A method for analyzing the consistency between scientific and technological innovation project results and goals

Abstract

The present invention proposes a method for analyzing the consistency between the results and goals of scientific and technological innovation projects. By integrating a large language model with a local knowledge base, the consistency between project results and project establishment goals is automatically evaluated. The method includes: importing the text of the project establishment document into the knowledge base to build a semantic database; reconstructing the topics of the project establishment document and project results text, extracting keywords and forming semantic chunks; retrieving the Top k text chunks most relevant to the project results in the semantic database; finally, inputting these text chunks together with the project results into the large language model to generate consistency evaluation results. This method optimizes long text matching, improves the efficiency and accuracy of scientific and technological innovation project audits, and supports offline deployment, ensures data security and privacy, and improves the transparency and trust of audits.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a method for analyzing the consistency between scientific and technological innovation project results and goals. Background Art

[0002] In the process of management and supervision of scientific and technological projects, especially in key links such as project establishment, development, and acceptance, the existing management system still has some problems that need to be solved urgently:

[0003] Subjective reliance: The project establishment and acceptance process relies too much on the subjective judgment of experts and lacks objective quantitative evaluation of the results. This evaluation method may be affected by the personal understanding and preference of experts, reducing the consistency and fairness of the review.

[0004] The tendency to prioritize quantity over quality in review: In the evaluation of project results, reviewers tend to prioritize the number of results (such as the number of published papers and patent applications), while neglecting the in-depth analysis of their quality and the consistency between the results and the initial goals of the project.

[0005] Lack of process supervision and system evaluation: The existing system lacks a mechanism for phased supervision and result evaluation during the project development process. Deviations from the initial goals during the implementation of the project cannot be discovered and corrected in a timely manner, affecting the overall benefits of the project.

[0006] The existing management system faces many challenges, especially in terms of technical implementation. It lacks tools that can effectively analyze and accurately match long texts. This is particularly evident when dealing with scientific research papers and project reports, making it difficult to achieve fine matching at the subject level. At the same time, the traditional project review process is difficult to strike a balance between efficiency and accuracy. The review process is not only time-consuming but also prone to errors. With the deepening of digital transformation, the security and privacy issues of project data have become increasingly prominent. The existing system is insufficient in protecting sensitive project information, bringing the risk of data leakage and abuse. In addition, the lack of explainability in the project review process makes it difficult for experts and decision makers to conduct an effective secondary review of the review results, affecting the transparency and trust of the review results. These issues together constitute challenges that need to be urgently addressed in the management of scientific and technological innovation projects.

[0007] It should be noted that the information disclosed in the above background technology section is only used for understanding the background of the present application, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the invention

[0008] The main purpose of the present invention is to overcome the defects existing in the above-mentioned background technology and provide a method for analyzing the consistency between the results and goals of scientific and technological innovation projects.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] A method for analyzing the consistency between scientific and technological innovation project results and goals, the method comprising the following steps:

[0011] S1. Import the text information of the Science and Technology Innovation Fund Project Approval Document into the large language model knowledge base;

[0012] S2. Reconstruct the topic of the imported project proposal text information. By extracting fine-grained keywords from the text, performing community detection and cluster analysis on the keywords, appending the sentences corresponding to the keywords to the clusters to form semantic chunks of highly relevant topics, and constructing a structured project proposal semantic database through vector representation.

[0013] S3, input the text of scientific and technological innovation project results into the large language model;

[0014] S4. Topic reconstruction of the scientific and technological innovation project results text, in which fine-grained keywords are extracted from the text, community detection and cluster analysis are performed on the keywords, and sentences corresponding to the keywords are attached to the clusters to form semantic blocks of highly relevant topics. After vectorization, the key themes and content blocks in the results text are extracted, thereby providing structured data for matching with the semantic information of the project proposal;

[0015] S5. In the project proposal semantic database, retrieve the top k text blocks that are most relevant to the scientific and technological innovation project results text, and provide them to the large language model to improve the matching accuracy;

[0016] S6. Input the matched Top k text blocks and the text of the scientific and technological innovation project results into the large language model for analysis, and the large language model generates an evaluation result on the consistency between the scientific and technological innovation project results and the project establishment document objectives.

[0017] Furthermore, the text information of the project proposal includes project objectives, references, work plan and expected results.

[0018] Furthermore, step S2 specifically includes:

[0019] Extracting fine-grained keywords of the text material;

[0020] Performing community detection and cluster analysis on the keywords, constructing a similarity matrix between keywords based on semantic similarity, and clustering the keywords using a clustering algorithm to form a topic cluster;

[0021] Attach the sentences corresponding to the keywords to the clusters to form a two-layer "small2big" structure;

[0022] Each sentence group is considered as a highly relevant topic and used as the corpus for constructing the “Chunk”;

[0023] Follow the subsequent steps of the original RAG to complete vectorization, generate semantic embedding, and build a semantic database of the project proposal for subsequent matching and retrieval.

[0024] S2.1, extract fine-grained keywords from text materials; let the text be , extract the keyword set

[0025] S2.2. Perform community detection and cluster analysis on keywords; construct a similarity matrix between keywords based on semantic similarity , use clustering algorithms (such as Louvain algorithm) to cluster keywords to form topic clusters ;

[0026] S2.3, attach the sentences corresponding to the keywords to the clusters in S2.2 to form a two-layer "small2big" structure; for each cluster , and its corresponding sentence set is represented as ;

[0027] S2.4. Each sentence group It can be regarded as a highly relevant topic and used as the corpus for constructing the semantic block "Chunk";

[0028] S2.5, follow the subsequent steps of the original RAG to complete the vectorization; The vector representation is , generate semantic embedding and build a semantic database of project proposal for subsequent matching and retrieval.

[0029] Furthermore, in step S2, an index structure is constructed and optimized during the segmentation of semantic blocks, and the segmentation size and content density of the semantic blocks are dynamically adjusted to reduce noise and cost while maintaining context integrity.

[0030] Furthermore, step S5 specifically includes:

[0031] S5.1. Calculate the semantic similarity between the project achievement text vector and each semantic block vector stored in the project proposal semantic database serving as an index library;

[0032] S5.2. Based on the calculated similarity, retrieve the top k semantic blocks that are most similar to the project outcome text from the project proposal semantic database;

[0033] S5.3. The retrieved Top k semantic blocks are attached to the output text as context information for subsequent consistency analysis by the large language model, thereby achieving efficient and accurate long text matching at the topic level.

[0034] Furthermore, it also includes integrating the framework for implementing prompt engineering with a large language model and a local knowledge base to support the long text matching requirements between project proposals and results in the project proposal.

[0035] Furthermore, it also includes integrating the project proposal information of the project establishment document into the offline knowledge base and building a persistent semantic index library for subsequent long text matching.

[0036] Furthermore, in step S6, the selected highly relevant text fragments are combined with the template-based query content to form a complete prompt word (Prompt), and are input into the large language model for consistency analysis to automatically generate an evaluation result of the consistency between the scientific and technological innovation project results and the project establishment objectives.

[0037] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method for analyzing the consistency between the results and goals of a scientific and technological innovation project.

[0038] A computer program product includes a computer program, which, when executed by a processor, implements the method for analyzing the consistency between scientific and technological innovation project results and goals.

[0039] The present invention has the following beneficial effects:

[0040] The present invention proposes a method for analyzing the consistency between the results and goals of scientific and technological innovation projects. By integrating advanced large language models and local knowledge bases, the present invention implements a topic-level long text matching mechanism, which divides long texts into highly relevant topic blocks, effectively maintains semantic integrity while reducing noise interference, and achieves more accurate topic consistency matching. In addition, the present invention optimizes the generate-retrieve-generate (RAG) process, and significantly improves the accuracy and computational efficiency of matching project proposals with results by building a semantic index library and retrieving the most relevant text blocks. The system supports offline deployment, ensures data security and privacy, and provides detailed matching context, making the audit results highly interpretable and convenient for experts to conduct secondary review. Compared with traditional methods, the present invention has obvious advantages in the accuracy of long text matching, processing efficiency and data security, and is particularly suitable for large-scale long text data processing, which significantly improves the audit efficiency and accuracy.

[0041] The method of the present invention designs an innovative evaluation system, which evaluates the correlation between project goals and results by matching and quantitatively analyzing the long text of the project's planned content and output. This system helps to provide a more scientific and objective evaluation basis, improve the evaluation efficiency and accuracy of project establishment, development, acceptance and other links, and support the optimal allocation of resources for scientific and technological innovation projects.

[0042] The present invention utilizes technologies such as data analysis, natural language processing, and intelligent matching to automatically analyze and evaluate the degree of fit between the project output (such as scientific research papers, research reports, etc.) and the project objectives, provide support for the review process of scientific and technological innovation projects, and improve review efficiency and accuracy.

[0043] The algorithm designed by the present invention achieves high-precision semantic matching by processing and analyzing long text content (such as full-text papers and project application materials). Through the natural language processing algorithm of the present invention, the system can automatically extract and compare key content, research topics, and technical points in long texts, thereby identifying the consistency and similarity between project results and project goals. This long text matching technology not only improves the accuracy of matching, but also can cope with the semantic complexity in long scientific research texts, effectively supporting the scientific review needs of scientific and technological innovation projects.

[0044] Other beneficial effects of the embodiments of the present invention will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The present invention provides an overall flow chart of a method for analyzing the consistency between scientific and technological innovation project results and objectives according to an embodiment of the present invention.

[0046] Figure 2 It is the RAG workflow involved in the embodiment of the present invention and the focus of the present invention.

[0047] Figure 3 A flowchart of exemplary long text reconstruction according to an embodiment of the present invention.

[0048] Figure 4 A schematic diagram of the framework of a method for analyzing the consistency between scientific and technological innovation project results and goals according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following is a detailed description of the embodiments of the present invention. It should be emphasized that the following description is only exemplary and is not intended to limit the scope and application of the present invention.

[0050] See also Figures 1 to 4 The embodiment of the present invention provides a method for analyzing the consistency between scientific and technological innovation project results and goals, and the method comprises the following steps:

[0051] Step S1: Import the text information of the science and technology innovation fund project proposal into the large language model knowledge base; the text information of the project proposal may include project goals, references, work plans and expected results, etc.

[0052] Step S2: reconstruct the topic of the imported project proposal text information, by extracting fine-grained keywords from the text, performing community detection and cluster analysis on the keywords, and attaching the sentences corresponding to the keywords to the clusters to form semantic chunks of highly relevant topics. After vectorization, a structured project proposal semantic database is constructed.

[0053] In a preferred embodiment, step S2 specifically includes:

[0054] Extracting fine-grained keywords of the text material;

[0055] Performing community detection and cluster analysis on the keywords, constructing a similarity matrix between keywords based on semantic similarity, and clustering the keywords using a clustering algorithm to form a topic cluster;

[0056] Attach the sentences corresponding to the keywords to the clusters to form a two-layer "small2big" structure;

[0057] Each sentence group is considered as a highly relevant topic and used as the corpus for building a semantic chunk “Chunk”;

[0058] Follow the subsequent steps of the original RAG to complete vectorization, generate semantic embedding, and build a semantic database of the project proposal for subsequent matching and retrieval.

[0059] Step S3: input the text of the scientific and technological innovation project results into the large language model.

[0060] Step S4: reconstruct the topic of the scientific and technological innovation project results text, wherein fine-grained keywords in the text are extracted, community detection and cluster analysis are performed on the keywords, sentences corresponding to the keywords are attached to the clusters to form semantic blocks of highly relevant topics, and vectorized representation is performed to extract key themes and content blocks in the results text, thereby providing structured data for matching with the semantic information of the project establishment document;

[0061] Step S5: retrieve the top k text blocks most relevant to the scientific and technological innovation project results text in the project proposal semantic database, and provide them to the large language model to improve matching accuracy;

[0062] In a preferred embodiment, step S5 specifically includes:

[0063] Calculate the semantic similarity between the project achievement text vector and each semantic block vector stored in the project proposal semantic database as an index library;

[0064] Based on the calculated similarity, retrieve the top k semantic blocks that are most similar to the project outcome text from the project proposal semantic database;

[0065] The retrieved Top k semantic blocks are attached to the result text as context information for subsequent consistency analysis by the large language model, thereby achieving efficient and accurate long text matching at the topic level.

[0066] Step S6: input the matched Top k text blocks and the scientific and technological innovation project results text into the large language model for analysis, and the large language model generates an evaluation result on the consistency between the scientific and technological innovation project results and the project establishment document objectives.

[0067] The core innovation of the evaluation method of the present invention lies in the topic-level long text matching mechanism, which divides long text into highly relevant topic blocks by constructing semantic chunks. This not only maintains the integrity of semantics, but also reduces noise interference, and achieves more accurate topic consistency matching. In addition, the system optimizes the generate-retrieve-generate (RAG) process, combines the large language model, builds a semantic index library and retrieves the most relevant text chunks, effectively improving the accuracy and computational efficiency of matching project proposals with results. The present invention effectively solves the pain point problems of relying on subjective judgment and difficulty in long text matching in the existing project review process, provides a systematic objective standard, and reduces the deviation and lack of consistency of the review results. It can also accurately identify complex semantic associations in different texts, especially in the precise topic matching between project proposals and scientific research results, which is difficult to achieve with traditional methods. Compared with the prior art, the present invention shows obvious advantages in topic-level long text matching, processing efficiency and data security. It constructs highly relevant topic chunks for precise topic matching and uses vectorized index libraries to optimize retrieval speed, which is particularly suitable for large-scale long text data processing. It supports offline deployment, meets the needs of scenarios with strict requirements on data security, and provides high explainability, which is convenient for expert secondary review, significantly improving the transparency and trust of the review. In the performance test (which will be further described later), the present invention simulated the paper review environment based on two data sets and compared it with four advanced methods, showing advantages in accuracy and macro-average F1 value. These test results further prove the accuracy of the present invention in long text matching ability and cross-domain content recognition.

[0068] The specific embodiments and experimental verification of the present invention are further described below.

[0069] One of the core challenges solved by the present invention is to achieve a topic-level match between scientific research project results and project proposals, rather than a global match of the entire article. In the project review process, the left side is the project proposal, and the right side is the project results (such as papers or patents, etc.). Each project result often only discusses a specific sub-aspect of the project proposal or application. If traditional natural language processing (NLP) technology is used directly to compare the overall semantic similarity of project results and project proposals, it usually results in a low similarity and cannot effectively identify the topic consistency between the project results and the project objectives.

[0070] To solve this problem, the present invention introduces a Generate-Retrieve-Generate (RAG) technology based on a large language model. Figure 2 The RAG workflow and the focus of the present invention are shown. Specifically, in the indexing stage of RAG, documents will be processed, segmented, and converted into embeddings and stored in a vector database. Building a high-quality index is the key to achieving efficient retrieval, which directly determines whether the correct context can be obtained and accurate matching can be achieved in the retrieval stage.

[0071] The present invention designs a method for constructing semantic blocks "Chunk" to optimize indexing. The traditional block segmentation method is to divide the document into blocks according to a fixed number of tags (such as 100, 256, 512). Larger blocks can contain more contextual information, which helps to capture more comprehensive semantics, but at the same time it will also introduce more noise, increase processing time and cost. Although smaller blocks have less noise and faster processing speed, they often cannot fully convey the necessary contextual content, affecting the accuracy of matching. In this regard, the present invention is designed to optimize the index structure during the block segmentation process to achieve efficient and accurate matching of long texts at the topic level. Furthermore, by dynamically adjusting the block size and content density, the system can reduce noise and cost while maintaining context integrity, thereby providing accurate context support for subsequent retrieval and evaluation.

[0072] Specifically, the method for analyzing the consistency between the scientific and technological innovation project results and goals of the present invention includes the following steps, among which a Chunk construction algorithm of the RAG process is specially designed.

[0073] Step S1: Import the text information of the science and technology innovation fund project proposal (such as project objectives, references, work plans and expected results, etc.) into the large language model knowledge base.

[0074] Step S2: Reconstruct the topic of the original project proposal text information:

[0075] Step S2.1: Extract fine-grained keywords from text materials. Suppose the text is , extract the keyword set

[0076] Step S2.2: Perform community detection and cluster analysis on keywords. Construct similarity matrix between keywords based on semantic similarity , use clustering algorithms (such as Louvain algorithm) to cluster keywords to form topic clusters ;

[0077] Step S2.3: Add the sentences corresponding to the keywords to the clusters in step S2.2 to form a two-layer "small2big" structure. , and its corresponding sentence set is represented as ;

[0078] Step S2.4: Each sentence group It can be regarded as a highly relevant topic and used as the corpus for constructing "Chunk";

[0079] Step S2.5: Follow the steps of the original RAG to complete the vectorization. The vector representation is , generate semantic embedding and build a semantic database for subsequent matching retrieval;

[0080] Step S3: Use the annual scientific and technological innovation project results (such as papers and patents) as input to the big model.

[0081] Step S4: Reconstruct the topic of the scientific and technological innovation project results text similar to step S2, and extract the key themes and content blocks of the results.

[0082] Step S5: In the project proposal semantic database, retrieve the top k text blocks that are most relevant to the achievement text and provide them to the large language model to improve the matching accuracy. Suppose the semantic vector of the achievement text to be analyzed is .

[0083] Step S6: The matched Top k text blocks and the scientific and technological innovation achievement text are input into the big model for analysis, and the big model generates a judgment on the consistency between the achievement and the project establishment document objectives.

[0084] This method combines the Generate-Retrieve-Generate (RAG) process of the large language model, and well implements the subject consistency assessment between the results of scientific and technological innovation projects (such as papers and patents) and the project objectives. First, the text information of the project proposal (such as project objectives, references, work plans, and expected results) is imported into the large model knowledge base. These texts are processed into topic chunks that are easy to match, forming a database with a semantic hierarchy.

[0085] During data preprocessing and topic reconstruction, fine-grained keywords in the project proposal text are extracted and grouped to form keyword clusters. Subsequently, sentences related to each keyword group are attached to the corresponding cluster to construct a two-layer "small2big" structure, that is, a semantic block consisting of keyword clusters and their corresponding sentences. Each block is defined as a chunk for subsequent vectorization and matching.

[0086] In the indexing phase of the RAG process, each chunk is converted into a semantic embedding vector and stored in a vector database. This structured index library ensures that the most relevant semantic blocks can be quickly found in the retrieval phase, thereby optimizing matching accuracy. Next, the text of annual project results (such as papers and patents) is imported into the system, and after similar reconstruction steps, its core topics are extracted and vector representations are generated. In the retrieval phase, the text vector of the project results is matched with the index database of the project proposal for semantic similarity, and the most similar Top k chunks are retrieved and attached to the text of the results as context.

[0087] Finally, the large language model receives the relevant Top k chunks and results text and generates comparative analysis text. With this contextual information, the large model can evaluate the consistency of project results with the goals of the project proposal, thereby providing accurate, topic-level evaluation. This process optimizes chunk construction and indexing in topic matching, and improves the accuracy and efficiency of judging the consistency between project results and initial goals.

[0088] An example is described below.

[0089] In the example, the present invention utilizes the Langchain framework and the ChatGLM model (especially ChatGLM-6B) to integrate with the local knowledge base to support the long text matching requirements between project proposals and outputs (such as papers). Figure 3 The diagram for implementing long text reconstruction is shown. The ChatGLM model is very suitable for this task because of its powerful language understanding ability, support for Chinese and English processing (adapting to the needs that project proposals are usually in Chinese and papers are mostly in English), and offline deployment. The Langchain framework is used to implement Prompt Engineering, thereby optimizing the model's understanding and generation of different content.

[0090] In the specific workflow, the present invention first integrates the information of the project proposal into the offline knowledge base of ChatGLM to form a persistent semantic index library. Research results (such as papers or patents) are used as query inputs to match the proposal content in the knowledge base. This process is based on vectorized text retrieval technology to embed the proposal text and store it in a vector database. Whenever a query (such as project results) is entered into the system, the query text will be vectorized and the similarity will be calculated with the text vector in the knowledge base to find the top k most similar text fragments in the vector space.

[0091] These selected highly relevant text fragments are combined with the template-based query content to form a complete prompt and fed back to ChatGLM. Through this design, the present invention can accurately match the subject content of project proposals and results, and automatically generate consistency analysis, ultimately helping to more efficiently evaluate the consistency of project results with project goals.

[0092] The performance test is as follows:

[0093] Based on two data sets, the present invention simulated the paper review environment, deployed the above method, and compared it with four advanced methods.

[0094] In the process of data set construction, the present invention uses PassageRetrieval-zh and PubMed data sets. PassageRetrieval-zh is a paragraph-level retrieval data set designed specifically for Chinese, containing a large number of Chinese short paragraphs and query pairs, which is suitable for training models in paragraph matching and information retrieval capabilities in the Chinese context. The PubMed data set consists of global biomedical and life science literature, covering more than 30 million medical paper abstracts and some full texts. It is one of the most commonly used English literature databases in the medical field and is suitable for text mining, information retrieval and semantic understanding tasks.

[0095] Based on these original data sets, the present invention randomly selected 3 articles in a specific field and used GPT-3.5-turbo to summarize them. Then, another 3 articles in the field were mixed with 7 articles in other fields to form a candidate set. The model to be evaluated needs to identify the article most relevant to the summary among these candidate articles, that is, the target article that matches the summary content. Such a construction method can not only test the model's long text matching ability, but also evaluate its accuracy in cross-domain content recognition.

[0096] The present invention uses Accuracy and macro-F1 as classification evaluation indicators, both of which can provide valuable performance evaluation information in different scenarios. Accuracy is one of the most commonly used evaluation indicators in classification tasks, indicating the proportion of samples correctly predicted by the model to the total samples. Its calculation is simple and clear, and can intuitively reflect the overall performance of the model on the entire data set. It is suitable for situations where the data set categories are evenly distributed, or when the present invention focuses on the overall performance of the model. In addition, the accuracy calculation is fast and takes up less resources, so it is very practical in real-time or large-scale applications.

[0097] Macro-F1 is a more balanced indicator, which is particularly useful when dealing with data sets with uneven category distribution. It can measure the average performance of the model in each category. The results are shown in Table 1. The method proposed in this paper outperforms other methods that require training in terms of accuracy and macro-average F1 value.

[0098] Table 1

[0099]

[0100] In addition to the advantages in key indicators, the method of the present invention has a series of unique advantages that meet the needs in the actual application of scientific and technological innovation project review:

[0101] Offline deployment ensures security and privacy: In the review of scientific and technological innovation projects, data often involves scientific research results, project planning, and high-value information, so the security and privacy of data are crucial. The method of the present invention supports the local deployment of models, and can complete the review work without relying on external servers or the cloud. This offline deployment mode ensures the security of data and avoids the risk of leakage of sensitive information. Offline deployment also allows the model to be customized according to the specific needs of the scientific and technological innovation project review, which improves the flexibility of use.

[0102] High interpretability, convenient for secondary review by experts: In the review of scientific and technological innovation projects, the transparency of the review process and the interpretability of the results are crucial for the trust and secondary review of experts. The method of the present invention outputs relevant topic blocks and semantic information during model prediction, and experts can clearly see the content that the model refers to during the text matching process. In this way, experts can make secondary judgments based on the analysis results of the model, especially for complex long text matching tasks, and can directly view the text paragraphs and keywords related to the matching results, further improving the accuracy of the review.

[0103] Improve audit efficiency and reduce expert workload: The review of scientific and technological innovation projects usually requires reviewing a large number of project proposals and research results, involving lengthy text content. The method of the present invention extracts the core content of project proposals and results through automated long text matching and performs subject-level analysis, making the review process more efficient. While maintaining high accuracy, it reduces the need for experts to manually compare the full text, greatly reducing the review time.

[0104] An embodiment of the present invention further provides a storage medium for storing a computer program, which at least performs the above method when executed.

[0105] An embodiment of the present invention further provides a control device, comprising a processor and a storage medium for storing a computer program; wherein the processor is configured to execute at least the method described above when executing the computer program.

[0106] An embodiment of the present invention further provides a processor, wherein the processor executes a computer program and at least executes the method described above.

[0107] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a ferromagnetic random access memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The storage medium described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0108] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0109] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0110] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0111] A person skilled in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, etc. Various media that can store program codes.

[0112] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0113] The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0114] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0115] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0116] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art of the present invention, several equivalent substitutions or obvious variations can be made without departing from the concept of the present invention, and the performance or use is the same, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for analyzing the consistency between scientific and technological innovation project results and goals, characterized in that: The method comprises the following steps: S1. Import the text information of the Science and Technology Innovation Fund Project Approval Document into the large language model knowledge base; S2. Topic reconstruction is performed on the imported project proposal text information. Fine-grained keywords are extracted from the text, community detection and cluster analysis are performed on the keywords, and sentences corresponding to the keywords are attached to the clusters to form semantic blocks of highly relevant topics. After vectorization, a structured project proposal semantic database is constructed. S3, input the text of scientific and technological innovation project results into the large language model; S4. Topic reconstruction of the scientific and technological innovation project results text, in which fine-grained keywords are extracted from the text, community detection and cluster analysis are performed on the keywords, and sentences corresponding to the keywords are attached to the clusters to form semantic blocks of highly relevant topics. After vectorization, the key themes and content blocks in the results text are extracted, thereby providing structured data for matching with the semantic information of the project proposal; S5. In the project proposal semantic database, retrieve the top k text blocks that are most relevant to the scientific and technological innovation project results text, and provide them to the large language model to improve the matching accuracy; S6. Input the matched Top k text blocks and the text of the scientific and technological innovation project results into the large language model for analysis, and the large language model generates an evaluation result on the consistency between the scientific and technological innovation project results and the project establishment document objectives.

2. The method for analyzing the consistency between scientific and technological innovation project results and goals as claimed in claim 1, characterized in that: The text information of the project proposal includes project objectives, references, work plan and expected results.

3. The method for analyzing the consistency between scientific and technological innovation project results and goals as claimed in claim 1, characterized in that: Step S2 specifically includes: Extracting fine-grained keywords of the text material; Performing community detection and cluster analysis on the keywords, constructing a similarity matrix between keywords based on semantic similarity, and clustering the keywords using a clustering algorithm to form a topic cluster; Attach the sentences corresponding to the keywords to the clusters to form a two-layer "small2big" structure; Each sentence group is considered as a highly relevant topic and used as corpus to build a semantic block; According to the subsequent steps of original generation-retrieval-generation of RAG, vectorization is completed, semantic embedding is generated, and a semantic database of the project proposal is built for subsequent matching and retrieval.

4. The method for analyzing the consistency between scientific and technological innovation project results and goals according to any one of claims 1 to 3, characterized in that: In step S2, an index structure is constructed and optimized during the segmentation of semantic blocks, and the segmentation size and content density of the semantic blocks are dynamically adjusted to reduce noise and cost while maintaining context integrity.

5. The method for analyzing the consistency between scientific and technological innovation project results and goals according to any one of claims 1 to 3, characterized in that: Step S5 specifically includes: S5.

1. Calculate the semantic similarity between the project achievement text vector and each semantic block vector stored in the project proposal semantic database as an index library; S5.

2. Based on the calculated similarity, retrieve the top k semantic blocks that are most similar to the project outcome text from the project proposal semantic database; S5.

3. The retrieved Top k semantic blocks are attached to the output text as context information for subsequent consistency analysis by the large language model, thereby achieving efficient and accurate long text matching at the topic level.

6. The method for analyzing the consistency between scientific and technological innovation project results and goals according to any one of claims 1 to 3, characterized in that: It also includes integrating the framework for implementing prompt engineering with large language models and local knowledge bases to support the long text matching requirements between project proposals and outcomes in project proposals.

7. The method for analyzing the consistency between scientific and technological innovation project results and goals according to any one of claims 1 to 3, characterized in that: It also includes integrating the project proposal information of the project establishment document into the offline knowledge base and building a persistent semantic index library for subsequent long text matching.

8. The method for analyzing the consistency between scientific and technological innovation project results and goals according to any one of claims 1 to 3, characterized in that: In step S6, the selected highly relevant text fragments are combined with the template-based query content to form a complete prompt word, and are input into the large language model for consistency analysis to automatically generate an evaluation result of the consistency between the scientific and technological innovation project results and the project establishment objectives.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements the method for analyzing the consistency between the results and goals of scientific and technological innovation projects as described in any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for analyzing the consistency between the results and goals of a scientific and technological innovation project as described in any one of claims 1 to 8 is implemented.

Images