Intelligent medical and invasive project evaluation method and system based on natural language processing

Abstract

The invention discloses an intelligent medical and invasive project evaluation method and system based on natural language processing. The method comprises the steps that professional corpora of multiple sources of an intelligent medical and invasive project are collected; the knowledge graph construction module comprises a term library construction module, a term clustering module and a relationship construction module; taking the professional corpus as a core to construct a professional word library; segmenting the professional lexicon into a plurality of clusters with professional terms as the center; constructing a knowledge graph by taking vocabularies in the clusters as nodes; and constructing an intelligent evaluation module which comprises an intelligent scoring module and an intelligent evaluation module, and outputting AI comments. According to the invention, the project plan is automatically processed through a natural language processing technology, so that the auditing efficiency is greatly improved; and in combination with a knowledge graph technology, a professional knowledge base is systematically constructed, background knowledge support is provided for a large language model, and the professionality and consistency of evaluation results are ensured.

Description

Technical Field

[0001] The present invention belongs to the field of natural language processing, and in particular relates to an intelligent scientific and technological innovation project evaluation method and system based on natural language processing. Background Art

[0002] With the strong support of society for scientific and technological innovation and the rapid development of the field of artificial intelligence, innovation and entrepreneurship competitions, as an important platform to stimulate innovation vitality, have become increasingly important in terms of project evaluation and review mechanisms. It is of great significance to improve the enterprise innovation service system and deepen innovation and entrepreneurship institutions.

[0003] In this context, how to accurately and efficiently tap into the potential market and deep value of a project has become a key issue that needs to be urgently addressed in various innovation and entrepreneurship competitions. As a carrier of the core value and goals of a project, the project plan can not only help innovative entrepreneurs more accurately position the project in the market and its competitive advantages during the writing process, but also allow review experts to intuitively understand the maturity, feasibility and future potential of the project, providing an important basis for the value assessment of innovative and entrepreneurial projects.

[0004] However, the traditional manual review model has problems such as large differences in subjective judgment, high review costs, and low efficiency, making it difficult to quickly respond to the growing demand for innovation and entrepreneurship. In addition, due to the high-risk and high-investment characteristics of scientific and technological innovation projects and the relative scarcity of scientific guidance resources, contestants face the dilemma of being unable to fully unleash their innovation potential, which affects the incubation and transformation of innovative results. Therefore, exploring a more efficient and fair review mechanism and strengthening precise support and professional guidance for innovative entrepreneurs have become urgent needs to solve the current dilemma and promote the healthy development of the innovation ecosystem. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide an intelligent scientific and technological innovation project evaluation method and system based on natural language processing, aiming to improve the efficiency of project book review, reduce interference from human factors, and provide more scientific and objective evaluation support for innovative entrepreneurs.

[0006] Technical solution: The intelligent scientific and technological innovation project evaluation method based on natural language processing of the present invention comprises the following steps:

[0007] Step 1: Collect professional corpus from various sources of intelligent science and technology innovation projects;

[0008] Step 2: Constructing the knowledge graph construction module includes a term base construction module, a term clustering module and a relationship construction module; through the term base construction module, a professional vocabulary is constructed with the professional corpus in step 1 as the core; through the term clustering module, the professional vocabulary is divided into multiple clusters centered on professional terms; through the relationship construction module, the vocabulary in the cluster is used as the node to construct the knowledge graph;

[0009] Step 3: Construct an intelligent evaluation module including an intelligent scoring module and an intelligent evaluation module; enhance and integrate the knowledge graph in step 2 with the knowledge graph and the external knowledge base through the intelligent scoring module; use a dynamic fusion strategy to calculate the score of each category and obtain the scoring result; obtain and optimize the comment set based on the scoring result and the key features of the knowledge graph through the intelligent evaluation module, and output the AI ​​comment;

[0010] Step 4: Further analyze the AI ​​comments through the expert evaluation module to obtain the final evaluation results.

[0011] Furthermore, step 1 is specifically as follows: the professional corpus from various sources of the intelligent scientific and technological innovation project includes academic papers, industry reports, technical documents and patents.

[0012] Furthermore, step 2 specifically includes the following steps:

[0013] Step 2.1: The terminology building module is used to build a professional vocabulary with the professional corpus collected by the information collection module as the core; set the document set as D = {d1, d2, ..., d m}, each document d i The vocabulary set in is T i ={t1,t2,…,t n}, for each word t j ∈T i , calculate its TF-IDF value in the document set D, set the threshold θ, and extract the term set C:

[0014] C={t i ∈T i |TF-IDF(t i ,D)>θ};

[0015] Step 2.2: The term clustering module is used to cluster the vocabulary set T i The vocabulary set T is divided into multiple clusters centered on professional terms, and the GloVe model is used to classify the vocabulary set T i Convert to a high-dimensional vector set Each of these Corresponding to T i The vocabulary in t j , with a term set The words in are cluster centers, and the vector of each word is calculated. With each term vector The distance between And classify it to the nearest cluster center The corresponding cluster C k middle:

[0016]

[0017] Among them, Cluster(t j ) represents vocabulary t j Belong to the cluster, cluster Contains and terms Similar vocabulary set, n k represents the number of terms in the kth cluster;

[0018] Step 2.3: The relationship building module is used to build a knowledge graph using the words in the cluster as nodes.

[0019] Furthermore, step 2.3 specifically includes the following steps:

[0020] Step 2.3.1: For the ki Perform dependency syntactic analysis on each sentence and record the relationship set R = {r1, r2, ..., r m}, will contain t ki Each sentence of is input into the BERT model to obtain t ki Dynamic context vector

[0021] Step 2.3.2: Cluster center As a central node For each relation r in the relation set R j , identify and t ki Related words j , and use these words as corresponding nodes V j =v j Add to the graph, at the central node V i With the relevant node V j A directed edge is established between them, and the edge label is relationship r j , represents the semantic relationship between edges;

[0022] Step 2.3.3: For each relevant word w j , if it has a dynamic context vector Then w j As the new central node, repeat the above steps to further parse its contextual relationship and form a multi-level knowledge graph structure; connect all central nodes and their related nodes according to the relationship, and finally organize them into a knowledge graph G.

[0023] Furthermore, step 3 specifically includes the following steps:

[0024] Step 3.1, the intelligent scoring module includes: a knowledge graph enhancement module and a fusion prediction module, which are used to receive the project book text submitted by the user, and extract the project book vocabulary set W = {w1, w2, ..., wn}, combine the feature vector and dynamic fusion strategy to calculate the intelligent score S of the project book;

[0025] Step 3.2: Generate the feature vector of the vocabulary set W by combining the internal knowledge graph G and the external knowledge base Wikidata through the knowledge graph enhancement module, and use the input project book text to supplement the knowledge graph G. For each word in the vocabulary set W, query the corresponding node N from the knowledge graph G. w , if node N w If it exists, get the adjacent node set N adj (w)={N1,N2,…,N k};

[0026] For each adjacent node N i ∈N adj (w), if N i If there is a match in the vocabulary set W, calculate its weight w i :

[0027]

[0028] Where, d(V(N i ),V(N w )) represents node N i and node N w The distance between the eigenvectors, d min and d max are the minimum and maximum values ​​of the distance, respectively, used for normalization;

[0029] For each word w i , update its feature vector V rel (w):

[0030]

[0031] Among them, N match Indicates that there is a matching set of adjacent nodes in the vocabulary set W;

[0032] Step 3.3: Integrate the feature information generated by the knowledge graph into the score prediction through the fusion prediction module to obtain the final score; input the vocabulary set W into the pre-trained FastText model to obtain the model prediction vector Among them, p i Indicates the probability that the text belongs to category i, according to the predicted probability p i Whether it exceeds the threshold γ, the basic score FIRST_SCORE is accumulated:

[0033]

[0034] in, is an indicator function, which takes the value 1 when the condition is true, otherwise it takes the value 0;

[0035] Step 3.4: Calculate the node set N generated by the knowledge graph enhancement module match The similarity between them gets an additional score ADDITIONAL_SCORE; the feature vector generated by combining the knowledge graph A dynamic fusion strategy is used to calculate the score s for each category. i :

[0036] s i =α·p i +β·v i ,

[0037] Among them, α is the FastText model prediction vector p i The weight coefficient represents the contribution of FastText’s initial prediction result to the final score, and Β is the knowledge graph feature vector v i The weight coefficient represents the contribution of the feature information generated by the knowledge graph to the final score, s i is the fused score vector;

[0038] The rating vector Input FastText model for prediction and generate probability distribution p=[p1,p2,…,p c ], combined with the competition scoring criteria, set the ideal distribution d = [d1, d2, …, d c ], where d i Represents the ideal score ratio of the i-th category. By calculating the deviation between p and the ideal distribution d, the second bonus score SECOND_SCORE is obtained:

[0039]

[0040] in, and They represent the minimum and maximum value of the bonus points, D represents the actual distance between p and the ideal distribution d, and D max Represents the maximum possible distance between p and the ideal distribution d. Add FIRST_SCORE, ADDITIONAL_SCORE and SECOND_SCORE to get the final score result;

[0041] Step 3.5: Through the intelligent evaluation module, according to the final scoring results and the extracted key features, historical expert comments are retrieved from the database and matched with the key features to obtain a set of comments and input them into the BERT model for integration and optimization. Combined with the coverage mechanism, AI comments with reference value are output.

[0042] Furthermore, in step 3.5, the key features include keywords, entities, and attributes related to the scoring criteria in the text.

[0043] Furthermore, step 4 is specifically as follows: input the keywords provided by the user into the pre-trained GPT large language model, accurately retrieve relevant historical project plan data through semantic analysis, and integrate to generate the prototype text of the scientific and technological innovation project; provide the project plan annotated by the intelligent evaluation module to experts for downloading and viewing, and analyze and evaluate the project based on the scores and comments generated by the system.

[0044] The present invention also discloses an intelligent scientific and technological innovation project evaluation system based on natural language processing, wherein the evaluation system comprises a database module, an information collection module, a knowledge graph construction module, an intelligent evaluation module, a scientific and technological innovation project prototype generation module, and an expert evaluation module;

[0045] The database module is used to store and manage project plans uploaded by users, user data, intelligent evaluation results and expert evaluation results;

[0046] The information collection module is used to collect professional corpus from various sources, and the knowledge graph construction module and intelligent evaluation module provide basic data support;

[0047] The knowledge graph construction module includes a term base construction module, a term clustering module and a relationship construction module;

[0048] The intelligent evaluation module includes an intelligent scoring module and an intelligent evaluation module;

[0049] The science and technology innovation project prototype generation module is used to input the keywords provided by the user into the pre-trained GPT large language model, accurately retrieve relevant historical project plan data through semantic analysis, and integrate and generate the science and technology innovation project prototype text;

[0050] The expert evaluation module is used to provide the project plan marked by the intelligent evaluation module to the experts for downloading and viewing.

[0051] The present invention also discloses a computer device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of the present invention.

[0052] The present invention also discloses a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method of the present invention are implemented.

[0053] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0054] 1. The present invention automates the processing of project plans through natural language processing technology, greatly improving the review efficiency; combined with knowledge graph technology, the system builds a professional knowledge base to provide background knowledge support for the large language model, ensuring the professionalism and consistency of the evaluation results; according to different review needs, the system can flexibly adjust the competition scoring standards and provide a flexible scoring mechanism to adapt to various review scenarios and requirements.

[0055] 2. The present invention ensures that the system's knowledge base can be dynamically updated through the information collection module and the knowledge graph construction module, keeps pace with the latest technologies, research results and market trends, continuously improves the accuracy and reliability of the system's intelligent evaluation, ensures that the review results can reflect current industry and technological trends, and improves the long-term practicality and accuracy of the system.

[0056] 3. The present invention integrates highly matched historical expert comments through deep learning models and intelligent scoring results, provides feedback with reference value, and helps project teams understand advantages and disadvantages and make improvements; with the help of pre-trained GPT large language model, the system can generate innovative and referenceable project templates from historical data, provide inspiration and reference for innovative entrepreneurs, and promote the conception and development of new projects.

[0057] 4. The present invention integrates an expert evaluation module to enable experts to conduct further in-depth analysis based on the intelligent evaluation results, improve the comprehensiveness and scientificity of the review, optimize the project screening and support processes, and provide researchers with new discoveries and trends in related fields, thereby promoting scientific research and technological innovation. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a system block diagram of the intelligent science and technology innovation project evaluation auxiliary system based on natural language processing of the present invention;

[0059] Figure 2 Constructing a module flow chart for the knowledge graph;

[0060] Figure 3 This is a flow chart of the intelligent scoring module. DETAILED DESCRIPTION

[0061] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0062] The purpose of the present invention can be achieved by the following technical solutions: an intelligent science and technology innovation project evaluation auxiliary system based on natural language processing, including a database module, an information collection module, a knowledge graph construction module, an intelligent evaluation module, a science and technology innovation project prototype generation module, and an expert evaluation module;

[0063] The database module is used to store and manage project plans, user data, intelligent evaluation results and expert evaluation results uploaded by users.

[0064] The information collection module is used to collect professional corpus from various sources, such as academic papers, industry reports, technical documents, patents, etc., to provide basic data support for subsequent knowledge graph construction modules and intelligent evaluation modules.

[0065] The knowledge graph construction module includes a term base construction module, a term clustering module and a relationship construction module.

[0066] Furthermore, the terminology library construction module is used to construct a professional vocabulary library with the professional corpus collected by the information collection module as the core. m}, each document d i The vocabulary set in is T i ={t1,t2,…,t n}, for each word t j ∈T i , calculate its TF-IDF value in the document set D. Set the threshold θ and extract the term set C:

[0067] C={t i ∈T i |TF-IDF(t i ,D)>θ}

[0068] Furthermore, the term clustering module is used to cluster the vocabulary set T i The vocabulary set T is divided into multiple clusters centered on professional terms. i Convert to a high-dimensional vector set Each of these Corresponding to T i The vocabulary in t j . By term collection The words in are cluster centers, and the vector of each word is calculated. With each term vector The distance between And classify it to the nearest cluster center The corresponding cluster C k middle:

[0069]

[0070] Among them, Cluster(t j ) represents vocabulary t j Belong to the cluster, cluster Contains and terms Similar vocabulary set, nk represents the number of terms in the k-th cluster.

[0071] Furthermore, the relationship building module is used to build a knowledge graph using the words in the cluster as nodes. The specific process is as follows:

[0072] 1. For the ki Perform dependency syntactic analysis on each sentence and record the relationship set R = {r1, r2, ..., r m}. Will contain t ki Each sentence of is input into the BERT model to obtain t ki The dynamic context vector

[0073] 2. Cluster center As a central node For each relation r in the relation set R j , identify and t ki Related words j , and use these words as corresponding nodes V j =v j Add to the diagram. At the center node V i With the relevant node V j A directed edge is established between them, and the edge label is relationship r j , indicating the semantic relationship between them.

[0074] 3. For each relevant word w j , if it has a dynamic context vector Then w j As the new central node, repeat the above steps to further parse its contextual relationship and form a multi-level knowledge graph structure.

[0075] Through the above steps, all central nodes and their related nodes are connected according to the relationship and finally organized into a knowledge graph G.

[0076] The intelligent evaluation module includes: an intelligent scoring module and an intelligent assessment module.

[0077] Furthermore, the intelligent scoring module includes: a knowledge graph enhancement module and a fusion prediction module, which are used to receive the project book text submitted by the user, and extract the project book vocabulary set W = {w1, w2, ..., w n}, combine the feature vector and dynamic fusion strategy to calculate the intelligent score S of the project book.

[0078] The knowledge graph enhancement module is used to combine the internal knowledge graph G and the external knowledge base Wikidata to generate the feature vector of the vocabulary set W, and use the input project book text to supplement the knowledge graph G. For each word in the vocabulary set W, the corresponding node N is queried from the knowledge graph G. w If node N w If it exists, get the adjacent node set N adj (w)={N1,N2,…,N k}.

[0079] For each adjacent node N i ∈N adj (w), if N i If there is a match in the vocabulary set W, calculate its weight w i :

[0080]

[0081] Where, d(B(N i ),V(N w )) represents node N i and node N w The distance between the eigenvectors, d min and d max They are the minimum and maximum distances, respectively, used for normalization.

[0082] For each word w i , update its feature vector V rel (w):

[0083]

[0084] Among them, N match Indicates that there is a matching set of adjacent nodes in the vocabulary set W.

[0085] The fusion prediction module is used to integrate the feature information generated by the knowledge graph into the score prediction to obtain the final score. Input the vocabulary set W into the pre-trained FastText model to obtain the model prediction vector Among them, p i Represents the probability that the text belongs to category i. According to the predicted probability p i Whether it exceeds the threshold γ, the basic score FIRST_SCORE is accumulated:

[0086]

[0087] in, is an indicator function that takes the value 1 if the condition is true, otherwise it takes the value 0.

[0088] By calculating the node set N generated by the knowledge graph enhancement module match The similarity between them gets an additional score ADDITIONAL_SCORE. Combined with the feature vector generated by the knowledge graph A dynamic fusion strategy is used to calculate the score s for each category. i :

[0089] s i =α·p i +β·v i ,

[0090] Among them, α is the FastText model prediction vector p i The weight coefficient represents the contribution of FastText’s initial prediction result to the final score. Β is the knowledge graph feature vector v i The weight coefficient represents the contribution of the feature information generated by the knowledge graph to the final score. i is the fused score vector.

[0091] The rating vector Further input FastText model for prediction, generating probability distribution p = [p1, p2, ..., p c ]. Combined with the competition scoring criteria, set the ideal distribution d = [d1, d2, …, d c ], where d i Represents the ideal score ratio of the i-th category. By calculating the deviation between p and the ideal distribution d, the second bonus score SECOND_SCORE is obtained:

[0092]

[0093] in, and They represent the minimum and maximum value of the bonus points, D represents the actual distance between p and the ideal distribution d, and D max It represents the maximum possible distance between p and the ideal distribution d. The final score is obtained by adding FIRST_SCORE, ADDITIONAL_SCORE and SECOND_SCORE.

[0094] Furthermore, the intelligent evaluation module is used to retrieve historical expert comments from the database and match them with these key features based on the scoring results and extracted key features, such as keywords, entities, and attributes related to the scoring criteria in the text, to obtain a set of comments and input them into the BERT model for integration and optimization, combined with a coverage mechanism to avoid duplication problems in the generation process, and output AI comments with reference value.

[0095] The science and technology innovation project prototype generation module is used to input the keywords provided by the user into the pre-trained GPT large language model, accurately retrieve relevant historical project plan data through semantic analysis, and integrate and generate innovative and referenceable science and technology innovation project prototype texts.

[0096] The expert evaluation module is used to provide the project plan marked by the intelligent evaluation module to the experts for downloading and viewing. The experts can conduct a more scientific and accurate analysis and evaluation of the project based on the scores and comments generated by the system.

Claims

1. An intelligent scientific and technological innovation project evaluation method based on natural language processing, characterized in that: The steps include: Step 1: Collect professional corpus from various sources of intelligent science and technology innovation projects; Step 2: Constructing the knowledge graph construction module includes a term base construction module, a term clustering module and a relationship construction module; through the term base construction module, a professional vocabulary is constructed with the professional corpus in step 1 as the core; through the term clustering module, the professional vocabulary is divided into multiple clusters centered on professional terms; through the relationship construction module, the vocabulary in the cluster is used as the node to construct the knowledge graph; Step 3: Construct an intelligent evaluation module including an intelligent scoring module and an intelligent evaluation module; enhance and integrate the knowledge graph in step 2 with the knowledge graph and the external knowledge base through the intelligent scoring module; use a dynamic fusion strategy to calculate the score of each category and obtain the scoring result; obtain and optimize the comment set based on the scoring result and the key features of the knowledge graph through the intelligent evaluation module, and output the AI ​​comment; Step 4: Further analyze the AI ​​comments through the expert evaluation module to obtain the final evaluation results.

2. According to the method for evaluating intelligent scientific and technological innovation projects based on natural language processing according to claim 1, it is characterized in that: Step 1 is specifically as follows: the professional corpus from various sources of the intelligent scientific and technological innovation project includes academic papers, industry reports, technical documents and patents.

3. According to the method for evaluating intelligent scientific and technological innovation projects based on natural language processing in claim 1, it is characterized in that: Step 2 specifically includes the following steps: Step 2.1: The terminology building module is used to build a professional vocabulary with the professional corpus collected by the information collection module as the core; set the document set as D = {d1, d2, ..., d m }, each document d i The vocabulary set in is T i ={t1,t2,…,t n }, for each word t j ∈T i , calculate its TF-IDF value in document set D, set threshold θ, and extract term set C: C={t i ∈T i ∣TF-IDF(t i ,D)>θ}; Step 2.2: The term clustering module is used to cluster the vocabulary set T i The vocabulary set T is divided into multiple clusters centered on professional terms, and the GloVe model is used to classify the vocabulary set T i Convert to a high-dimensional vector set Each of these Corresponding to T i The vocabulary in t j , with a term set The words in are cluster centers, and the vector of each word is calculated. With each term vector The distance between And classify it to the nearest cluster center The corresponding cluster C k middle: Among them, Cluster(t j ) represents vocabulary t j The cluster to which it belongs, cluster Contains and terms Similar vocabulary set, n k represents the number of terms in the kth cluster; Step 2.3: The relationship building module is used to build a knowledge graph using the words in the cluster as nodes.

4. According to claim 3, a method for evaluating intelligent scientific and technological innovation projects based on natural language processing is characterized in that: Step 2.3 specifically includes the following steps: Step 2.3.1: For the ki Perform dependency syntactic analysis on each sentence and record the relationship set R = {r1, r2, ..., r m }, will contain t ki Each sentence of is input into the BERT model to obtain t ki Dynamic context vector Step 2.3.2: Cluster center As a central node For each relation r in the relation set R j , identify and t ki Related words j , and use these words as corresponding nodes V j =v j Add to the graph, at the central node V i With the relevant node V j A directed edge is established between them, and the label of the edge is relationship r j , represents the semantic relationship between edges; Step 2.3.3: For each relevant word w j , if it has a dynamic context vector Then w j As the new central node, repeat the above steps to further parse its contextual relationship and form a multi-level knowledge graph structure; connect all central nodes and their related nodes according to the relationship, and finally organize them into a knowledge graph G.

5. According to claim 1, a method for evaluating intelligent scientific and technological innovation projects based on natural language processing is characterized in that: Step 3 specifically includes the following steps: Step 3.1, the intelligent scoring module includes: a knowledge graph enhancement module and a fusion prediction module, which are used to receive the project book text submitted by the user, and extract the project book vocabulary set W = {w1, w2, ..., w n }, combine the feature vector and dynamic fusion strategy to calculate the intelligent score S of the project book; Step 3.2: Generate the feature vector of vocabulary set w by combining the internal knowledge graph G and the external knowledge base Wikidata through the knowledge graph enhancement module, and use the input project book text to supplement the knowledge graph G. For each word in the vocabulary set W, query the corresponding node N from the knowledge graph G. w , if node N w If it exists, get the adjacent node set N adj (w)={N1,N2,…,N k }; For each adjacent node N i ∈N adj (w), if N i If there is a match in the vocabulary set W, calculate its weight w i : Where, d(V(N i ),V(N w )) represents node N i and node N w The distance between the eigenvectors, d min and d max are the minimum and maximum values ​​of the distance, respectively, used for normalization; For each word w i , update its feature vector V rel (w): Among them, N match Indicates that there is a matching set of adjacent nodes in the vocabulary set W; Step 3.3: Integrate the feature information generated by the knowledge graph into the score prediction through the fusion prediction module to obtain the final score; input the vocabulary set W into the pre-trained FastText model to obtain the model prediction vector Among them, p i Indicates the probability that the text belongs to category i, according to the predicted probability p i Whether it exceeds the threshold γ, the basic score FIRST_SCORE is accumulated: in, is an indicator function, which takes the value 1 when the condition is true, otherwise it takes the value 0; Step 3.4: Calculate the node set N generated by the knowledge graph enhancement module match The similarity between them gets an additional score ADDITIONAL_SCORE; the feature vector generated by combining the knowledge graph A dynamic fusion strategy is used to calculate the score s for each category. i : s i =α·p i +β·v i , Among them, α is the FastText model prediction vector p i The weight coefficient represents the contribution of FastText’s initial prediction result to the final score, and Β is the knowledge graph feature vector v i The weight coefficient represents the contribution of the feature information generated by the knowledge graph to the final score, s i is the fused score vector; The rating vector Input FastText model for prediction and generate probability distribution p=[p1,p2,…,p c ], combined with the competition scoring criteria, set the ideal distribution d = [d1, d2, …, d c ], where d i Represents the ideal score ratio of the i-th category. By calculating the deviation between p and the ideal distribution d, the second bonus score SECOND_SCORE is obtained: in, and They represent the minimum and maximum value of the bonus points, D represents the actual distance between p and the ideal distribution d, and D max Represents the maximum possible distance between p and the ideal distribution d. Add FIRST_SCORE, ADDITIONAL_SCORE and SECOND_SCORE to get the final score result; Step 3.5: Through the intelligent evaluation module, according to the final scoring results and the extracted key features, historical expert comments are retrieved from the database and matched with the key features to obtain a set of comments and input them into the BERT model for integration and optimization. Combined with the coverage mechanism, AI comments with reference value are output.

6. According to claim 5, a method for evaluating intelligent scientific and technological innovation projects based on natural language processing is characterized in that: In step 3.5, the key features include keywords, entities, and attributes related to the scoring criteria in the text.

7. According to claim 1, a method for evaluating intelligent scientific and technological innovation projects based on natural language processing is characterized in that: Step 4 is as follows: input the keywords provided by the user into the pre-trained GPT large language model, accurately retrieve relevant historical project plan data through semantic analysis, and integrate them to generate the prototype text of the scientific and technological innovation project; provide the project plan annotated by the intelligent evaluation module to experts for download and viewing, and analyze and evaluate the project based on the scores and comments generated by the system.

8. An intelligent scientific and technological innovation project evaluation system based on natural language processing, used to implement the method according to any one of claims 1 to 7, characterized in that: The evaluation system includes a database module, an information collection module, a knowledge graph construction module, an intelligent evaluation module, a scientific and technological innovation project prototype generation module, and an expert evaluation module; The database module is used to store and manage project plans uploaded by users, user data, intelligent evaluation results and expert evaluation results; The information collection module is used to collect professional corpus from various sources, and the knowledge graph construction module and intelligent evaluation module provide basic data support; The knowledge graph construction module includes a term base construction module, a term clustering module and a relationship construction module; The intelligent evaluation module includes an intelligent scoring module and an intelligent evaluation module; The science and technology innovation project prototype generation module is used to input the keywords provided by the user into the pre-trained GPT large language model, accurately retrieve relevant historical project plan data through semantic analysis, and integrate and generate the science and technology innovation project prototype text; The expert evaluation module is used to provide the project plan marked by the intelligent evaluation module to the experts for downloading and viewing.

9. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of claim 1.

10. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

Images