Paper quality intelligent evaluation system based on multi-model fusion

Abstract

The invention discloses a paper quality intelligent evaluation system based on multi-model fusion, and belongs to the technical field of paper evaluation systems. Comprising a paper uploading module, a format evaluation module, a graphic formula content recognition module, a paper cleaning module, a paper partitioning module, a text duplicate checking evaluation module, a research method evaluation text generation module, an abstract evaluation module, a reference evaluation module, a research method evaluation module and a paper quality evaluation module. In this way, comprehensive quality scoring of the papers is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of paper evaluation systems, and in particular to an intelligent paper quality evaluation system based on multi-model fusion. Background Art

[0002] The paper quality assessment system is a tool or platform used to evaluate the quality of academic papers, usually analyzing the innovation and technicality of the paper based on a series of indicators or parameters; Among many existing technologies, Chinese patent application CN202411559273.X discloses a method and related equipment for evaluating the quality of scientific research paper abstract generation, which includes obtaining the original text, target abstract and references of the target scientific research paper and performing bidirectional encoding in sequence to obtain the original text embedding vector, the abstract text embedding vector, and the reference text embedding vector, extracting a pseudo target abstract based on the original text and encoding it to obtain a pseudo abstract text embedding vector; calculating each text embedding vector to obtain a comprehensive semantic evaluation score; and calculating a natural coherence score based on the calculated word-by-word probability and perplexity; concatenating the target abstract, the original text and the indicator set and inputting the prompt words into a large language model to score the target abstract to obtain a self-scoring result; weightedly fusing the comprehensive semantic evaluation score, the natural coherence score, the self-scoring result and the Rouge indicator to obtain a quality score of the target abstract, and evaluating the generation quality of the target abstract based on the quality score; improving the evaluation efficiency of the abstract quality.

[0003] However, this patent application only evaluates the abstract of the paper and cannot conduct a comprehensive quality assessment of the paper. Based on this, the present invention designs an intelligent paper quality evaluation system based on multi-model fusion to solve the above problems. Summary of the Invention

[0004] In view of the above shortcomings of the prior art, the present invention provides an intelligent paper quality evaluation system based on multi-model fusion.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: An intelligent paper quality assessment system based on multi-model fusion, including a paper upload module, a format assessment module, a graphic formula content recognition module, a paper cleaning module, a paper segmentation module, a main text duplication check and assessment module, a research method assessment text generation module, an abstract assessment module, a reference assessment module, a research method assessment module, and a paper quality assessment module; The paper uploading module is in communication connection with the format evaluation module, the format evaluation module is in communication connection with the graphic formula content recognition module, the graphic formula content recognition module is in communication connection with both the paper cleaning module and the paper quality evaluation module, the paper cleaning module is in communication connection with the paper segmentation module, the paper segmentation module is in communication connection with the main text duplicate checking evaluation module, the abstract evaluation module, and the reference evaluation module, the abstract evaluation module and the reference evaluation module are in communication connection with the paper quality evaluation module, the main text duplicate checking evaluation module is in communication connection with the research method evaluation text generation module and the paper quality evaluation module, the research method evaluation text generation module is in communication connection with the research method evaluation module, and the research method evaluation module is in communication connection with the paper quality evaluation module; The paper uploading module is used to upload the paper, and the paper uploading module sends the paper to the format evaluation module; The format assessment module is used for paper format review, and the format assessment module sends the papers that have passed the review to the graphic formula content recognition module; The format assessment module identifies whether the paper is fully equipped with the title, subject, table of contents, abstract, main text and reference headings; If yes, the format assessment module sends the paper to the graphic formula content recognition module; If not, the format assessment module will prompt that there is missing title content and terminate the assessment of the paper; The graphic formula content recognition module is used to identify and extract images and formulas in the research method part of the paper; And use the VILA large model to convert images into natural language; The formulas are converted into natural language 2 through the MathText large model; The graphic formula content recognition module evaluates the relevance of natural language 1 and natural language 2 with the context of their location and generates an image formula relevance score E. The graphic formula content recognition module sends E to the paper quality assessment module. The graphic formula content recognition module sends the paper to the paper cleaning module; The paper cleaning module is used to clean the paper content. The paper cleaning module deletes the main titles and their contents except the abstract, main text, and reference titles in the paper, and generates a cleaned paper. The paper cleaning module sends the cleaned paper to the paper segmentation module; The paper segmentation module is used to segment the cleaned paper into blocks according to the main title, obtaining the abstract block, the main text block, and the reference block; the abstract block, the main text block, and the reference block all include the main title and its content; The paper chunking module sends the summary chunks to the summary evaluation module; The paper block module sends the text blocks to the text plagiarism assessment module; The paper block module sends the text block and reference block to the reference evaluation module; The main text duplication checking and evaluation module is used to check the main text block of the paper for duplication and obtain the duplication checking rate A, and generate a duplication checking score C. The main text duplication checking and evaluation module sends C to the paper quality assessment module; The main text plagiarism check and evaluation module checks the main text content based on the paper plagiarism check algorithm and marks the plagiarized content. The main text plagiarism check and evaluation module sends the plagiarized text to the research method evaluation text generation module; The research method evaluation text generation module is used to identify the duplicate content marked in the research method content of the text after the duplicate check, and extract the duplicate content in the research method and generate a research method evaluation text. The research method evaluation text generation module sends the research method evaluation text to the research method evaluation module; The reference evaluation module is used to retrieve references and extract their abstracts. The reference evaluation module compares the abstracts of the references with the corresponding content in the literature review in the main text and generates a literature relevance score V. The reference evaluation module sends V to the paper quality evaluation module. The abstract evaluation module is used to evaluate the abstract content and generate an abstract score Z, which is sent to the paper quality assessment module; The research method evaluation module is used to evaluate the research method text and generate a research method score J. The research method evaluation module sends J to the paper quality assessment module; The paper quality assessment module is used to calculate the paper quality score K based on C, J, V, Z, and E.

[0006] Furthermore, the format assessment module identifies whether the paper has a title, subject, table of contents, abstract, main text, and references; If yes, the format assessment module sends the paper to the graphic formula content recognition module; If not, the format assessment module will prompt that there is missing content and terminate the assessment of the paper.

[0007] Furthermore, the graphic formula content recognition module identifies whether there are images or formulas in the research method section; If so, the graphic formula content recognition module calculates the absolute value G of the similarity between the natural language 1, the natural language 2 and the context of their location based on the cosine similarity algorithm; The graphical formula content recognition module calculates E and uses the following formula: E=5(G1+G2+...+Gn) / n; Where n represents the number of G; If no, skip.

[0008] Furthermore, the C is calculated using the following formula: C=5(1-A).

[0009] Furthermore, the reference evaluation module (9) calculates the absolute value Q of the similarity between the abstract contents of the multiple references and the literature review contents in the corresponding text based on the cosine similarity algorithm. The reference evaluation module (9) calculates the average similarity value q based on the multiple similarity values Q, and calculates the literature relevance score V based on q. And use the following formula: V=5q.

[0010] Furthermore, the Z is calculated using the following formula: Z=z1+z2+z3+z4+z5; Where z1 represents the completeness score of the summary; z2 represents the accuracy score of the summary; z3 represents the normative score of the abstract; z4 represents the conciseness score of the abstract; z5 represents the fluency score of the summary.

[0011] Furthermore, the values of z1, z2, z3, z4, and z5 are determined by the following steps: In step 1, the abstract evaluation module uses a large natural language model to determine whether the terminology used in the abstract is accurate. If so, z1 takes the value of 1; If not, z1 takes the value of 0; In step 2, the abstract evaluation module uses a natural language model to determine whether the abstract comprehensively summarizes the background, methods, main findings, and conclusions of the research. If yes, z2 takes the value of 1; If not, z2 takes the value of 0; Step 3: The abstract evaluation module determines whether the number of characters in the abstract is greater than or equal to 200; If so, z3 takes the value of 1; If not, z3 takes the value of 0; Step 4: The abstract evaluation module uses the natural language model to check whether the number of keywords in the abstract is between 3 and 5; If so, z4 takes the value of 1; If not, z4 takes the value of 0; Step 5: The abstract evaluation module judges the fluency of the English abstract based on the natural language model; If yes, z5 takes the value of 1; If not, z5 takes the value of 0.

[0012] Furthermore, the research method evaluation module identifies theoretical texts in the duplicate check content of the research method text based on the natural language large model, obtains B theoretical texts, and judges the hotness of each theoretical text respectively; The research method evaluation module retrieves the total number of citations (S) of a theoretical text in authoritative journals from its publication date R1 to the evaluation date R2, and calculates the total number of citations (S1) in authoritative journals within the time period (R2-R1) / 3 before R2. It then uses the formula W=S1 / S to determine whether W is greater than 50%, where W represents the interval percentage. If so, the theoretical text is judged to be hot topic; If not, it is determined that the theoretical text is not hot topic.

[0013] Furthermore, the research method evaluation module counts the number of theoretical texts judged as hot topics B1 and the total number of theoretical texts B, and calculates J using the following formula: J=5B1 / B.

[0014] Furthermore, K adopts the following formula: K=u1C+u2J+u3V+u4Z+u5E; Among them, u1+u2+u3+u4+u5=1.

[0015] Compared with the existing technology, the present invention has the following advantages: it can recognize and extract images and formulas in the research method part of the paper through the graphic formula content recognition module, interpret the images and formulas into natural language one and natural language two based on Math2Text, and evaluate their relevance in the context, thereby realizing the evaluation of the relevance between images and formulas in the paper and the paper, and improving the comprehensiveness of the evaluation; The paper is divided into blocks through the paper block module, and the abstract and references are evaluated through the abstract evaluation module and the reference evaluation module. The main text research method part is evaluated through the main text duplication evaluation module, the research method evaluation text generation module, and the research method evaluation module, thus realizing the evaluation of the paper blocks. The paper is processed through multi-model fusion, which improves the efficiency and accuracy of paper evaluation; The paper quality assessment module is used to comprehensively score the paper's quality based on the above assessment results. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort.

[0017] Figure 1 It is a block diagram of an intelligent paper quality evaluation system based on multi-model fusion of the present invention; Figure 2 A statistical chart of citations of theoretical texts by the research method evaluation module of the present invention; Figure 3 This is a schematic diagram of the paper after cleaning of the present invention; Figure 4 A schematic diagram of the text for evaluating the research method of the present invention.

[0018] The numbers in the figure represent: 1. Paper upload module 2. Format assessment module 3. Graphic and formula content recognition module 4. Paper cleaning module 5. Paper segmentation module 6. Text duplication check and assessment module 7. Research method assessment text generation module 8. Abstract assessment module 9. Reference assessment module 10. Research method assessment module 11. Paper quality assessment module DETAILED DESCRIPTION

[0019] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0020] Example 1: In some embodiments, please refer to the accompanying drawings of the specification. Figures 1-4 , an intelligent paper quality evaluation system based on multi-model fusion, including paper uploading module 1, format evaluation module 2, graphic formula content recognition module 3, paper cleaning module 4, paper segmentation module 5, text duplication check and evaluation module 6, research method evaluation text generation module 7, abstract evaluation module 8, reference evaluation module 9, research method evaluation module 10 and paper quality evaluation module 11; The paper uploading module 1 is in communication connection with the format evaluation module 2, the format evaluation module 2 is in communication connection with the graphic formula content recognition module 3, the graphic formula content recognition module 3 is in communication connection with the paper cleaning module 4 and the paper quality evaluation module 11, the paper cleaning module 4 is in communication connection with the paper segmentation module 5, the paper segmentation module 5 is in communication connection with the main text duplicate checking and evaluation module 6, the abstract evaluation module 8, and the reference evaluation module 9, the abstract evaluation module 8 and the reference evaluation module 9 are in communication connection with the paper quality evaluation module 11, the main text duplicate checking and evaluation module 6 is in communication connection with the research method evaluation text generation module 7 and the paper quality evaluation module 11, the research method evaluation text generation module 7 is in communication connection with the research method evaluation module 10, and the research method evaluation module 10 is in communication connection with the paper quality evaluation module 11; The paper uploading module 1 is used to upload the paper, and the paper uploading module 1 sends the paper to the format evaluation module 2; The format assessment module 2 is used for paper format review. The format assessment module 2 sends the papers that have passed the review to the graphic formula content recognition module 3; Format Assessment Module 2 identifies whether the paper is fully equipped with title, subject, table of contents, abstract, main text and references; If yes, the format assessment module 2 sends the paper to the graphic formula content recognition module 3; If not, the format evaluation module 2 will prompt that there is missing title content and terminate the evaluation of the paper; Graphic formula content recognition module 3 is used to identify and extract images and formulas in the research method part of the paper; And use the VILA large model to convert images into natural language; The formulas are converted into natural language 2 through the Math2Text model; Natural language refers to the natural language description of text data or image content in the image; Natural language 2 refers to describing the formula in natural language.

[0021] The graphic formula content recognition module 3 evaluates the relevance of the natural language 1 and the natural language 2 with the context of their location and generates an image formula relevance score E. The graphic formula content recognition module 3 sends E to the paper quality assessment module 11; Graphics and formula content recognition module 3 identifies whether there are images or formulas in the research method section; If so, the graphic formula content recognition module 3 calculates the absolute value G of the similarity between the natural language 1, the natural language 2 and the context of their location based on the cosine similarity algorithm; The graphic formula content recognition module 3 calculates E and uses the following formula: E=5(G1+G2+...+Gn) / n; Where n represents the number of G; If no, skip.

[0022] Context refers to the previous and next sentences where the image or formula is located.

[0023] The graphic formula content recognition module 3 sends the paper to the paper cleaning module 4; The paper cleaning module 4 is used to clean the paper content. The paper cleaning module 4 deletes the main titles and their contents except the abstract, main text, and reference titles in the paper, and generates a cleaned paper. The paper cleaning module 4 sends the cleaned paper to the paper segmentation module 5; The paper segmentation module 5 is used to segment the cleaned paper into blocks according to the main title, obtaining the abstract block, the main text block, and the reference block; the abstract block, the main text block, and the reference block all include the main title and its content; The paper block module 5 sends the summary block to the summary evaluation module 8; The paper block module 5 sends the text block to the text duplication checking and evaluation module 6; The paper block module 5 sends the text block and reference block to the reference evaluation module 9; The main text duplication checking and evaluation module 6 is used to check the main text blocks of the paper for duplication and obtain the duplication checking rate A, and generate a duplication checking score C. The main text duplication checking and evaluation module 6 sends C to the paper quality evaluation module 11; The C is calculated using the following formula: C=5(1-A); The main text duplication checking and evaluation module 6 checks the main text content for duplication based on the paper duplication checking algorithm and marks the duplication-checked content. The main text duplication checking and evaluation module 6 sends the duplication-checked main text to the research method evaluation text generation module 7; The paper duplication checking algorithm adopts HowNet paper duplication checking; The research method evaluation text generation module 7 is used to identify the duplicate content marked in the research method content of the main text after the duplicate check, and extract the duplicate content in the research method to generate a research method evaluation text. The research method evaluation text generation module 7 sends the research method evaluation text to the research method evaluation module 10; The reference evaluation module 9 is used to retrieve references and extract their abstracts. The reference evaluation module 9 compares the abstracts of the references with the corresponding contents in the literature review in the main text and generates a literature relevance score V. The reference evaluation module 9 sends V to the paper quality evaluation module 11. The reference evaluation module 9 calculates the absolute value Q of the similarity between the abstract contents of multiple references and the literature review contents in the corresponding text based on the cosine similarity algorithm. The reference evaluation module 9 calculates the average similarity value q based on the multiple similarity values Q, and calculates the literature relevance score V based on q; And use the following formula: V=5q; The abstract evaluation module 8 is used to evaluate the abstract content and generate an abstract score Z. The abstract evaluation module 8 sends Z to the paper quality evaluation module 11; Z is calculated using the following formula: Z=z1+z2+z3+z4+z5; Where z1 represents the completeness score of the summary; z2 represents the accuracy score of the summary; z3 represents the normative score of the abstract; z4 represents the conciseness score of the abstract; z5 represents the fluency score of the summary; The values of z1, z2, z3, z4, and z5 are determined using the following steps: Step 1: The abstract evaluation module 8 determines whether the terminology in the abstract is accurate based on the natural language model; If so, z1 takes the value of 1; If not, z1 takes the value of 0; Step 2: The abstract evaluation module 8 uses the natural language model to determine whether the abstract fully summarizes the background, methods, main findings, and conclusions of the research; If yes, z2 takes the value of 1; If not, z2 takes the value of 0; Step 3: the abstract evaluation module 8 determines whether the number of characters in the abstract is greater than or equal to 200; If so, z3 takes the value of 1; If not, z3 takes the value of 0; Step 4: The abstract evaluation module 8 checks whether the number of keywords in the abstract is between 3 and 5 based on the natural language model; If so, z4 takes the value of 1; If not, z4 takes the value of 0; Step 5: The abstract evaluation module 8 judges the fluency of the English abstract in the abstract based on the natural language model; If yes, z5 takes the value of 1; If not, z5 takes the value of 0.

[0024] The research method evaluation module 10 is used to evaluate the research method text and generate a research method score J. The research method evaluation module 10 sends J to the paper quality evaluation module 11; The research method evaluation module 10 identifies theoretical texts in the duplicate content of the research method text based on the natural language model, obtains B theoretical texts, and judges the hotness of each theoretical text; Research method evaluation module 10 searches for the total number of citations S of the theoretical text in authoritative journals from the publication date R1 to the evaluation date R2, and calculates the total number of citations S1 in authoritative journals within the time period (R2-R1) / 3 before R2, and uses the formula W=S1 / S to determine whether W is greater than 50%, where W represents the interval proportion; If so, the theoretical text is judged to be hot topic; If not, the theoretical text is judged to be not hot topic; The research method evaluation module 10 counts the number of theoretical texts B1 that are judged to be hot topics and the total number of theoretical texts B, and calculates J using the following formula: J=5B1 / B; In this invention, the authoritative journal is HowNet; The W value can reflect the proportion of the theoretical text's recent citations to the total number of citations. A larger W value can indicate that the citations of the theoretical text are on the rise, thereby determining the popularity of the theoretical text.

[0025] The paper quality assessment module 11 is used to calculate the paper quality score K based on C, J, V, Z, and E.

[0026] K uses the following formula: K=u1C+u2J+u3V+u4Z+u5E Among them, u1+u2+u3+u4+u5=1.

[0027] The values of u1, u2, u3, u4, and u5 are 0.3, 0.2, 0.2, 0.1, and 0.2 respectively.

[0028] When the present invention is used, the image and formula content recognition module 3 is used to recognize and extract the images and formulas in the research method part of the paper, and the images and formulas are converted into natural language 1 and natural language 2, and their relevance in the context is evaluated, thereby realizing the evaluation of the relevance between the images and formulas in the paper and the paper, and improving the comprehensiveness of the evaluation; The paper is divided into blocks through the paper block module 5, and the paper abstract and references are evaluated through the abstract evaluation module 8 and the reference evaluation module 9. The main text research method part is evaluated through the main text duplication evaluation module 6, the research method evaluation text generation module 7, and the research method evaluation module 10, thus realizing the evaluation of the paper blocks; The paper quality assessment module 11 is used to comprehensively assess the paper's quality based on the above assessment results.

[0029] When the present invention is used, the relevance between the references of the paper and the content of the literature review in the main text is evaluated, thereby realizing the relevance between the cited documents and the content of the literature review.

[0030] When the present invention is used, the comprehensiveness of the abstract scoring is improved by scoring the completeness, accuracy, standardization, conciseness and fluency of the paper abstract.

[0031] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An intelligent paper quality assessment system based on multi-model fusion, characterized by: It includes the paper uploading module (1), format evaluation module (2), graphic formula content recognition module (3), paper cleaning module (4), paper segmentation module (5), text duplication check and evaluation module (6), research method evaluation text generation module (7), abstract evaluation module (8), reference evaluation module (9), research method evaluation module (10) and paper quality assessment module (11); The paper uploading module (1) is in communication connection with the format evaluation module (2), the format evaluation module (2) is in communication connection with the graphic formula content recognition module (3), the graphic formula content recognition module (3) is in communication connection with the paper cleaning module (4) and the paper quality evaluation module (11), the paper cleaning module (4) is in communication connection with the paper segmentation module (5), the paper segmentation module (5) is in communication connection with the main text duplicate checking evaluation module (6), the abstract evaluation module (8), and the reference evaluation module (9), the abstract evaluation module (8) and the reference evaluation module (9) are in communication connection with the paper quality evaluation module (11), the main text duplicate checking evaluation module (6) is in communication connection with the research method evaluation text generation module (7) and the paper quality evaluation module (11), the research method evaluation text generation module (7) is in communication connection with the research method evaluation module (10), and the research method evaluation module (10) is in communication connection with the paper quality evaluation module (11); The paper uploading module (1) is used for uploading the paper, and the paper uploading module (1) sends the paper to the format evaluation module (2); The format evaluation module (2) is used for reviewing the paper format. The format evaluation module (2) sends the papers that have passed the review to the graphic formula content recognition module (3); The format assessment module (2) identifies whether the paper is fully equipped with the title, subject, table of contents, abstract, main text and references; If yes, the format assessment module (2) sends the paper to the graphic formula content identification module (3); If not, the format evaluation module (2) will prompt that there is a missing title content and terminate the evaluation of the paper; The graphic formula content recognition module (3) is used to identify and extract images and formulas in the research method part of the paper; And use the VILA large model to convert images into natural language; The formulas are converted into natural language 2 through the Math2Text model; The graphic formula content recognition module (3) evaluates the relevance of the natural language 1 and the natural language 2 with the context of their location and generates an image formula relevance score E. The graphic formula content recognition module (3) sends E to the paper quality assessment module (11); The graphic formula content recognition module (3) sends the paper to the paper cleaning module (4); The paper cleaning module (4) is used to clean the paper content. The paper cleaning module (4) deletes the main titles and contents of the paper except the abstract, main text, and reference titles, and generates a cleaned paper. The paper cleaning module (4) sends the cleaned paper to the paper segmentation module (5); The paper segmentation module (5) is used to segment the cleaned paper according to the main title to obtain the abstract block, the main text block, and the reference block; the abstract block, the main text block, and the reference block all include the main title and its content; The paper chunking module (5) sends the summary chunk to the summary evaluation module (8); The paper block module (5) sends the text blocks to the text duplication assessment module (6); The paper block module (5) sends the text block and reference block to the reference evaluation module (9); The main text duplication checking and evaluation module (6) is used to check the main text blocks of the paper for duplication and obtain the duplication checking rate A, and generate a duplication checking score C. The main text duplication checking and evaluation module (6) sends C to the paper quality evaluation module (11); The main text duplication checking and evaluation module (6) checks the main text content for duplication based on the paper duplication checking algorithm and marks the duplication-checked content. The main text duplication checking and evaluation module (6) sends the duplication-checked main text to the research method evaluation text generation module (7); The research method evaluation text generation module (7) is used to identify the duplicate content marked in the research method content of the main text after the duplicate check, and extract the duplicate content in the research method to generate a research method evaluation text. The research method evaluation text generation module (7) sends the research method evaluation text to the research method evaluation module (10); The reference evaluation module (9) is used to retrieve references and extract their abstract contents. The reference evaluation module (9) compares the relevance of the abstract contents of the references with the corresponding contents in the literature review in the main text and generates a literature relevance score V. The reference evaluation module (9) sends V to the paper quality evaluation module (11); The abstract evaluation module (8) is used to evaluate the abstract content and generate an abstract score Z. The abstract evaluation module (8) sends Z to the paper quality evaluation module (11); The research method evaluation module (10) is used to evaluate the research method text and generate a research method score J. The research method evaluation module (10) sends J to the paper quality evaluation module (11); The paper quality assessment module (11) is used to calculate the paper quality score K based on C, J, V, Z, and E.

2. The intelligent paper quality assessment system based on multi-model fusion according to claim 1 is characterized in that: The format assessment module (2) identifies whether the paper has a title, subject, table of contents, abstract, main text, and references; If yes, the format assessment module (2) sends the paper to the graphic formula content identification module (3); If not, the format evaluation module (2) will prompt that there is missing content and terminate the evaluation of the paper.

3. The intelligent paper quality assessment system based on multi-model fusion according to claim 2 is characterized in that: The graphic formula content recognition module (3) identifies whether there are images or formulas in the research method section; If it exists, the graphic formula content recognition module (3) calculates the absolute value G of the similarity between the natural language 1, the natural language 2 and the context of their location based on the cosine similarity algorithm; The graphical formula content recognition module (3) calculates E and uses the following formula: E=5(G1+G2+...+Gn) / n; Where n represents the number of G; If no, skip.

4. The paper quality intelligent evaluation system based on multi-model fusion according to claim 3 is characterized in that: The C is calculated using the following formula: C=5(1-A).

5. The intelligent paper quality assessment system based on multi-model fusion according to claim 4 is characterized in that: The reference evaluation module (9) calculates the absolute value Q of the similarity between the abstract contents of multiple references and the literature review contents in the corresponding text based on the cosine similarity algorithm. The reference evaluation module (9) calculates the average similarity value q based on the multiple similarity values Q, and calculates the literature relevance score V based on q; And use the following formula: V=5q.

6. The intelligent paper quality assessment system based on multi-model fusion according to claim 5 is characterized in that: The Z is calculated using the following formula: Z=z1+z2+z3+z4+z5; Where z1 represents the completeness score of the summary; z2 represents the accuracy score of the summary; z3 represents the normative score of the abstract; z4 represents the conciseness score of the abstract; z5 represents the fluency score of the summary.

7. The intelligent paper quality assessment system based on multi-model fusion according to claim 6 is characterized in that: The values of z1, z2, z3, z4, and z5 are determined by the following steps: Step 1: The abstract evaluation module (8) determines whether the terminology in the abstract is accurate based on the natural language model; If so, z1 takes the value of 1; If not, z1 takes the value of 0; Step 2: The abstract evaluation module (8) uses the natural language model to determine whether the abstract fully summarizes the background, methods, main findings, and conclusions of the study; If yes, z2 takes the value of 1; If not, z2 takes the value of 0; Step 3, the abstract evaluation module (8) identifies whether the number of characters in the abstract is greater than or equal to 200; If so, z3 takes the value of 1; If not, z3 takes the value of 0; Step 4: The abstract evaluation module (8) checks whether the number of keywords in the abstract is between 3 and 5 based on the natural language model; If so, z4 takes the value of 1; If not, z4 takes the value of 0; Step 5, the abstract evaluation module (8) judges the fluency of the English abstract in the abstract based on the natural language model; If yes, z5 takes the value of 1; If not, z5 takes the value of 0.

8. The intelligent paper quality assessment system based on multi-model fusion according to claim 7 is characterized in that: The research method evaluation module (10) identifies theoretical texts in the duplicate content of the research method text based on the natural language large model, obtains B theoretical texts, and judges the hotness of each theoretical text respectively; Research method evaluation module (10) retrieves the total number of citations S of the theoretical text in authoritative journals from the publication date R1 to the evaluation date R2, and calculates the total number of citations S1 in authoritative journals within the time period (R2-R1) / 3 before R2, and uses the formula W=S1 / S to determine whether W is greater than 50%, where W represents the interval proportion; If so, the theoretical text is judged to be hot topic; If not, it is determined that the theoretical text is not hot topic.

9. The intelligent paper quality assessment system based on multi-model fusion according to claim 8 is characterized in that: The research method evaluation module (10) counts the number of theoretical texts judged as hot topics B1 and the total number of theoretical texts B, and calculates J using the following formula: J=5B1 / B.

10. The intelligent paper quality assessment system based on multi-model fusion according to any one of claims 1 to 9 is characterized in that: K uses the following formula: K=u1C+u2J+u3V+u4Z+u5E; Among them, u1+u2+u3+u4+u5=1.

Images