document materials was extracted. The core theme information of text materials is the key words and key paragraphs. The core theme information of chart materials is the chart title, the meaning of the coordinate axis and the description of the data trend. The core theme information of formula materials is the purpose of the formula and the definition of the parameters. Each type of historical tender material is treated as a multimodal material node, and each multimodal material node contains a material type identifier, core theme information, and material storage path; Analyze the bidding demand change records of similar projects in the industry, extract the time series information and demand change content from the bidding demand change records, perform trend feature identification on the demand change content, determine the key change direction and change frequency of demand in similar projects at different time stages, and take the key change direction of demand in each time stage as a trend association node. Each trend association node includes time stage identifier, demand change direction description and change frequency statistics. Calculate the semantic similarity value between the demand node and the multimodal material node, determine the initial semantic association strength information between the demand node and the multimodal material node based on the semantic similarity value, calculate the demand matching degree value between the demand node and the trend association node, and determine the initial semantic association strength information between the demand node and the trend association node based on the demand matching degree value; wherein, the value of the initial semantic association strength information is calculated by the corresponding semantic similarity value or demand matching degree value through a preset linear or nonlinear mapping function; A dynamic association update rule is constructed. Based on the latest trend characteristics of the bidding requirement change record, the initial semantic association strength information between requirement nodes, multimodal material nodes, and trend association nodes is adjusted periodically to generate dynamic semantic association strength information. All requirement nodes, all multimodal material nodes, all trend association nodes, and dynamic semantic association strength information are input into the graph construction tool to generate a bidding scenario semantic graph with nodes as the core and dynamic semantic association strength information as the connecting edges. The update frequency of the connecting edges in the bidding scenario semantic graph is consistent with the change frequency of the bidding requirement change record.

3. The method for generating electronic tender documents based on artificial intelligence according to claim 2, characterized in that, The process involves analyzing bidding demand change records for similar projects within the industry, extracting time-series information and demand change content from these records, performing trend feature identification on the demand change content, determining the key change direction and frequency of demand changes for similar projects at different time stages, and treating the key change direction of demand at each time stage as a trend correlation node, including: Text structure processing is performed on the bidding requirement change records of similar projects in the industry to remove redundant descriptions and duplicate information in the bidding requirement change records, and retain the core content including time stamps, requirement change items and reasons for change, to generate a structured requirement change record list; Extract time series information from the structured requirement change record list, divide the records in the structured requirement change record list into multiple consecutive time stages based on the time identifier, the duration of each time stage is determined according to the bidding cycle of similar projects, and assign a unique time stage identifier to each time stage. For each time period, summarize the content of all requirement change items within that time period, perform topic clustering on the content of the requirement change items, and group requirement change items that describe the same type of requirement change into a change topic. Each change topic corresponds to a type of requirement change direction. Count the number of times each change topic appears in the time period, calculate the proportion of the number of times it appears to the total number of change items in the time period, identify the change topic with the highest proportion as the key change direction of the requirement in the time period, and generate a description of the change direction of the requirement. Calculate the ratio of the total number of demand change items to the duration of each time period to obtain the demand change frequency for that time period. Integrate the time period identifier, the description of the direction of demand change, and the change frequency statistics to generate complete information for each trend-related node.

4. The method for generating electronic tender documents based on artificial intelligence according to claim 2, characterized in that, The aforementioned dynamic correlation update rules are constructed based on the latest trend characteristics of the bidding demand change records. Periodically adjust the initial semantic association strength information between demand nodes, multimodal material nodes, and trend association nodes to generate dynamic semantic association strength information, including: Set the associated update cycle, and the duration of the associated update cycle is determined based on the frequency of changes in the bidding requirements change record; Within each associated update cycle, extract the latest trend features from the bidding demand change records, determine the latest key demand change direction and the latest change frequency, and calculate the similarity value between the latest key demand change direction and the demand of each demand node. The initial semantic association strength information between the demand node and the trend association node is adjusted based on the calculated demand similarity value. When the demand similarity value is greater than the preset first threshold, the initial semantic association strength information value between the corresponding demand node and the trend association node increases. When the demand similarity value is less than the preset second threshold, the initial semantic association strength information value between the corresponding demand node and the trend association node decreases. Extract the multimodal material requirements under the latest trend characteristics, determine the multimodal material types and core themes that adapt to the latest requirements, and calculate the matching degree between the core theme information of each multimodal material node and the theme of the latest requirements. The initial semantic association strength information between the demand node and the multimodal material node is adjusted based on the calculated subject matching degree value. When the subject matching degree value is greater than the preset third threshold, the initial semantic association strength information value between the corresponding multimodal material node and the demand node increases. When the subject matching degree value is less than the preset fourth threshold, the initial semantic association strength information value between the corresponding multimodal material node and the demand node decreases. Summarize all adjusted initial semantic association strength information, generate dynamic semantic association strength information, record the time of each adjustment, the basis for each adjustment, and the changes in the initial semantic association strength information values ​​before and after the adjustment, and generate a dynamic association update log.

5. The method for generating electronic tender documents based on artificial intelligence according to claim 1, characterized in that, Based on the hierarchical relationship of demand nodes and the evolutionary characteristics of trend-related nodes in the semantic graph of the bidding scenario, the overall structure of the electronic tender document is adaptively decomposed to generate a dynamic framework for the electronic tender document containing multiple functional module units and dynamic connection rules between functional module units, including: Analyze the hierarchical relationship of demand nodes in the semantic graph of the bidding scenario, and use a semantic hierarchical clustering algorithm to divide demand nodes with subordinate relationships into different levels. The demand nodes at the highest level are core demand nodes, the demand nodes at the next lower level are secondary demand nodes, and so on, to generate a demand node hierarchical structure. By combining the evolution characteristics of trend-related nodes, the demand evolution direction of each core demand node at different time stages is determined. Based on the demand evolution direction, a corresponding trend-related node is matched for each core demand node to generate target association pairs. Based on each target association pair, select multimodal material nodes in the bidding scenario semantic graph whose dynamic semantic association strength information value with the core demand node and the trend association node is greater than or equal to the preset association strength threshold, and form an association node group by combining the core demand node, the matched trend association node and the selected multimodal material nodes. The required content, trend information, and multimodal material information corresponding to each associated node group are integrated to generate a functional module unit. Each functional module unit includes the functional module unit theme name, the core requirement description of the functional module unit, the trend adaptation description of the functional module unit, and the multimodal material index list of the functional module unit. The multimodal material index list of the functional module unit records the identifiers of all multimodal material nodes corresponding to the functional module unit and the storage paths of all multimodal material nodes. Analyze the semantic relationships between adjacent functional module units, and generate dynamic connection rules between functional module units based on the number of overlapping nodes in the associated node group and the dynamic semantic relationship strength information of the associated node group. The dynamic connection rules between functional module units include the expression logic of the connection content, the modal conversion method of the connection content, and the information supplementation requirements of the connection content. All functional modules are ranked and scored according to the importance of core requirement nodes. Combined with the dynamic connection rules between functional modules, a dynamic framework for electronic tender documents is generated.

6. The method for generating electronic tender documents based on artificial intelligence according to claim 5, characterized in that, The analysis of the hierarchical relationship of demand nodes in the semantic graph of the bidding scenario involves using a semantic hierarchical clustering algorithm to divide demand nodes with subordinate relationships into different levels. The demand nodes at the highest level are core demand nodes, the demand nodes at the next lower level are secondary demand nodes, and so on, generating a hierarchical structure of demand nodes, including: Extract the core requirement words and requirement constraints of all requirement nodes in the semantic graph of the bidding scenario, and concatenate the core requirement words and requirement constraints of each requirement node to form the requirement feature text. The pre-trained semantic encoding model is invoked to encode the requirement feature text of each requirement node, generating a semantic vector for each requirement node. The dimension of the semantic vectors of all requirement nodes is kept consistent. Calculate the cosine similarity between the semantic vectors of any two demand nodes, and construct a demand node association matrix based on the cosine similarity. The element values ​​in the demand node association matrix are the semantic similarity between the corresponding two demand nodes. Hierarchical clustering algorithm is used to perform clustering processing on demand nodes. Taking the demand node association matrix as input, demand nodes with semantic similarity higher than the clustering distance threshold are grouped into the same cluster, and each cluster corresponds to a demand level. For each cluster, the importance of the demand nodes is evaluated. The sum of the dynamic semantic association strength information between each demand node and other demand nodes outside the cluster is calculated. The demand node with the largest sum of dynamic semantic association strength information is the core node of the cluster, which is the core demand node of the corresponding demand level. The other demand nodes in the cluster are the secondary demand nodes of the corresponding demand level. Based on the hierarchical relationship of the clusters, the demand levels are arranged from high to low to generate a demand node hierarchy structure. The core demand nodes of each demand level, the secondary demand nodes of each demand level, and the subordinate relationships between the core demand nodes and the secondary demand nodes are recorded.

7. The method for generating electronic tender documents based on artificial intelligence according to claim 5, characterized in that, The analysis of semantic relationships between adjacent functional module units, based on the number of overlapping nodes in associated node groups and the dynamic semantic relationship strength information of associated node groups, generates dynamic connection rules between functional module units, including: Extract the associated node groups corresponding to two adjacent functional module units, count the number of overlapping demand nodes, overlapping trend associated nodes, and overlapping multimodal material nodes in the two associated node groups, calculate the ratio of the number of overlapping nodes to the total number of nodes in the two associated node groups, and use this ratio as the node overlap degree. Extract the dynamic semantic association strength information between overlapping nodes in two associated node groups, calculate the average value of the dynamic semantic association strength information, and use the average value as the mean association strength. Based on the node overlap value and the average association strength value, the association level of adjacent functional module units is determined. Specifically, adjacent functional module units with node overlap values ​​greater than or equal to a first overlap threshold and an average association strength value greater than or equal to a first strength threshold are classified as having a first association level; adjacent functional module units with node overlap values ​​less than the first overlap threshold but greater than or equal to a second overlap threshold, and an average association strength value less than the first strength threshold but greater than or equal to the second strength threshold, are classified as having a second association level; adjacent functional module units with node overlap values ​​less than the second overlap threshold and an average association strength value less than the second strength threshold are classified as having a third association level. Different association levels correspond to different connection strategies. For adjacent functional module units at the first level of association, the logic for generating connecting content is that the latter functional module unit, while inheriting the core content of the former functional module unit, supplements new requirements based on trend evolution; the modal transformation method for generating connecting content is to maintain the consistency of the main modal types of the two functional module units, and only supplement auxiliary modalities to adapt to new requirements; the information supplementation requirements for generating connecting content are to clearly mark the connection points and differences between the contents of the two functional module units. For adjacent functional module units at the second association level, the expression logic for generating connecting content is that the two functional module units describe the requirements of different dimensions respectively, and establish a connection through common nodes; the modal conversion method for generating connecting content is to adjust the modal type according to the theme of the functional module unit; the information supplementation requirement for generating connecting content is to mark the content association corresponding to the common nodes. For adjacent functional module units at the third association level, the logic for generating connecting content is to connect the two functional module units by adding transitional requirements; the modal conversion method for generating connecting content is to reselect the modal type based on the theme of the next functional module unit; the information supplementation requirements for generating connecting content are to provide a detailed introduction to the transitional content and the association logic between the two functional module units, integrate the connecting strategies of different association levels, and generate dynamic connecting rules between functional module units.

8. The method for generating electronic tender documents based on artificial intelligence according to claim 1, characterized in that, The process involves calling a pre-trained cross-modal AI content generation model, inputting the dynamic framework of the electronic tender document, the dynamic semantic association strength information of the bidding scenario semantic graph, and multimodal material parsing rules. Cross-modal material association generation processing is then performed on each functional module unit to obtain a preliminary draft of the multimodal module content for each functional module unit, including: For each functional module, extract the functional module topic name, core requirement description, trend adaptation description, and multimodal material index list from the electronic tender dynamic framework. Based on the material identifiers and storage paths in the multimodal material index list of functional module units, the corresponding multimodal materials are retrieved from the historical tender material collection. According to the material type identifier, the multimodal materials are divided into a text material subset, a chart material subset, and a formula material subset. Generate multimodal material parsing rules. Text material parsing rules extract key arguments and logical reasoning from the text. Chart material parsing rules extract data relationships and trend conclusions from the charts. Formula material parsing rules extract application scenarios and calculation logic from the formulas. Perform parsing processing on the text material subset, chart material subset, and formula material subset according to the multimodal material parsing rules to obtain the material parsing results. The cross-modal AI content generation model inputs the functional module unit theme name, core requirement description, trend adaptation description, material analysis results, and dynamic semantic association strength information corresponding to the functional module unit. Based on the dynamic semantic association strength information, the cross-modal AI content generation model determines the reference weight of each part of the material analysis results. The higher the dynamic semantic association strength information of the material analysis content, the greater its reference weight. The cross-modal AI content generation model integrates the material analysis results with the core requirements of functional modules and the trend adaptation descriptions of functional modules according to the citation weights. For text content, it generates text descriptions that conform to the logic of the requirements; for chart content, it generates text descriptions and data interpretations corresponding to the charts; and for formula content, it generates descriptions of the application scenarios of the formulas and explanations of the calculation steps of the formulas. The generated text descriptions, chart descriptions, and formula application instructions are integrated to generate the initial draft of the multimodal module content corresponding to the functional module unit. Each initial draft of the multimodal module content includes the text portion, the charts and corresponding descriptions, and the formulas and their corresponding explanations.

9. An artificial intelligence-based electronic tender generation system, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the AI-based electronic tender generation method according to any one of claims 1 to 8 by executing the machine-executable instructions.

10. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. A processor of a computer device reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the computer device to perform the artificial intelligence-based electronic tender generation method as described in any one of claims 1 to 8.

Images