A method for intelligently recommending reviewers to a system to prevent cheating

Abstract

The present invention discloses a method for intelligently recommending reviewers to a system to prevent cheating, including obtaining relevant information of experts, combing the obtained relevant information of experts to build an expert information base; analyzing the data in the expert information base to build an expert knowledge map; retrieving experts based on the expert knowledge map and the information of the documents to be reviewed and the authors, and sorting the retrieved experts according to the institutional demand parameters, and taking the top several experts to recommend to the system. On the one hand, this method enables the system to consider not only the research direction of the experts but also the research depth and breadth of the experts when recommending reviewers; on the other hand, it further explores the relationship between experts and authors, making the manuscript review more fair and just, and the recommendation indicators and calculation parameters can be flexibly set according to the needs of the institution.

Description

Technical Field

[0001] The present invention relates to the technical field of recommendation systems, and in particular to a method for intelligently recommending reviewers to a system to prevent cheating. Background Art

[0002] The process that a document needs to go through from the author writing to the institution publishing includes the author writing (papers, patents, conference manuscripts, etc.), selecting the submission institution (journals, conferences, patent offices, etc.), institution review, institution acceptance, and institution publication. Among them, institution review is the key factor in determining whether a document can be published. Institutional review means that the institution selects appropriate reviewers based on the submitted document, and finally determines whether to accept the document based on the review results of the reviewer. The document will only be published after the institution accepts it. In the whole process, the reviewer plays a key role in whether the document is suitable for publication in the submission institution. For the institution, the ideal reviewer is one who gives the value of the paper from a professional perspective in a fair and impartial manner. The usual methods of recommending reviewers for documents are:

[0003] 1) Based on document content matching: match submitted documents with expert documents, and recommend experts corresponding to the matched documents to the system;

[0004] 2) Based on the matching of literature and reviewers: extract the key information of literature and reviewers (obtained from the scholar information network, scholars' articles and review literature), search and match based on the key information, and then recommend the reviewers corresponding to the matched key information to the system;

[0005] The above methods all recommend reviewers based on the relevance of research directions. Not only do they ignore the requirement of fairness and impartiality for reviewers, but the relevance of research directions also fails to reflect the difference between major peers and minor peers, which is not smart enough. Summary of the invention

[0006] In order to solve the above technical problems, the purpose of the present invention is to provide a method for intelligently recommending reviewers to a system to prevent cheating.

[0007] The purpose of the present invention is achieved through the following technical solutions:

[0008] A method for intelligently recommending reviewers to a system to prevent cheating, comprising:

[0009] Obtain relevant information about experts, sort out the obtained relevant information about experts and build an expert information database;

[0010] Analyze the data in the expert information database and build an expert knowledge graph;

[0011] Experts are retrieved based on the expert knowledge graph, pending documents and author information, and the retrieved experts are ranked according to the institution's demand parameters, and the top several experts are recommended to the system.

[0012] Compared with the prior art, one or more embodiments of the present invention may have the following advantages:

[0013] On the one hand, this method enables the system to consider not only the expert's research direction but also the expert's research depth and breadth when recommending reviewers; on the other hand, it further explores the relationship between experts and authors, making manuscript review more fair and just, and the recommendation indicators and calculation parameters can be flexibly set according to the needs of the institution. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the method of intelligently recommending reviewers to the system to prevent cheating;

[0015] Figure 2 It is an expert knowledge graph. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below in conjunction with embodiments and drawings.

[0017] like Figure 1 As shown, the method for intelligently recommending reviewers to the system to prevent cheating includes:

[0018] 1) Obtain relevant information about experts, sort out the obtained information about experts and build an expert information database;

[0019] The expert information database sorts out expert information from various databases, including first obtaining the expert's default research direction, graduation school and major, work experience, and family member information from the expert database and talent database, then obtaining the expert's published documents from the HowNet database, and then obtaining the expert's historical review manuscripts and expert-under-review documents (referring to the documents submitted by experts that are awaiting review) from the editing and compiling database. Finally, the expert information database ExpertDataBase is constructed based on the obtained expert-related information. The expert information database is in a state of continuous updating.

[0020] 2) Analyze the data in the expert information database and construct an expert knowledge graph;

[0021] By analyzing the data in ExpertDataBase, we build the expert knowledge graph Expert-KG. As the expert information base is updated, the expert knowledge graph is also in a state of continuous updating. The construction process is as follows:

[0022] Analyze the experts' published literature, historical review manuscripts, default research directions, and graduate school majors to obtain the words / sequences representing the experts' research directions, and obtain the triples of the experts' research directions: (expert, research, word / sequence);

[0023] Analyze the author information of the experts' published papers to obtain the expert triples with cooperative relationships: (expert, cooperation, expert);

[0024] According to (expert, research, word / sequence) and (expert, cooperation, expert), we get the triples: (expert, research direction includes, expert), (expert, research direction is consistent, expert);

[0025] The experts’ graduate schools and work experience were analyzed to obtain the triples: (expert, affiliation (past and present), unit / school / institution);

[0026] Analyze the expert’s family member information to obtain the triples: (expert, family relationship, person), (person, subordinate, unit / school / institution);

[0027] Analyze the experts' under-reviewed documents to obtain the triple: (expert, under-review, document feature set);

[0028] According to the triples, we can construct Figure 2 Expert knowledge graph Expert-KG.

[0029] 3) Retrieve experts based on the expert knowledge graph, pending documents and author information, rank the retrieved experts according to the institution’s demand parameters, and recommend the top experts to the system.

[0030] The parameters related to institutional needs include: the relevance between the literature to be reviewed and the expert's research direction, the breadth of expert research, the depth of expert research, and the relevance between the author of the literature to be reviewed and the expert. The specific search and sorting process is as follows:

[0031] (1) Calculate the correlation vector sims_doc_experts_vector between the literature to be reviewed and the expert research direction:

[0032] (1.1) Extract keywords / sequences from the documents to be reviewed: Extract the top-n keywords / sequences keywords_seqs_list by extracting features from the documents to be reviewed and reducing the semantic dimension of the features. The default value of n is 5 and can be adjusted according to the document type.

[0033] (1.2) Calculate the semantic matching between the extracted keywords_seqs_list and the "words / sequences" in the expert knowledge graph Expert-KG, and then calculate the correlation between the literature and the expert research direction based on the "research" relationship between the "words / sequences" and "experts" in Expert-KG: sims_doc_experts_vector = [sim_0, sim_1,......., sim_j,......, sim_n-1], where sim_j is the similarity between the literature to be reviewed and the research direction of the jth expert, and there are n experts in total.

[0034] (2) Calculate the expert research breadth vector breadth_experts_vector: calculate the number of research directions based on the expert research direction;

[0035] (2.1) Calculate the "experts" with the "research direction inclusion" relationship starting from the "expert i" node in Expert-KG

[0036] experts_from_expert_i=[expert_10,...,expert_j],j≠i;

[0037] (2.2) Reduce the dimensionality of experts in experts_from_expert_i and merge experts with "same research direction";

[0038] (2.3) Calculate the number of experts in experts_from_expert_i, which is the research breadth b_i of expert_i. If we calculate "expert 1" (see Figure 2 )’s research breadth, b_1=3

[0039] (2.4)breadth_experts_vector=[b_0,b_1,......,b_j,......,b_n-1], b_j is the research breadth of the jth expert, breadth_experts_vector is normalized, and b_j is normalized to a value between 0 and 1.

[0040] (3) Calculate the expert research depth vector depth_experts_vector:

[0041] (3.1) Calculate the number of layers d_i (exp: Calculate "Expert 5" (see Figure 2 )’s research depth d_5=2, counting from 0);

[0042] (3.2)depth_experts_vector=[d_0,d_1,......,d_j,......,d_n-1], d_j is the research depth of the jth expert, depth_experts_vector is normalized, and d_j is normalized to a value between 0 and 1.

[0043] (4) Calculate the correlation attribute value vector Correlation_vector between the author and the expert of the document to be reviewed: Calculate from the following aspects

[0044] (4.1) Are the expert and the author the same person? j,0 ={0|1}, 0 represents no, 1 represents yes;

[0045] (4.2) Whether the literature under review by the expert is similar or almost the same as the literature to be reviewed, c j,1 ={0|1}, 0 represents no, 1 represents yes;

[0046] (4.3) Have the expert and the author published literature together? j,2 ={0|1}, 0 represents no, 1 represents yes;

[0047] (4.4) Are the expert and the author affiliated with the same school / unit / institution? j,3 ={0|1}, 0 represents no, 1 represents yes;

[0048] (4.5) Do the expert and the author have a family relationship, or do the expert's family members and the author belong to the same school / unit / institution? j,4 ={0|1}, 0 represents no, 1 represents yes;

[0049] (4.6)c j =1, if w0c j,0 +w1c j,1 +w2c j,2 +w3c j,3 +w4c j,4 =0

[0050] c j =0, if w0c j,0 +w1c j,1 +w2c j,2 +w3c j,3 +w4c j,4 >0

[0051] in:

[0052] w0=0|1, by default w0=1, if set to 0, the system will prompt the existence of the relationship

[0053] w1=0|1, by default w1=1, if set to 0, the system will prompt the existence of the relationship

[0054] w2=0|1, by default w2=1, if set to 0, the system will prompt the existence of the relationship

[0055] w3=0|1, by default w3=1, if set to 0, the system will prompt the existence of the relationship

[0056] w4=0|1, by default w4=1, if set to 0, the system will prompt the existence of the relationship

[0057] w0, w1, w2, w3, w4 are independent of each other and can be set separately;

[0058] (4.7)Correlation_vector=[c0,c1,...,c j ,.......,c n-1 ], c j is the correlation between the author and the jth expert, with a total of n experts.

[0059] (5) The expert recommendation index vector is F recommend :

[0060] F recommend =(w0X+w1Y+w2Z)*Correlation_vector=[f0,f1,......,f j ,......,f n-1 ]

[0061] in:

[0062] X = sims_doc_experts_vector, 0≤w0≤1 is the importance coefficient, adjustable;

[0063] Y = breadth_experts_vector, 0≤w1≤1 is the importance coefficient, adjustable;

[0064] Y = depth_experts_vector, 0≤w2≤1 is the importance coefficient, adjustable;

[0065] 0≤w0+w1+w2≤1

[0066] * indicates the multiplication of corresponding elements of two vectors, f j is the recommendation index of the jth expert, and there are n experts in total.

[0067] (6) For F recommend Sort in reverse order, take the top-n experts to recommend to the system, and prompt the system for the existing relationship.

[0068] The specific implementation is as follows:

[0069] Task: Author A submitted a paper on "How to solve the nested problem in sequence annotation" to an institution. Now recommend a suitable reviewer for this paper. The process is as follows:

[0070] Build an expert information database: The expert information database is established in advance and is in a state of continuous updating.

[0071] Constructing an expert knowledge graph: An expert knowledge graph is constructed in advance based on the expert information database and is continuously updated as the expert information database is updated. Here, we assume that there are only five experts. Figure 2 The "word / sequence" is instantiated as follows:

[0072] Recommend reviewers to the system:

[0073] (1) Calculate the correlation vector between the literature to be reviewed and the expert research direction sims_doc_experts_vector = [0.90, 0.42, 0.42, 0.95, 0.99];

[0074] (2) Calculate the expert research breadth vector breadth_experts_vector = [3, 0, 0, 1, 0], normalize (divide by the maximum value 3, other methods can also be used) breadth_experts_vector = [1.0, 0.0, 0.0, 0.33, 0.0];

[0075] (3) Calculate the expert research depth vector depth_experts_vector = [0, 1, 1, 1, 2], normalize (divide by the maximum value 2, or other methods can be used) depth_experts_vector = [0.0, 0.5, 0.5, 0.5, 1.0];

[0076] (4) Calculate the correlation attribute value vector Correlation_vector between the author and the expert of the document to be reviewed: Here we assume the following:

[0077] The author is not one of the 5 experts in the atlas;

[0078] The author's pending literature is similar or almost identical to the literature under review by expert 1, c 0,1 =1

[0079] The author has co-published papers with expert 4 3,2 =1;

[0080] The author and expert 3 belong to the same school / unit / institution, c 2,3 =1;

[0081] The author has no family relationship with any expert, nor does the author belong to the same school / unit / institution as any family member of any expert;

[0082] c j =1, if w0c j,0 +w1c j,1 +w2c j,2 +w3c j,3 +w4c j,4 =0

[0083] c j =0, if w0c j,0 +w1c j,1 +w2c j,2 +w3c j,3 +w4c j,4 >0

[0084] Assuming that the user only filters out experts with three relationships i, ii, and iii, the parameters are set as follows

[0085] w0=1,w1=1,w2=1,w3=0,w4=0

[0086] Correlation_vector=[0, 1, 0, 0, 1]

[0087] (5) Calculate the expert recommendation index vector F recommend :

[0088] Here we assume that the user's parameter settings are: w0 = 0.5, w1 = 0.3, w2 = 0.2

[0089] F recommend =(w0X+w1Y+w2Z)*Correlation_vector=[0.0, 0.31, 0.0, 0.0, 0.695]

[0090] (6) For F recommend Reverse sorting, taking the top-n experts to recommend to the system, and prompting the system about the existing relationship: Here, the top-1 expert is taken to the system, that is, expert 5;

[0091] Although the embodiments disclosed in the present invention are as above, the above contents are only embodiments adopted for facilitating the understanding of the present invention and are not intended to limit the present invention. Any technician in the technical field to which the present invention belongs can make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in the present invention, but the patent protection scope of the present invention shall still be subject to the scope defined in the attached claims.

Claims

1. A method for intelligently recommending reviewers to a system to prevent cheating, characterized in that: The method comprises: A. Obtain relevant information of experts and build an expert information database based on the obtained relevant information of experts; B. Analyze the data in the expert information database and build an expert knowledge graph; C. Retrieve experts based on the expert knowledge graph, pending documents, and author information, and rank the retrieved experts according to the institution’s demand parameters, and select the top experts to recommend to the system; In A, the default research direction, graduate school and major, work experience, and family member information of the expert are obtained from the expert database and talent database, and then the published documents of the expert are obtained from the CNKI database, and the historical review manuscripts and the documents under review by the expert are obtained from the editing database. Finally, the expert information database ExpertDataBase is constructed based on the obtained expert-related information. In B, the expert knowledge graph is continuously updated as the expert information database is updated; the construction process of the expert knowledge graph includes: b1) Analyze the expert's published literature, historical review manuscripts, default research direction, and graduate school major to obtain the word / sequence representing the expert's research direction, and obtain the triple of the expert's research direction: "expert", "research", "word / sequence"; b2) Analyze the author information of the experts’ published papers to obtain the expert triples with cooperative relationships: “expert”, “cooperation”, “expert”; b3) According to the triples of "expert", "research", "word / sequence" and "expert", "cooperation", "expert", we get the triples of "expert", "research direction includes", "expert" and the triples of "expert", "research direction consistent", "expert"; b4) Analyze the experts’ graduation schools and work experience to obtain the triples: “expert”, “affiliation”, “unit / school / institution”; b5) analyzing the expert's family member information to obtain the "expert", "family relationship", "person" triples and the "person", "subordinate", "unit / institution" triples; b6) Analyze the experts' documents under review to obtain the triple of "expert", "under review", and "document feature set"; b7) Construct an expert knowledge graph Expert-KG based on all the triples obtained; The institutional demand parameters in C include the relevance between the literature to be reviewed and the expert's research direction, the breadth of expert research, the depth of expert research, and the relevance between the author of the literature to be reviewed and the expert; The sorting process in C includes: c1) Calculate the correlation vector between the literature to be reviewed and the research direction of experts; c2) Calculate the expert research breadth vector, that is, calculate the number of research directions based on the expert research direction; c3) calculating expert research depth vector; c4) Calculate the relevant attribute value vectors of the authors and experts of the documents to be reviewed; c5) calculating the expert recommendation index vector; c6) Reversely sort the expert recommendation index vector, take the first several experts to recommend to the system, and prompt the system for the existing relationship.

2. The method for intelligently recommending reviewers to a system for preventing cheating according to claim 1, characterized in that: The c1 includes: 1) Extract keywords / sequences of the documents to be reviewed by extracting features and reducing feature semantics; 2) Calculate the semantic matching between the extracted keywords / sequences and the words / sequences in the expert knowledge graph, and then calculate the relevance of the literature and the expert research direction sims_doc_experts_vector based on the "research" relationship between "words / sequences" and "experts" in the expert knowledge graph; 3)sims_doc_experts_vector=[sim_0,sim_1,.......,sim_j,......,sim_n-1], sim_j is the similarity between the document to be reviewed and the research direction of the jth expert, and there are n experts in total.

3. The method for intelligently recommending reviewers to a system for preventing cheating according to claim 1, characterized in that: The c2 calculates the number of research directions based on the expert research direction; including: 1) Calculate the "experts" with the "research direction includes" relationship from the "expert i" node in the expert knowledge graph: experts_from_expert_i = [expert_10,......,expert_j], j≠i; 2) Reduce the dimensionality of experts in experts_from_expert_i and merge experts with "same research direction"; 3) Calculate the number of experts in experts_from_expert_i, which is the research breadth b_i of expert_i; 4) breadth_experts_vector = [b_0, b_1, ..., b_j, ..., b_n-1], b_j is the research breadth of the jth expert, breadth_experts_vector is normalized, and b_j is normalized to a value between 0 and 1.

4. The method for intelligently recommending reviewers to a system for preventing cheating according to claim 1, characterized in that: The c3 includes: 1) Calculate the number of layers included in an expert's research direction; 2) depth_experts_vector = [d_0, d_1, ..., d_j, ..., d_n-1], d_j is the research depth of the jth expert, depth_experts_vector is normalized, and d_j is normalized to a value between 0 and 1.

5. The method for intelligently recommending reviewers to a system for preventing cheating according to claim 1, characterized in that: The c4 is calculated by the following aspects: 1) Whether the expert and the author are the same person, c j,0 ={0|1}, 0 represents no, 1 represents yes; 2) Whether the literature under review by the experts is similar or almost the same as the literature to be reviewed, c j,1 ={0|1}, 0 represents no, 1 represents yes; 3) Whether the expert and the author have published literature together, j,2 ={0|1}, 0 represents no, 1 represents yes; 4) Whether the expert and the author belong to the same school / unit / institution, j,3 ={0|1}, 0 represents no, 1 represents yes; 5) Whether the expert and the author have a family relationship, or whether the expert's family members and the author belong to the same school / unit / institution j,4 ={0|1}, 0 represents no, 1 represents yes; 6)c j =1, if w0c j,0 +w1c j,1 +w2c j,2 +w3c j,3 +w4c j,4 =0 c j =0, if w0c j,0 +w1c j,1 +w2c j,2 +w3c j,3 +w4c j,4 >0 in: w0=0|1, by default w0=1, if set to 0, the system will prompt the existence of the relationship w1=0|1, by default w1=1, if set to 0, the system will prompt the existence of the relationship w2=0|1, by default w2=1, if set to 0, the system will prompt the existence of the relationship w3=0|1, by default w3=1, if set to 0, the system will prompt the existence of the relationship w4=0|1, by default w4=1, if set to 0, the system will prompt the existence of the relationship w0, w1, w2, w3, w4 are independent of each other and can be set separately; 7)Correlation_vector=[c0,c1,...,c j ,.......,c n-1 ], c j is the correlation between the author and the jth expert, with a total of n experts.

6. The method for intelligently recommending reviewers to a system for preventing cheating according to claim 1, characterized in that: In c5, the expert recommendation index vector is F recommend : F recommend =(w0X+w1Y+w2Z)*Correlation_vector=[f0,f1,......,f j ,......,f n-1 ] in: X = sims_doc_experts_vector, 0≤w0≤1 is the importance coefficient, adjustable; Y = breadth_experts_vector, 0≤w1≤1 is the importance coefficient, adjustable; Y = depth_experts_vector, 0≤w2≤1 is the importance coefficient, adjustable; 0≤w0+w1+w2≤1 * indicates the multiplication of corresponding elements of two vectors, f j is the recommendation index of the jth expert, and there are n experts in total.

Images