Intelligent expert recommendation method for blind examination of academic papers based on expert portraits

Abstract

The invention discloses an expert portrait-based academic thesis blind examination expert intelligent recommendation method, which comprises the following steps of: constructing an expert behavior and social dimension system according to factors such as expert examination probability, historical review book quality, thesis review duration, historical review cycle and the like obtained in a blind examination system historical expert matching service; and better initial recommendation is provided for the user. According to the method, the defects of high manual matching cost, long period and incapability of quickly adjusting a submission strategy are reduced, so that the cost of the paper blind examination is reduced, the period is shortened, customized service can be realized in each batch, and a part of contribution is made to the paper blind examination in the academic circle.

Description

Technical Field

[0001] The present invention belongs to the technical field of recommendation systems, and in particular, is an intelligent recommendation method for blind review experts of degree theses based on expert portraits. Background Art

[0002] With the rapid development of informatization and the Internet, in order to solve the problem of information overload and provide people with effective information, recommendation technology has become a technology that many researchers have paid attention to. On major e-commerce websites, many fast and efficient recommendation methods are used to improve service quality. For example, based on the user's browsing history, favorite products and other information, users are recommended items that may be purchased. Such recommendation technology is also applied to news information, music, movies, books, etc. In the review of degree thesis, there is also a similar application scenario, that is, recommending appropriate review experts for degree thesis, reviewing the thesis and giving quality comments and suggestions. In order to improve the efficiency and accuracy of reviewer selection, existing systems and research usually combine large models to analyze the deep semantic relationship between thesis keywords and experts, better capture the semantic relationship of the text, and improve the accuracy and richness of semantic representation. However, it also has certain limitations: most of the existing methods are based solely on the similarity of text information such as thesis keywords or reviewer-related information, but ignore the behavioral characteristics of expert review and user feedback. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent recommendation method for blind review experts of degree thesis based on expert portraits.

[0004] The technical solution to achieve the purpose of the present invention is: an intelligent recommendation method for blind review experts of degree thesis based on expert portraits, comprising the following steps:

[0005] Step 1: Establish and maintain expert database:

[0006] Data collection: crawler technology and API interfaces are used to crawl experts’ personal information from the websites of major universities, including but not limited to name, title, affiliated institution, contact information, and academic achievements.

[0007] Data cleaning and integration: Clean the collected data, remove duplicate information, standardize the format, and integrate it into a unified database structure.

[0008] Continuous update: regularly perform data crawling tasks, update expert information, and ensure the timeliness and accuracy of the database.

[0009] Step 2: Build an expert field label system:

[0010] Information extraction: Extract key information from experts’ academic papers, including research directions, publication journals, co-authors, etc.

[0011] Labeling: Based on this information, a label system containing specific research fields is generated for each expert to facilitate subsequent expert matching.

[0012] Institutional affiliation: record the educational institution and department where the expert is employed to understand his / her position in the academic circle.

[0013] Step 3: Build an expert behavior and social dimension system:

[0014] Historical data collection: Collect the work records of each expert who participated in the blind review in the past, such as the probability of accepting the review, the review quality score, the time required to complete the review, etc.

[0015] Data analysis: Analyze the above data through statistical methods to form the behavior pattern of experts.

[0016] Social network analysis: Use social network analysis techniques to explore the collaboration network among experts and their influence.

[0017] Step 4: Match the paper keywords, calculate the similarity with the expert field tags, and select experts with higher similarity.

[0018] 4.1 Based on the basic information of the paper submitted for review, the paper structure data can be obtained:

[0019]

[0020] Among them, p_i represents the i-th paper in the paper library, title represents the title of the paper, subject is the subject to which the blind review paper belongs, and keywords are the keywords of the paper.

[0021] Basic structural data of the expert:

[0022]

[0023] Among them, e_i represents the i-th expert, subject represents the subject to which the expert belongs, direction represents the research direction of the expert, and keywords represents the keywords of the expert's published papers.

[0024] 4.2 According to the Sentence-Bert model, encode the subject of the matching paper and the subject of the candidate expert to obtain paper_major_embedding and expert_major_embedding, and perform similarity test on them:

[0025]

[0026] Among them, P and E are paper_major_embedding and expert_major_embedding vectors respectively, and the first-level discipline matching screening results are obtained.

[0027] 4.3 According to the first-level discipline screening results, the matching paper keywords, candidate expert research directions and published paper keywords are encoded according to the Sentence-Bert model to obtain paper_keyword_embedding and expert_keywords_embedding, and the similarity test is performed on them:

[0028]

[0029] Among them, P' and E' are paper_keyword_embedding and expert_keywords_embedding vectors respectively.

[0030] Thus, the secondary matching screening results are obtained, and the candidate experts are further screened.

[0031] Step 5: Comprehensive evaluation and ranking:

[0032] TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) is a multi-attribute decision-making method used to evaluate the comprehensive indicators of alternative solutions to find the optimal solution. It can help decision makers make decisions quickly and objectively while considering the influence of multiple factors. This paper adopts the entropy weight method-Topsis comprehensive evaluation method, and comprehensively evaluates candidate experts based on expert behavior and social dimension data, and obtains the optimal reviewer recommendation result.

[0033] Step 6: Blind review expert recommendation:

[0034] Results presentation: The experts with the highest final rankings will be recommended to organizations or individuals who need blind review services.

[0035] Feedback mechanism: Establish a feedback mechanism to collect users’ satisfaction with receiving recommendations and further optimize the recommendation algorithm.

[0036] Further, step 5 comprehensive evaluation and ranking

[0037] First, construct the candidate expert raw data matrix:

[0038] X ij =(X ij ) n×m

[0039] Among them, n is the number of candidate experts, m is the number of entropy weight method-Topsis evaluation indicators, including expert acceptance rate, quality of historical review books, review time, and historical review cycle.

[0040] The original data is dimensionless processed to make it comparable, and then the positive indicators are obtained:

[0041]

[0042] Negative indicators:

[0043]

[0044] Calculate the weight matrix of the indicator:

[0045]

[0046] Calculate the entropy value of each indicator:

[0047]

[0048] in The larger the entropy value, the more chaotic it is, which means it contains more information.

[0049] Calculate the coefficient of variance:

[0050] g j =1-e j

[0051] Calculate indicator weights:

[0052]

[0053] Vector normalization:

[0054]

[0055] Construct the weighting matrix:

[0056]

[0057] The weighted matrix results are as follows:

[0058]

[0059] The weighted matrix Z is obtained by entropy weight method * Finally, the Topsis comprehensive evaluation model was used to conduct a comprehensive evaluation.

[0060] First, normalize the weight matrix:

[0061]

[0062] Find the optimal and worst values ​​of each indicator and establish the optimal value vector z + and the worst value vector z - :

[0063]

[0064] Calculate each evaluation object and the optimal value and the worst value Distance between:

[0065]

[0066] Calculate the relative closeness of each evaluation index to the optimal value:

[0067]

[0068] Require The smaller the better, The bigger the better.

[0069] Finally, according to C i Sort by size, C i The larger it is, the closer the evaluation object is to the optimal value, and the higher the score is, the better the comprehensive evaluation result of the candidate expert is.

[0070] Furthermore, in step 3, based on the above expert information, the basic structure data of the expert can be obtained:

[0071]

[0072] Among them, e_i represents the i-th expert, subject represents the subject to which the expert belongs, direction represents the research direction of the expert, and keywords represents the keywords of the expert's published papers.

[0073] The expert behavior and social dimension data:

[0074]

[0075] Among them, t_i represents the i-th expert, a_1, a_2, a_3, and a_4 are the data of expert acceptance rate, quality of historical review books, review time, and historical review cycle, respectively.

[0076] Compared with the prior art, the present invention has the following significant advantages: the present invention constructs an expert behavior and social dimension system based on factors such as the expert acceptance probability, historical review book quality, review paper duration, and historical review cycle obtained in the historical expert matching business of the blind review system, and provides users with better initial recommendations through methods such as behavior analysis and evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is a flow chart of the thesis review expert recommendation method based on multi-graph neural network of the present invention. DETAILED DESCRIPTION

[0078] The present invention will be further described below in conjunction with the accompanying drawings.

[0079] 1. Establishment and maintenance of expert database:

[0080] Data collection: Using crawler technology and API interfaces, we crawl experts’ personal information from major university websites (such as Peking University Graduate Student Website, Nanjing University of Science and Technology Graduate Student Website, etc.), including but not limited to name, title, affiliated institution, contact information, and academic achievements.

[0081] Data cleaning and integration: Clean the collected data, remove duplicate information, standardize the format, and integrate it into a unified database structure.

[0082] Continuous updating: Regularly perform data crawling tasks, update expert information, and ensure the timeliness and accuracy of the database.

[0083] 2. Build an expert field label system:

[0084] Information extraction: Extract key information from experts’ academic papers, including research directions, publication journals, co-authors, etc.

[0085] Labeling: Based on this information, a label system containing specific research fields is generated for each expert to facilitate subsequent expert matching.

[0086] Institutional affiliation: Record the educational institution and department where the expert is affiliated to understand his or her position in the academic community.

[0087] 3. Build an expert behavior and social dimension system:

[0088] Historical data collection: Collect the work records of each expert who participated in the blind review in the past, such as the probability of accepting the review, the review quality score, the time required to complete the review, etc.

[0089] Data analysis: Analyze the above data through statistical methods to form the behavior pattern of experts.

[0090] Social network analysis: Use social network analysis techniques to explore the collaboration network among experts and their influence.

[0091] Based on the above expert information, the basic structure data of the expert can be obtained:

[0092]

[0093] Among them, e_i represents the i-th expert, subject represents the expert's discipline, direction represents the expert's research direction, and keywords are the keywords of the expert's published papers.

[0094] The expert behavior and social dimension data:

[0095]

[0096] Among them, t_i represents the i-th expert, a_1, a_2, a_3, and a_4 are the data of expert acceptance rate, quality of historical review books, review time, and historical review cycle, respectively.

[0097] 4. Paper keyword matching:

[0098] 4.1 Based on the basic information of the paper submitted for review, the paper structure data can be obtained:

[0099]

[0100] Among them, p_i represents the i-th paper in the paper library, title represents the title of the paper, subject is the subject to which the blind review paper belongs, and keywords are the keywords of the paper.

[0101] Basic structural data of the expert:

[0102]

[0103] Among them, e_i represents the i-th expert, subject represents the expert's discipline, direction represents the expert's research direction, and keywords are the keywords of the expert's published papers.

[0104] 4.2 According to the Sentence-Bert model, encode the subject of the matching paper and the subject of the candidate expert to obtain paper_major_embedding and expert_major_embedding, and perform similarity test on them:

[0105]

[0106] Among them, P and E are paper_major_embedding and expert_major_embedding vectors respectively.

[0107] Get the first-level discipline matching screening results.

[0108] 4.3 According to the first-level discipline screening results, the matching paper keywords, candidate expert research directions and published paper keywords are encoded according to the Sentence-Bert model to obtain paper_keyword_embedding and expert_keywords_embedding, and the similarity test is performed on them:

[0109]

[0110] Among them, P' and E' are paper_keyword_embedding and expert_keywords_embedding vectors respectively.

[0111] Thus, the secondary matching screening results are obtained, and the candidate experts are further screened.

[0112] 5. Comprehensive evaluation and ranking:

[0113] TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) is a multi-attribute decision-making method used to evaluate the comprehensive indicators of alternative solutions to find the optimal solution. It can help decision makers make decisions quickly and objectively while considering the influence of multiple factors. This paper adopts the entropy weight method-TOPSIS comprehensive evaluation method, and comprehensively evaluates candidate experts based on expert behavior and social dimension data, and obtains the optimal reviewer recommendation result.

[0114] First, construct the candidate expert raw data matrix:

[0115] X ij =(X ij ) n×m

[0116] Among them, n is the number of candidate experts, and m is the number of entropy weight method-Topsis evaluation indicators, including expert acceptance rate, quality of historical review books, review time, and historical review cycle.

[0117] The original data is dimensionless processed to make it comparable, and then the positive indicators are obtained:

[0118]

[0119] Calculate the weight matrix of the indicator:

[0120]

[0121] Calculate the entropy value of each indicator:

[0122]

[0123] in The larger the entropy value, the more chaotic it is, which means it contains more information.

[0124] Calculate the coefficient of variance:

[0125] g j =1-e j

[0126] Calculate indicator weights:

[0127]

[0128] Vector normalization:

[0129]

[0130] Construct the weighting matrix:

[0131]

[0132] The weighted matrix results are as follows:

[0133]

[0134] The weighted matrix Z is obtained by entropy weight method * Finally, the Topsis comprehensive evaluation model is used to conduct a comprehensive evaluation.

[0135] First, normalize the weight matrix:

[0136]

[0137] Find the optimal and worst values ​​of each indicator and establish the optimal value vector z + and the worst value vector z - :

[0138]

[0139] Calculate the distance between each evaluation object and the optimal value and the worst value:

[0140]

[0141] Calculate the relative closeness of each evaluation index to the optimal value:

[0142]

[0143] Require The smaller the better, The bigger the better.

[0144] Finally, according to C iSort by size, C i The larger it is, the closer the evaluation object is to the optimal value, and the higher the score is, the better the comprehensive evaluation result of the candidate expert is.

[0145] 6. Blind review expert recommendation:

[0146] Result presentation: The experts with the highest final ranking will be recommended to organizations or individuals in need of blind review services.

[0147] Feedback mechanism: Establish a feedback mechanism to collect users’ satisfaction with receiving recommendations and further optimize the recommendation algorithm.

[0148] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent recommendation method for blind review experts of degree theses based on expert portraits, characterized by: The following steps, Step 1: Establish and maintain expert database: Data collection: crawler technology and API interfaces are used to crawl experts’ personal information from the websites of major universities, including but not limited to name, title, affiliated institution, contact information, and academic achievements. Data cleaning and integration: Clean the collected data, remove duplicate information, standardize the format, and integrate it into a unified database structure. Continuous update: regularly perform data crawling tasks, update expert information, and ensure the timeliness and accuracy of the database. Step 2: Build an expert field label system: Information extraction: Extract key information from experts’ academic papers, including research directions, publication journals, co-authors, etc. Labeling: Based on this information, a label system containing specific research fields is generated for each expert to facilitate subsequent expert matching. Institutional affiliation: record the educational institution and department where the expert is employed to understand his / her position in the academic circle. Step 3: Build an expert behavior and social dimension system: Historical data collection: Collect the work records of each expert who participated in the blind review in the past, such as the probability of accepting the review, the review quality score, the time required to complete the review, etc. Data analysis: Analyze the above data through statistical methods to form the behavior pattern of experts. Social network analysis: Use social network analysis techniques to explore the collaboration network among experts and their influence. Step 4: Match the paper keywords, calculate the similarity with the expert field tags, and select experts with higher similarity. 4.1 Based on the basic information of the paper submitted for review, the paper structure data can be obtained: P=「p1,p2,p3,…,p m ],p i =(title,subject,keywords); Among them, p_i represents the i-th paper in the paper library, title represents the title of the paper, subject is the subject to which the blind review paper belongs, and keyWords is the keywords of the paper. Basic structural data of the expert: And=「e1,e2,....and n ],and i =(subject,direction,keywords); Among them, e_i represents the i-th expert, subject represents the subject to which the expert belongs, direction represents the research direction of the expert, and keyWords represents the keywords of the expert's published papers. 4.2 According to the Sentence-Bert model, encode the subject of the matching paper and the subject of the candidate expert to obtain paper_major_embedding and expert_major_embedding, and perform similarity test on them: Among them, P and E are paper_major_embedding and expert_major_embedding vectors respectively. Get the first-level discipline matching screening results, 4.3 According to the first-level discipline screening results, the matching paper keywords, candidate expert research directions and published paper keywords are encoded according to the Sentence-Bert model to obtain paper_keyword_embedding and expert_keywords_embedding, and the similarity test is performed on them: Among them, P' and E' are paper_keyword_embedding and expert_keywords_embedding vectors respectively. Thus, the secondary matching screening results are obtained, and the candidate experts are further screened. Step 5: Comprehensive evaluation and ranking: TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) is a multi-attribute decision-making method used to evaluate the comprehensive indicators of alternative solutions to find the optimal solution. It can help decision makers make decisions quickly and objectively while considering the influence of multiple factors. This paper adopts the entropy weight method-Topsis comprehensive evaluation method, and comprehensively evaluates candidate experts based on expert behavior and social dimension data, and obtains the optimal reviewer recommendation result. Step 6: Blind review expert recommendation: Results presentation: The experts with the highest final rankings will be recommended to organizations or individuals who need blind review services. Feedback mechanism: Establish a feedback mechanism to collect users’ satisfaction with receiving recommendations and further optimize the recommendation algorithm.

2. The method for recommending degree thesis review experts based on multi-graph neural network according to claim 1, characterized in that: Step 5 Comprehensive evaluation and ranking First, construct the candidate expert raw data matrix: X ij =(X ij ) n×m Among them, n is the number of candidate experts, m is the number of entropy weight method-Topsis evaluation indicators, including expert acceptance rate, quality of historical review books, review time, and historical review cycle. The original data is dimensionless processed to make it comparable, and then the positive indicators are obtained: Negative indicators: Calculate the weight matrix of the indicator: Calculate the entropy value of each indicator: in The larger the entropy value, the more chaotic it is, which means that it contains more information. Calculate the coefficient of variance: g j =1-e j Calculate indicator weights: Vector normalization: Construct the weighting matrix: The weighted matrix results are as follows: The weighted matrix Z is obtained by the entropy weight method * Finally, the Topsis comprehensive evaluation model was used to conduct a comprehensive evaluation. First, normalize the weight matrix: Find the optimal and worst values ​​of each indicator and establish the optimal value vector z + and the worst value vector z - : Calculate each evaluation object and the optimal value and the worst value Distance between: Calculate the relative closeness of each evaluation index to the optimal value: Require The smaller the better. The bigger the better. Finally, according to C i Sort by size, C i The larger it is, the closer the evaluation object is to the optimal value, and the higher the score is, the better the comprehensive evaluation result of the candidate expert is.

3. The method for recommending degree thesis review experts based on multi-graph neural network according to claim 1, characterized in that: In step 3, based on the above expert information, the basic structure data of the expert can be obtained: And=「e1,e2,...,e n ],and i =(subject,direction,keywords); Among them, e_i represents the i-th expert, subject represents the subject to which the expert belongs, direction represents the research direction of the expert, and keywords represents the keywords of the expert's published papers. The expert behavior and social dimension data: <h2 style=";text-align:left;direction:ltr">T=「t1,t2,t3,...,t<h2 style=";text-align:left;direction:ltr"> n <h2 style=";text-align:left;direction:ltr"> ],t<h2 style=";text-align:left;direction:ltr"> i <h2 style=";text-align:left;direction:ltr"> (a1, a2, a3, a4) Among them, t_i represents the i-th expert, a_1, a_2, a_3, and a_4 are the data of expert acceptance rate, quality of historical review books, review time, and historical review cycle, respectively.

Images