Bidding document data processing method and system

Abstract

The invention discloses a bidding file data processing method and system, and the system employs a MoE mixed expert model architecture, and comprises a multi-type bidding file recognition module, a routing network, a professional sub-model library, a bidding file analysis engine module, and an automatic bidding file generation module. Wherein the bid invitation file is input into the multi-type bid invitation file identification module so as to identify the type of the bid invitation file; according to the type of the bid invitation file, calling a corresponding sub-model in a professional sub-model library; then the bid invitation file analysis engine module analyzes the bid invitation file; and the automatic bidding document generation module generates a bidding bidding document.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of artificial intelligence technology, and belongs to the field of natural language processing and intelligent document analysis. Specifically, it relates to a bidding document data processing method and system, and more particularly to a bidding document intelligent analysis and bid document automatic generation system and method based on a mixed expert model (MoE) architecture BACKGROUND

[0002] With the popularization of electronic bidding, the number and complexity of bidding documents are growing exponentially. The traditional processing method of bidding documents mainly relies on manual reading and sorting, which has the problems of low processing efficiency, limited professional ability, poor consistency, and compliance risks. The existing large model technology also has challenges in processing bidding documents: the general large model does not understand professional terms deeply, which is prone to "AI illusion", a single large model cannot simultaneously consider the special requirements of multiple bidding types, and the consumption of computing resources is large when processing large-scale bidding documents, and the reasoning speed is slow. SUMMARY

[0003] In order to solve the problem that the general large model does not understand professional terms in the prior art, which leads to the inaccuracy of the general large model in the field of bidding document processing, the present application provides a bidding document data processing method and system, which uses a mixed expert model (MoE) as an innovative form of large model architecture, and through the design idea of "specialization in a certain field", the task is allocated to multiple "experts" for solution.

[0004] In order to achieve the above-mentioned purpose, the embodiment of the present application provides a bidding document data processing system, which adopts a MoE mixed expert model architecture: a MoE architecture with sparse activation, including a multi-specialty sub-model library and a routing network; professional sub-models are deployed for different bidding types, and each sub-model focuses on a specific type of bidding document processing; the routing network dynamically selects the activated professional sub-model according to the characteristics of the input bidding document; wherein the bidding types include government procurement, engineering bidding, and service bidding;

[0005] Multi-type bidding document identification module: a lightweight large model is used to identify the type of the input bidding document, accurately distinguishing between different types such as government procurement, engineering bidding, and service bidding; intelligent classification is performed based on multi-dimensional features such as document title, keywords, and format characteristics; dynamic update of classification rules is supported to adapt to new types of bidding documents;

[0006] Routing network, dynamically selects the activated professional sub-model according to the characteristics of the input bidding document;

[0007] Professional sub-model library: supports dynamic expansion and updating of sub-models, and can manually add new professional sub-models; the professional sub-model library includes: government procurement sub-model, engineering bidding sub-model, and service bidding sub-model; among them, the government procurement sub-model focuses on parsing relevant provisions of the Government Procurement Law, with a built-in knowledge base of the Government Procurement Law and its implementing regulations; the engineering bidding sub-model focuses on processing engineering bidding documents such as bills of quantities, technical specifications, and construction requirements, incorporating knowledge from the field of engineering auditing; and the service bidding sub-model focuses on processing service bidding documents;

[0008] The bidding document parsing engine module is used to execute a multi-stage parsing process, including: document structure recognition → key information extraction → semantic understanding → structured output. It supports the combination of OCR recognition and natural language processing to process scanned documents and electronic documents, and accurately extracts technical parameters, commercial terms, and scoring criteria from bidding documents.

[0009] Automated bidding document generation module: Automatically generates bidding document framework based on parsing results; supports parallel generation of multiple versions to accommodate different bidding strategies; built-in compliance checks ensure that generated documents meet bidding requirements;

[0010] Among them, the bidding documents are input into the multi-type bidding document recognition module to identify the type of bidding documents; according to the type of bidding documents, the sub-model in the corresponding professional sub-model library is called; then the bidding documents are parsed by the bidding document parsing engine module; and the bidding documents are generated by the automated bidding document generation module.

[0011] Furthermore, the bidding document parsing engine module includes: a document structure recognition unit, a key information extraction unit, a semantic understanding unit and a structured output unit.

[0012] Furthermore, the system further comprises:

[0013] Knowledge update and optimization mechanism module: Continuously optimize the model by combining online learning and offline training; support user feedback mechanism, automatically use parsing errors after annotation for model optimization; regularly update the knowledge base of professional sub-models to ensure synchronization with the latest regulations and policies.

[0014] At the same time, an embodiment of the present invention proposes a method for processing bidding document data, including:

[0015] The bidding document is input into the multi-type bidding document recognition module to identify the type of bidding document; according to the type of bidding document, the sub-model in the corresponding professional sub-model library is called; then the bidding document is parsed by the bidding document parsing engine module; and the bidding document is generated by the automated bidding document generation module.

[0016] Further, the method further comprises: performing compliance checking on the generated file.

[0017] The beneficial effects of the above technical solutions of the present application are as follows: The scheme of the embodiments of the present application is particularly suitable for solving the problem of professional demand in the processing of bidding documents. The method includes the steps of bidding document type identification, professional sub-model routing activation, multi-stage parsing processing (document structure identification, key information extraction, semantic understanding, structured output), automatic bidding document generation, and compliance checking. The present application solves the technical problems of low processing efficiency, insufficient professionalism, consistency problems, and large consumption of computing resources in the prior art by deploying professional sub-models for different bidding types. The present application has the technical advantages of professional processing, high efficiency, scalability, consistency, and compliance guarantee, can significantly improve the efficiency and quality of bidding work, and reduce the compliance risk.

[0018] Compared with the prior art, the present application has the following significant advantages:

[0019] Professional processing: professional sub-models are deployed for different bidding types to ensure accurate understanding of professional terms and special requirements.

[0020] Efficiency: MoE architecture uses sparse activation, which has faster pre-training speed and inference speed than dense models; the routing mechanism ensures that computing resources are concentrated on relevant sub-models, improving computing efficiency.

[0021] Scalability: The sub-model library supports dynamic expansion and can easily adapt to the processing needs of new types of bidding documents.

[0022] Consistency: Bidding documents of the same type are processed by the same professional sub-model, ensuring consistency and accuracy of the processing results.

[0023] Compliance guarantee: built-in professional regulation knowledge base automatically checks the compliance of the generated file, reducing the compliance risk.

[0024] Lightweight: lightweight large models and MoE architecture are used to reduce the demand for computing resources and facilitate deployment. BRIEF DESCRIPTION OF DRAWINGS

[0025] The following drawings are used to provide a further understanding of the present application, and the illustrative examples of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0026] Figure 1 is a structural schematic diagram of a bidding document data processing system according to an embodiment of the present application;

[0027] Figure 2 is a flowchart of a bidding document data processing method according to an embodiment of the present application; DETAILED DESCRIPTION

[0028] It should be noted that the examples in the present application and the features in the examples can be combined with each other without conflict, unless specifically required, individual components and functions are optional, and the order of operation can be changed. Some parts and features of the embodiments can be included or replaced by parts and features of other embodiments. The scope of the embodiments of the present application includes the entire scope of the claims, and all available equivalents of the claims. The present application will be described in detail below with reference to the accompanying drawings and in conjunction with examples.

[0029] The embodiments of the present application aim to solve the technical problems of low processing efficiency, insufficient professionalism, consistency problems, and large consumption of computing resources in the prior art, and provide a tender document intelligent analysis and bid document automatic generation system and method based on MoE architecture, which can deploy professional sub-models for different tender types (government procurement, engineering tendering, etc.), and realize efficient, accurate, and compliant tender document analysis.

[0030] According to the embodiments of the present application, a tender and bidding document data processing system is provided, which is based on MoE architecture and can intelligently analyze tender documents and automatically generate bid documents. The tender and bidding document data processing system of the embodiments of the present application comprises:

[0031] 1. Multi-type tender document identification module

[0032] A lightweight MoE fine-tuning model (such as DeepSeek-R1 architecture) is used to realize multi-dimensional feature recognition, including:

[0033] Feature extraction layer: the input file generates a vector representation through an OCR multi-modal engine (integrating text, table, and image feature extraction). The Bi-LSTM+Attention mechanism is combined to capture the context association of the title, keywords, and paragraph structure, for example, to identify the specific clause expressions of the “Government Procurement Law” in the “government procurement” file;

[0034] Dynamic classifier: the routing network uses a noisy Top-K gating mechanism to calculate the matching degree of the feature vector and the expert sub-model:

[0035] G(x)=Sofmtax(TopK(2 g x+∈,k=2))

[0036] Wherein

[0037] x is the input vector, i.e. the data sample to be processed, which is usually a feature vector (for example: embedded representation of text);

[0038] W gFor trainable weights, the learnable parameter matrix of the gating network, shaped [hidden dimension, number of experts], is equivalent to a linear transformation;

[0039] ∈ is a random noise vector, usually sampled from a standard normal distribution, to avoid routing solidification;

[0040] TopK selects the k largest elements from the score vector, and the rest are set to -∞ (zero probability in Softmax);

[0041] Softmax is a normalization function that normalizes the score vector after TopK filtering (only non-∞ elements are calculated).

[0042] Support incremental learning: When a new type of bidding document is detected (such as "Carbon Neutral Special Procurement"), trigger the rule engine to update the classification tree and add a new expert sub-model branch.

[0043] 2, MoE hybrid expert model architecture, including:

[0044] Government procurement sub-model: Embedding the "Government Procurement Law Implementation Regulations" knowledge graph, using Conditional Random Fields (CRF) to parse structured fields such as "evaluation criteria" and "complaint provisions", and outputting compliance scores;

[0045] Engineering bidding sub-model: Integrating a Bill of Materials (BOM) parser, combining BERT-BiLSTM models to identify technical parameters and construction process requirements, and linking to a CAD drawing library to verify logical consistency;

[0046] Service bidding sub-model: Based on Service Level Agreement (SLA) template matching algorithm, extract "response time" and "KPI indicators" clauses, and generate deviation tables.

[0047] Routing network dynamic scheduling: Gating network outputs expert weight distribution, only activates Top-2 experts (such as government procurement + engineering bidding sub-model), and the rest of the expert parameters are not loaded into the video memory; Expert capacity factor (Capacity Factor) is set to 1.25, and overload tokens are transferred to the next layer through residual connection to avoid information loss;

[0048] 3, Bidding document parsing engine and automatic bid document generation module:

[0049] Used to perform a multi-stage parsing process:

[0050]

[0051] Bid document generation and compliance assurance; including: dynamic template filling and compliance checking;

[0052] Dynamic template filling: according to the weight of the evaluation standard, the enterprise knowledge base (qualification certificate, historical case) is called to automatically fill in the "technical scheme" and "implementation plan" chapters, and different strategy versions are adapted.

[0053] Compliance check: the integrated rule engine verifies the signature position, directory level, and tamper-proof key clauses of the blockchain storage.

[0054] 4. Knowledge updating and optimization mechanism module, including:

[0055] Dual-mode training strategy, including online learning and offline training;

[0056] Online learning: user annotation errors trigger incremental training, and the importance sampling (Importance Sampling) is used to update the expert weight, and the load balancing loss function is:

[0057] L aux = alpha * CV (Importance) 2 + beta * CV (Load) 2

[0058] where

[0059] CV is the coefficient of variation, the ratio of the standard deviation to the mean of the bid standard, which evaluates the balance of expert load or importance, and the smaller the value, the more balanced;

[0060] Importance is the importance of an expert, which is calculated according to the sum of the probability of being selected by a certain expert in a batch;

[0061] Load is the expert load, that is, the number of samples assigned to an expert in a batch (usually considering the case where the weight is higher than the threshold), which reflects the amount of calculation burden;

[0062] alpha and beta are balance hyperparameters used to adjust the relative importance between the two optimization objectives (Importance and Load).

[0063] Offline training: synchronize the regulatory library (such as the new procurement policy of the Ministry of Finance) every month, and use knowledge distillation to compress the policy expert sub-model.

[0064] According to the embodiment of the application, a kind of bidding document data processing method is provided, comprising:

[0065] The bidding document is input to the multi-type bidding document identification module to identify the type of the bidding document; according to the type of the bidding document, the sub-model in the corresponding professional sub-model library is called; then the bidding document is parsed by the bidding document parsing engine module; the bid document is generated by the automatic bidding document generation module.

[0066] Further, the method further comprises: performing compliance checking on the generated file.

[0067] The application is classified and explained below through specific embodiments.

[0068] Example 1 : Government procurement tender document processing

[0069] After the system receives the government procurement tender document, the type identification module determines that it is of the government procurement type.

[0070] The routing network activates the government procurement professional sub-model.

[0071] The sub-model performs in-depth analysis on sections such as "Bidder Qualification Requirements", "Procurement Needs", "Evaluation Criteria" in the tender document.

[0072] Key information such as "Supplier Qualification", "Procurement Budget", "Technical Parameters" is extracted.

[0073] An automatically generated tender document framework that meets the requirements of the "Government Procurement Law" is generated.

[0074] Compliance checking is performed on the generated file to ensure that all necessary clauses are included.

[0075] Example 2 : Engineering tender document processing

[0076] After the system receives the engineering tender document, the type identification module determines that it is of the engineering tender type.

[0077] The routing network activates the engineering tender professional sub-model.

[0078] The sub-model performs professional analysis on sections such as "Bill of Quantities", "Technical Specifications", "Construction Requirements" in the tender document.

[0079] Key information such as "Quantities", "Material Requirements", "Time Requirements" is extracted.

[0080] Intelligent interpretation of technical parameters is combined with engineering audit field knowledge.

[0081] An automatically generated tender document containing technical solutions, commercial offers, etc.

[0082] Example 3 : System expansion and optimization

[0083] When encountering new types of tender documents, system administrators can add new professional sub-models.

[0084] Through an online learning mechanism, sub-models can continuously learn from processed documents.

[0085] The user can feed back the analysis result, mark errors or make suggestions for improvement.

[0086] The system automatically updates the knowledge base of all sub-models periodically, ensuring synchronization with the latest regulatory policies.

[0087] In the above embodiments, all or part can be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part can be implemented in the form of a computer program product. The computer program product includes one or more computer programs. When the computer programs are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer programs can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer programs can be transferred from one website, computer, server or data center to another via wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available media sets. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as high-density digital video disc (digital video disc, DVD)), or semiconductor media (such as solid state disk (solid state disk, SSD)), etc.

[0088] Those of ordinary skill in the art can understand that the various numbers such as first, second, etc. involved in the present application are only for the convenience of description and do not limit the scope of the embodiments of the present application, nor indicate the order.

[0089] At least one of the present application can also be described as one or more, and the plurality can be two, three, four or more, which is not limited in the present application. In the embodiments of the present application, for a technical feature, the technical features in the technical feature are distinguished by "first", "second", "third", "A", "B", "C" and "D". There is no order or size order between the technical features described by "first", "second", "third", "A", "B", "C" and "D".

[0090] The correspondence relationship shown in each table in the present application can be configured or predefined. The values of the information in each table are merely examples, and other values can be configured, and the present application is not limited thereto. When configuring the correspondence relationship of the information and each parameter, it is not necessarily required to configure all the correspondence relationships shown in each table. For example, the correspondence relationship shown in some rows in the table in the present application can also not be configured. For another example, the above tables can be appropriately deformed, for example, split, merged, and the like. The names of the parameters shown in the titles of the above tables can also use other names understandable by the communication device, and the values or representation manners of the parameters can also use other values or representation manners understandable by the communication device. The above tables can also use other data structures when implemented, for example, an array, a queue, a container, a stack, a linear table, a pointer, a linked list, a tree, a graph, a structure, a class, a heap, a hash table, or the like.

[0091] The predefinition in the present application can be understood as defining, predefining, storing, pre-storing, pre-negotiating, pre-configuring, solidifying, or pre-burning.

[0092] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0093] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0094] The above is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A bidding document data processing system, characterized in that: The system adopts a MoE hybrid expert model architecture: it adopts a sparsely activated MoE architecture, including a multi-professional sub-model library and a routing network; it deploys specialized sub-models for different bidding types, with each sub-model focusing on processing a specific type of bidding document; The routing network dynamically selects an activated professional sub-model based on the characteristics of the input bidding document; wherein the bidding types include: government procurement, engineering bidding, and service bidding; Multi-type bidding document recognition module: This module uses a lightweight large model to identify the type of input bidding documents, accurately distinguishing between different types such as government procurement, project bidding, and service bidding; intelligently classifies documents based on multi-dimensional features such as document titles, keywords, and format characteristics; and supports dynamic updating of classification rules to adapt to new types of bidding documents. The routing network dynamically selects the activated specialized sub-model based on the characteristics of the input bidding documents; Professional sub-model library: supports dynamic expansion and updating of sub-models, and can manually add new professional sub-models; the professional sub-model library includes: government procurement sub-model, engineering bidding sub-model, and service bidding sub-model; among them, the government procurement sub-model focuses on parsing relevant provisions of the Government Procurement Law, with a built-in knowledge base of the Government Procurement Law and its implementing regulations; the engineering bidding sub-model focuses on processing engineering bidding documents such as bills of quantities, technical specifications, and construction requirements, incorporating knowledge from the field of engineering auditing; and the service bidding sub-model focuses on processing service bidding documents; The bidding document parsing engine module is used to execute a multi-stage parsing process, including: document structure recognition → key information extraction → semantic understanding → structured output. It supports the combination of OCR recognition and natural language processing to process scanned documents and electronic documents, and accurately extracts technical parameters, commercial terms, and scoring criteria from bidding documents. Automated bidding document generation module: Automatically generates bidding document framework based on parsing results; supports parallel generation of multiple versions to accommodate different bidding strategies; built-in compliance checks ensure that generated documents meet bidding requirements; Among them, the bidding documents are input into the multi-type bidding document recognition module to identify the type of bidding documents; according to the type of bidding documents, the sub-model in the corresponding professional sub-model library is called; then the bidding documents are parsed by the bidding document parsing engine module; and the bidding documents are generated by the automated bidding document generation module.

2. The bidding document data processing system according to claim 1, characterized in that: The bidding document parsing engine module includes: a document structure recognition unit, a key information extraction unit, a semantic understanding unit and a structured output unit.

3. The bidding document data processing system according to claim 1, characterized in that: The system further comprises: Knowledge update and optimization mechanism module: Continuously optimize the model by combining online learning and offline training; support user feedback mechanism, automatically use parsing errors after annotation for model optimization; regularly update the knowledge base of professional sub-models to ensure synchronization with the latest regulations and policies.

4. A method for processing bidding document data using the bidding document data processing system according to any one of claims 1 to 3, characterized in that: include: The bidding documents are input into a multi-type bidding document identification module to identify the type of the bidding documents; According to the type of bidding document, call the sub-model in the corresponding professional sub-model library; The bidding document parsing engine module then parses the bidding document; the automated bidding document generation module generates the bidding document.

5. The method for processing bidding document data by the bidding document data processing system according to claim 4, characterized in that: The method further includes: performing compliance check on the generated file.

Images