Bidding document dynamic template matching method and system based on natural language processing

Abstract

The invention relates to the technical field of business data management, in particular to a bidding document dynamic template matching method and system based on natural language processing. The method comprises the steps that multiple types of bidding document templates are collected, multi-dimensional features are extracted to form a multi-dimensional template feature system, and the initial weight of each dimension is determined; obtaining current bid invitation demand data to generate vectors, screening current bid invitation core features in combination with the multi-dimensional template feature system, and dynamically adjusting feature dimension weights; based on the current bid invitation core features and the adjusted weights, calculating similarities, and outputting a target template according to a similarity sequence; the system comprises a multi-dimensional template feature system construction module, a bidding demand feature extraction module and a bidding document template dynamic matching module. In this way, the bidding document can meet the requirements of different industries and different types of bidding projects, and therefore the bidding document manufacturing efficiency and quality are improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of business data management, in particular to a bid document dynamic template matching method and system based on natural language processing. BACKGROUND

[0002] In the bidding activity, bid document preparation is one of the core links, and selecting a suitable bid document template is the key to improving the efficiency and quality of bid document preparation. The traditional bid document template matching method mainly relies on manual screening. The staff needs to browse the templates in the template library one by one according to the bidding requirements, and compare the industry types, project requirements and the matching degree of the bidding requirements. This process not only consumes time and effort, but also is affected by the experience and subjective judgment of the staff, and the matching accuracy is poor.

[0003] With the development of the bidding industry, the types of bidding projects are increasingly diversified, involving engineering construction, goods procurement, service outsourcing and other fields. The bid document templates in different fields have significant differences in chapter structure, technical requirements and commercial terms. In the prior art, the template matching method mainly adopts keyword matching. This method can only realize simple text matching and cannot adapt to the requirements of different industries and different types of bidding projects, affecting the efficiency and quality of bid document preparation. SUMMARY

[0004] The purpose of the present application is to provide a bid document dynamic template matching method and system based on natural language processing, which aims to solve the technical problems in the prior art that only simple text matching can be realized, which cannot adapt to the requirements of different industries and different types of bidding projects, and affects the efficiency and quality of bid document preparation.

[0005] To achieve the above-mentioned purpose, the bid document dynamic template matching method based on natural language processing adopted by the present application comprises the following steps: Collecting multiple types of bid document templates, extracting multiple dimensions of features to form a multiple dimension template feature system, and determining the initial weight of each dimension; Generating a vector by obtaining current bidding requirement data, screening the current bidding core features in combination with the multiple dimension template feature system, and dynamically adjusting the feature dimension weight; Based on the current bidding core features and the adjusted weight, the similarity is calculated, and the target template is output according to the similarity.

[0006] Among them, in the step of collecting multiple types of bid document templates, extracting multiple dimensions of features to form a multiple dimension template feature system, and determining the initial weight of each dimension: Collecting multiple types of bid document templates to form a template basic library; Structural analysis is performed on each template in the template basic library; Based on the degree of influence of each dimension on template matching, the initial weights of each dimension are determined using the analytic hierarchy process (AHP).

[0007] Among the steps involved in performing structured parsing of each template in the template base library: Extract multi-dimensional features such as chapter structure, keywords, format, constraints, and industry attributes.

[0008] After the step of performing structured parsing on each template in the template base library: The extracted multi-dimensional features are integrated to construct a multi-dimensional template feature system that includes first-level dimensions and second-level sub-dimensions.

[0009] Among the steps, in obtaining the current bidding requirements data to generate a vector, and combining it with a multi-dimensional template feature system to filter the core features of the current bidding, and dynamically adjusting the feature dimension weights: Obtain the current bidding requirements document, remove special characters, page numbers, headers and footers, and use a stop word list to remove meaningless words to complete text cleaning; Semantic segmentation and part-of-speech tagging are performed on the cleaned text to generate word-level semantic vectors; Based on the constructed multi-dimensional template feature system, the correlation between semantic vectors is calculated.

[0010] In the step of calculating the correlation between semantic vectors based on the constructed multi-dimensional template feature system: Features are selected as the core features of the current tender based on their relevance, and the text proportion and frequency of each feature in the document are recorded.

[0011] In this process, after calculating the correlation between semantic vectors based on the constructed multi-dimensional template feature system: Feature attention scores are calculated based on the text proportion, frequency of occurrence, and industry importance of each core feature, and the weights of each feature dimension are dynamically adjusted based on the scores.

[0012] In the step of calculating similarity based on the current core features of the tender and the adjusted weights, and then sorting and outputting the target templates according to similarity: For each template in the template library, based on the current core features of the tender and the adjusted weights, the similarity of each dimension, including chapter structure, keywords, format, constraints, and industry attributes, is calculated. The total similarity is calculated by combining the adjusted weights of each dimension with the similarity of the corresponding dimension.

[0013] After calculating the total similarity by combining the adjusted weights of each dimension and the similarity of the corresponding dimensions: Sort the templates according to their total similarity and output the target template.

[0014] This invention also provides a dynamic tender document template matching system based on natural language processing, including a multi-dimensional template feature system construction module, a tender requirement feature extraction module, and a dynamic tender document template matching module; wherein: The multi-dimensional template feature system construction module is used to collect multiple types of tender templates, extract multi-dimensional features to form a multi-dimensional template feature system, and determine the initial weight of each dimension. The bidding requirement feature extraction module is used to obtain the current bidding requirement data generation vector, and combine it with a multi-dimensional template feature system to filter the current bidding core features, and dynamically adjust the feature dimension weights. The dynamic matching module for tender templates is used to calculate similarity based on the core features of the current tender and the adjusted weights, and output the target templates in order of similarity.

[0015] This invention discloses a dynamic template matching method and system for tender documents based on natural language processing. The method comprises a multi-dimensional template feature system construction module, a tender requirement feature extraction module, and a dynamic template matching module, which perform the following steps: collecting multiple types of tender document templates, extracting multi-dimensional features to form a multi-dimensional template feature system, and determining the initial weights of each dimension; obtaining the current tender requirement data generation vector, and combining it with the multi-dimensional template feature system to filter the current core tender features, dynamically adjusting the feature dimension weights; calculating the similarity based on the current core tender features and the adjusted weights, and outputting the target templates in order of similarity. This approach can adapt to the needs of different industries and types of tender projects, thereby improving the efficiency and quality of tender document preparation. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the steps of the dynamic template matching method for tender documents based on natural language processing of the present invention.

[0018] Figure 2 This is a flowchart of steps S100 of the present invention.

[0019] Figure 3 This is a flowchart of steps S200 of the present invention.

[0020] Figure 4 This is a flowchart of steps S300 of the present invention.

[0021] Figure 5 This is a schematic diagram of the structural principle of the dynamic template matching system for tenders based on natural language processing of the present invention.

[0022] Figure 6 This is a schematic diagram of the electronic device of the present invention.

[0023] 401 - Multi-dimensional template feature system construction module, 402 - Bidding requirement feature extraction module, 403 - Bidding document template dynamic matching module. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0025] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0026] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0027] Please see Figures 1-4 This invention provides a dynamic template matching method for tender documents based on natural language processing, comprising the following steps: S100: Collect various types of tender document templates, extract multi-dimensional features to form a multi-dimensional template feature system, and determine the initial weights of each dimension.

[0028] In this implementation, multiple types of tender document templates are collected, multi-dimensional features are extracted to form a multi-dimensional template feature system, and the initial weights of each dimension are determined. The specific process is as follows: S101: Collect various types of tender document templates to form a template library; S102: Perform structured parsing on each template in the template base library, extracting multi-dimensional features such as chapter structure, keywords, format, constraints, and industry attributes; S103: Integrate the extracted multi-dimensional features to construct a multi-dimensional template feature system that includes first-level dimensions and second-level sub-dimensions; S104: Based on the degree of influence of each dimension on template matching, the initial weights of each dimension are determined using the analytic hierarchy process (AHP).

[0029] In the above process, we extensively collected tender document templates from different industries and different types of tenders, such as open bidding, competitive negotiation, and single-source procurement. After deduplication and verification, we formed a template library covering multiple fields to ensure the comprehensiveness of template types.

[0030] Each template in the template base library is structured and parsed. The hierarchical relationship of chapters such as "Tender Letter", "Technical Solution" and "Business Quotation" is sorted out by document parsing tools. Core terms in each chapter (such as "concrete strength" and "service response time") are extracted and their frequency of occurrence is counted. The format requirements such as quotation form format and construction period unit are recorded. Required constraints such as "qualification certificate" and "project implementation plan" are marked. At the same time, industry-specific terms are extracted (such as "seismic resistance level" in the construction industry and "data security certification" in the IT industry).

[0031] The extracted features are divided into five primary dimensions: “chapter structure”, “keywords”, “format”, “constraints”, and “industry attributes”. Each primary dimension is further subdivided into secondary sub-dimensions (such as the keyword dimension, which includes “technical terms”, “business terms”, etc.), forming a hierarchical multi-dimensional template feature system.

[0032] The importance of each dimension in template matching is evaluated, and the analytic hierarchy process (AHP) is used to quantify the expert evaluation results into weight values. For example, the keyword dimension has a greater impact on matching, so its initial weight is set to 0.3, and the format dimension has its initial weight set to 0.15. Finally, the initial weight allocation of each dimension is determined.

[0033] S200: Obtain the current bidding requirements data to generate a vector, and combine it with a multi-dimensional template feature system to filter the core features of the current bidding, and dynamically adjust the feature dimension weights.

[0034] In this implementation, a vector of current bidding requirements data is generated, and the core features of the current bidding are selected using a multi-dimensional template feature system, with the feature dimension weights dynamically adjusted. The specific process is as follows: S201: Obtain the current tender requirements document, remove special characters, page numbers, headers and footers, and use a stop word list to remove meaningless words to complete text cleaning; S202: Perform semantic segmentation and part-of-speech tagging on the cleaned text to generate word-level semantic vectors; S203: Based on the constructed multi-dimensional template feature system, calculate the correlation degree between semantic vectors, screen features according to the correlation degree as the current core features of the tender, and record the text proportion and occurrence frequency of each feature in the document; S204: Calculate the feature attention scores according to the text proportion, occurrence frequency and industry importance of each core feature, and dynamically adjust the weights of each feature dimension according to the scores.

[0035] In the above process, obtain the tender requirement document uploaded by the user (supporting formats such as PDF and Word), use text processing tools to remove special symbols (such as #, @), page number identifiers (such as "Page X"), and header and footer content in the document, and then filter out meaningless words such as "of", "in", "le" through the stop word list, and retain the core text content related to the project.

[0036] Use a Chinese pre-trained model based on BERT to process the cleaned text, split the text into word units with semantic associations (such as "automatic biochemical analyzer" as a complete term), and at the same time mark the词性 (noun, verb, determiner, etc.) of each word, and convert these words into vectors that can represent semantic information (word-level semantic vectors).

[0037] Compare the generated semantic vectors with the multi-dimensional template feature system constructed in step S100, calculate the degree of tightness of the association between the vectors, and screen out the features with high correlation as the core features of the current tender requirements (such as "after-sales service life ≥ 3 years" in the medical equipment procurement requirements is determined as the core feature), and at the same time record the proportion of the text length and the number of occurrences of this feature in the document.

[0038] Integrate the text proportion of each core feature (such as a certain feature accounts for 15% of the total text), the occurrence frequency (such as a cumulative occurrence of 5 times), and the industry importance coefficient (such as the coefficient of "medical device business license" in the medical industry is 1.2), calculate the attention score of each feature, and the higher the score, the greater the impact of the feature on the matching. Accordingly, dynamically adjust the weight of the corresponding dimension (such as the weight of the dimension where the core feature is located is increased from 0.2 to 0.3).

[0039] S300: Calculate the similarity based on the current tender core features and the adjusted weights, and output the target template sorted by similarity.

[0040] In this embodiment, calculate the similarity based on the current tender core features and the adjusted weights, and output the target template sorted by similarity. The specific process is as follows: S301: For each template in the template library, calculate the similarity of each dimension of chapter structure, keywords, format, constraints, and industry attributes based on the current tender core features and the adjusted weights; S302: Calculate the total similarity by combining the adjusted weights of each dimension with the similarity of the corresponding dimension; S303: Sort the templates according to the total similarity and output the target template.

[0041] In the above process, for each template in the template library, based on the core bidding features extracted in step S200 and the adjusted weights, similarity is calculated in different dimensions: for chapter structure, the consistency between the suggested chapter order of bidding requirements and the chapter order of the template is compared; for keywords, the number of overlaps between the core bidding terms and the template terms is counted; for format and constraints, whether the amount unit, required fields, etc. meet the requirements is checked; for industry attributes, the matching degree of industry-specific terms is compared.

[0042] The similarity results of each dimension are multiplied by the adjusted weights of the corresponding dimensions, and then all the product results are summed to obtain the total similarity between the template and the current bidding requirements. The higher the total similarity value, the higher the matching degree between the template and the requirements.

[0043] All templates are sorted from highest to lowest total similarity, and the top 3 to 5 templates are output as target templates. If the highest total similarity is lower than a preset threshold (e.g., 0.65), the user is prompted to supplement the bidding requirements details (e.g., technical parameters, business terms), and the administrator is advised to update the template library to improve the subsequent matching effect.

[0044] In this invention, firstly, multiple types of tender document templates are collected, multi-dimensional features are extracted to form a multi-dimensional template feature system, and the initial weights of each dimension are determined; then, the current bidding requirement data is obtained to generate a vector, and the core features of the current bidding are screened in combination with the multi-dimensional template feature system, and the feature dimension weights are dynamically adjusted; finally, based on the current bidding core features and the adjusted weights, the similarity is calculated, and the target template is output in order of similarity. Through the above method, it is possible to adapt to the needs of different industries and different types of bidding projects, thereby improving the efficiency and quality of tender document preparation.

[0045] Corresponding to the aforementioned embodiments of the dynamic template matching method for tenders based on natural language processing, this application also provides embodiments of a dynamic template matching system for tenders based on natural language processing.

[0046] Figure 5 This is a block diagram of a dynamic template matching system for tender documents based on natural language processing, according to an exemplary embodiment. (Refer to...) Figure 5 The system may include: a multi-dimensional template feature system construction module 401, a bidding requirement feature extraction module 402, and a bid document template dynamic matching module 403; wherein: The multi-dimensional template feature system construction module 401 is used to collect multiple types of tender templates, extract multi-dimensional features to form a multi-dimensional template feature system, and determine the initial weight of each dimension. The bidding requirement feature extraction module 402 is used to obtain the current bidding requirement data generation vector, and combine it with a multi-dimensional template feature system to filter the current bidding core features, and dynamically adjust the feature dimension weights. The dynamic matching module 403 for bid templates is used to calculate similarity based on the current core features of the bid and the adjusted weights, and output the target templates in order of similarity.

[0047] In this embodiment, the multi-dimensional template feature system construction module 401 collects various types of tender document templates, extracts multi-dimensional features to form a multi-dimensional template feature system, and determines the initial weights of each dimension; the tender requirement feature extraction module 402 obtains the current tender requirement data generation vector, and combines it with the multi-dimensional template feature system to filter the current tender core features, and dynamically adjusts the feature dimension weights; the tender document template dynamic matching module 403 calculates the similarity based on the current tender core features and the adjusted weights, and outputs the target templates in order of similarity; through the above methods, it is possible to adapt to the needs of different industries and different types of tender projects, thereby improving the efficiency and quality of tender document preparation.

[0048] Regarding the system in the above embodiments, the specific ways in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0049] For the system embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0050] Accordingly, this application also provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the above-described dynamic template matching method for tender documents based on natural language processing. Figure 6 The diagram shown is a hardware structure diagram of any device with data processing capabilities, used in an embodiment of the present invention to provide a dynamic template matching system for tender documents based on natural language processing. (Except for...) Figure 6In addition to the processor, memory, and network interface shown, any data processing device in the embodiment may also include other hardware depending on the actual function of the data processing device, which will not be described in detail here.

[0051] Accordingly, this application also provides a computer-readable storage medium storing computer instructions thereon, which, when executed by a processor, implement the above-described dynamic template matching method for tenders based on natural language processing. The computer-readable storage medium can be an internal storage unit of any data-processing device as described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data-processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data-processing device, and can also be used to temporarily store data that has been output or will be output.

[0052] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0053] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A dynamic template matching method for tender documents based on natural language processing, characterized in that, Includes the following steps: Collect various types of tender document templates, extract multi-dimensional features to form a multi-dimensional template feature system, and determine the initial weights of each dimension; Obtain the current bidding requirements data to generate vectors, and combine them with a multi-dimensional template feature system to filter the core features of the current bidding, and dynamically adjust the feature dimension weights; Based on the core features of the current bidding process and the adjusted weights, the similarity is calculated, and the target templates are output in order of similarity.

2. The dynamic template matching method for tender documents based on natural language processing as described in claim 1, characterized in that, In the steps of collecting various types of tender document templates, extracting multi-dimensional features to form a multi-dimensional template feature system, and determining the initial weights of each dimension: Collect various types of tender document templates to form a basic template library; Perform structured parsing on each template in the template base library; Based on the degree of influence of each dimension on template matching, the initial weights of each dimension are determined using the analytic hierarchy process (AHP).

3. The dynamic template matching method for tender documents based on natural language processing as described in claim 2, characterized in that, In the process of performing structured parsing on each template in the template base library: Extract multi-dimensional features such as chapter structure, keywords, format, constraints, and industry attributes.

4. The dynamic template matching method for tender documents based on natural language processing as described in claim 3, characterized in that, After performing structured parsing on each template in the template base library: The extracted multi-dimensional features are integrated to construct a multi-dimensional template feature system that includes first-level dimensions and second-level sub-dimensions.

5. The dynamic template matching method for tender documents based on natural language processing as described in claim 1, characterized in that, In the steps of obtaining current bidding requirements data to generate vectors, combining multi-dimensional template feature systems to filter core features of the current bidding, and dynamically adjusting feature dimension weights: Obtain the current bidding requirements document, remove special characters, page numbers, headers and footers, and use a stop word list to remove meaningless words to complete text cleaning; Semantic segmentation and part-of-speech tagging are performed on the cleaned text to generate word-level semantic vectors; Based on the constructed multi-dimensional template feature system, the correlation between semantic vectors is calculated.

6. The dynamic template matching method for tender documents based on natural language processing as described in claim 5, characterized in that, In the step of calculating the correlation between semantic vectors based on the constructed multi-dimensional template feature system: Features are selected as the core features of the current tender based on their relevance, and the text proportion and frequency of each feature in the document are recorded.

7. The dynamic template matching method for tender documents based on natural language processing as described in claim 6, characterized in that, After calculating the correlation between semantic vectors based on the constructed multi-dimensional template feature system: Feature attention scores are calculated based on the text proportion, frequency of occurrence, and industry importance of each core feature, and the weights of each feature dimension are dynamically adjusted based on the scores.

8. The dynamic template matching method for tender documents based on natural language processing as described in claim 1, characterized in that, In the step of calculating similarity based on the core features of the current bidding process and the adjusted weights, and then sorting and outputting the target templates according to similarity: For each template in the template library, based on the current core features of the tender and the adjusted weights, the similarity of each dimension, including chapter structure, keywords, format, constraints, and industry attributes, is calculated. The total similarity is calculated by combining the adjusted weights of each dimension with the similarity of the corresponding dimension.

9. The dynamic template matching method for tender documents based on natural language processing as described in claim 8, characterized in that, After combining the adjusted weights for each dimension and the similarity of the corresponding dimensions, and calculating the total similarity: Sort the templates according to their total similarity and output the target template.

10. A dynamic template matching system for tender documents based on natural language processing, applied to the dynamic template matching method for tender documents based on natural language processing as described in claim 1, characterized in that, This includes a multi-dimensional template feature system construction module, a bidding requirement feature extraction module, and a bid document template dynamic matching module; among which: The multi-dimensional template feature system construction module is used to collect multiple types of tender templates, extract multi-dimensional features to form a multi-dimensional template feature system, and determine the initial weight of each dimension. The bidding requirement feature extraction module is used to obtain the current bidding requirement data generation vector, and combine it with a multi-dimensional template feature system to filter the current bidding core features, and dynamically adjust the feature dimension weights. The dynamic matching module for tender templates is used to calculate similarity based on the core features of the current tender and the adjusted weights, and output the target templates in order of similarity.

Images