CN109919463A - Technical journal manuscript quality evaluation system based on SVM learning model - Google Patents

Abstract

The invention discloses a kind of step 1 the technical journal manuscript quality evaluation system based on SVM learning model obtains the expert opinion index of several experts；Step 2, the evaluation index of each expert is established following collection resultant: step 3, using several expert's set C as training sample, input obtains SVM classifier equipped in the computer of SVM learning model；Step 4, the SVM classifier obtained according to step 3, determines whether the contribution employs；Greatly reduce the interference of human factor in refereeing procedure in this way, the standard strictly accepted or rejected using Article quality as contribution, meanwhile, also improve the conscientious degree of going over a manuscript or draft of peer-reviewers, and shortening is gone over a manuscript or draft the period, and the quality of technical journal is improved.

Description

Technical journal manuscript quality evaluation system based on SVM learning model

Technical field

The present invention relates to computer application field of engineering technology more particularly to technical journal manuscript quality evaluation systems.

Background technique

Technical journal is to deliver based on the academic paper of systematicness, technicality, creativeness, with reflection high level, high-quality The periodical that the research and teaching achievement of amount is attached most importance to.One vital task of Editors in Charge of Journals of Natural Sciences is exactly to the total of technical journal Weight is effectively controlled, and most effective approach is exactly specialist reading and expert opinion.Manuscript is examined by expert It reads, the manuscript of high quality is filtered out for periodical.Expert assessment method is that occur relatively early and apply a kind of wider evaluation method.Expert Evaluation assessment is exactly to select evaluation index according to the concrete condition of evaluation object, makes evaluation to each index respectively by several experts Grade, ranking score preferably, preferably, generally, poor four class, and being indicated with score value 100,80,60,0；Traditional refereeing procedure is It is calculated through artificial rough Statistics, total score reaches the contribution of some score value or more, is determined as the contribution that can be published.This tradition side Method often by the interference of human factor, makes opinion rating not meet actual state, influences quality of going over a manuscript or draft；Simultaneously by largely, examine The efficiency of original text is also low.

The present invention is related with SVM.The entitled support vector machines of SVM (Support Vector Machine) Chinese, is common A kind of method of discrimination.It is that data are dug commonly used to carry out pattern-recognition, classification and regression analysis in machine learning field The classical learning model for having supervision in pick field.Make one below simply to introduce:

Svm classifier problem is described as follows: given sample training collection,, whereinIt is defeated Enter indicator vector,It is output-index, determinesOn a real-valued function, so as to any input, all can be by decision functionIt is inferred to its corresponding output, whereinFor symbol letter Number, etc..

Classification problem can be described as follows: givenA classification based training sample, Wherein And, it is based on one classification function of above-mentioned sample architecture。 There is certain corresponding relationship between more classification and two classification problems: ifClassification problem can divide completely, thenAny two class in class Can centainly it divide；Conversely, if can divide between its any two class, by certain combination or ballot rule, can by two all right one way or the other points Lai It is final to realizeClass can divide.

The basic thought of SVM multi-classification algorithm based on binary tree is that all categories are divided into two subclasses, then by subclass Two secondary subclasses are divided into, until to be performed repeatedly until all nodes only include an individual classification.This method will be former Some multi-class problems have equally resolved into a series of two classes classification problem, and the classification function between two of them subclass uses SVM.

Currently, there are no people to establish the evaluation system based on SVM learning model in terms of technical journal manuscript quality evaluation.

Summary of the invention

In order to improve quality of going over a manuscript or draft, separates expert opinion index with the evaluation result of comprehensive evaluation index, avoid going over a manuscript or draft As a result it is influenced by other factors, while improving the efficiency and convenience gone over a manuscript or draft, realize the shared of long-distance draft reviewing and Internet resources, It is proposed a kind of technical journal manuscript quality evaluation system based on SVM learning model.

The present invention uses following technical scheme to solve above-mentioned technical problem:

A kind of technical journal manuscript quality evaluation system based on SVM learning model, specifically includes the following steps:

Step 1, obtain several experts'；The expert opinion index is the evaluation obtained according to traditional expert assessment method Index, including 6 political nature, ideological content, novelty, academic nature, science and practicability evaluation contents, each evaluation index point Preferably, preferably, generally, 4 opinion ratings such as difference, assign 4 opinion ratings of 6 evaluation contents to corresponding qualitative fuzzy comment The qualitative fuzzy evaluation magnitude of valence magnitude, 4 opinion ratings is respectively 100,80,60,0；

Step 2, the evaluation index of each expert is established into following collection resultant:

A={ good, preferably, generally, poor };B={ 100,80,60,0 };C=(a₁, a₂, a₃, a₄, a₅, a_6, b)

Wherein, set A is the fuzzy set of evaluation index, and set B is quantization set corresponding with set A, and set C is According to the vector that evaluation index obtains journal appraising, wherein a₁, a₂, a₃, a₄, a₅, a₆For six evaluation index quantized values, b is Evaluate the variable after classification quantifies comprehensive evaluation index；

Step 3, using several expert's set C as training sample, equipped in the computer of SVM learning model, computer is pressed for input It is trained, examines according to SVM learning model, obtain SVM classifier；

Step 4, contribution employs decision:

The SVM classifier obtained according to step 3, determines whether the contribution employs, specifically:

Four sorting algorithms based on binary tree, first category training dataset be divided into two subset SetA (80,100) and SetB (60) assigns label+1 to the data of SetA, and the data of SetB assign label -1, with this two classes data configuration classification function SVM1 further executes identical step to the training dataset in SetA, obtains another classification function SVM2；

The sample new for one, is classified with SVM1: if result is 1, showing that the sample may belong to 100,80 One of class is then classified with SVM2, if result is -1, result 60；If result when being classified with SVM2 It is 1, then shows that the sample class is 100, be otherwise 80；100 and 80 be respectively that high-quality and quality preferably evaluates magnitude；It is logical The above calculating is crossed, the manuscript of high quality is filtered out.

The positive effect of the present invention is, in Sci-tech Periodical Editorial Department door refereeing procedure, by the Article quality evaluation system It is combined with periodical platform of gathering and editing, as long as go over a manuscript or draft index system and corresponding factor of evaluation that peer-reviewers are provided according to editor Corresponding value is chosen, editorial department can input computer, pass through model meter according to the reviewers' commentss of several peer-reviewers It calculates, provides desired go over a manuscript or draft as a result, such interference for greatly reducing human factor in refereeing procedure, is strictly made with Article quality For the standard that contribution is accepted or rejected, meanwhile, it also improves the conscientious degree of going over a manuscript or draft of peer-reviewers, and shorten and go over a manuscript or draft the period, improves science and technology The quality of periodical.

Detailed description of the invention

Fig. 1 is the four sorting algorithm schematic diagrames based on binary tree.

Specific embodiment

Below by taking " X X college journal " as an example, method of the invention is further illustrated.

So-and-so college journal periodical guarantee, reaching following quality requirement just to the contribution received can publish, ideological content and Political nature will reach higher level or more, and academic, science will reach preferable level, and innovation and practicality will reach one As it is horizontal or more.In order to improve quality of going over a manuscript or draft, separates expert opinion index with the evaluation result of comprehensive evaluation index, avoid Result of going over a manuscript or draft is influenced by other factors, while improving the efficiency and convenience gone over a manuscript or draft, realizes long-distance draft reviewing and Internet resources It is shared, using following technical scheme:

Step 1,

Obtain the evaluation index of several experts；The expert opinion index, be obtained according to traditional expert assessment method it is every The evaluation index of a expert, including 6 political nature, ideological content, novelty, academic nature, science and practicability evaluation contents, often A evaluation index is divided into, preferably, generally, difference etc., 4 opinion ratings of 6 evaluation contents are assigned corresponding qualitative fuzzy Evaluation magnitude, respectively 100,80,60,0；As shown in table 1,4 evaluations are selected in 6 evaluation contents of each expert in table 1 A certain magnitude in grade.

1 evaluation index quantized value of table

Step 2, the evaluation table of each expert is established into following collection resultant:

C=(a₁, a₂, a₃, a₄, a₅, a_6, b)

Wherein, set A is the fuzzy set of evaluation index, and set B is quantization set corresponding with set A, and set C is According to the vector that evaluation index obtains journal appraising, wherein a₁, a₂, a₃, a₄, a₅, a₆For six evaluation index quantized values, b is Evaluate the variable after classification quantifies comprehensive evaluation index；To a₁, a₂, a₃, a₄, a₅, a₆According to trained vector machine model Determine weight；

Step 3,

From the set C of several peer-reviewers of step 2, that chooses the basic publication requirement of 100 satisfactions is used as Article quality data Sample (100 different paper samples of selection as shown in Table 2), input are equipped in the computer of SVM learning model, computer It is trained, examines according to SVM learning model, obtain SVM classifier；Program is write with matlab7.10 language, experiment porch Pentium (R), 2G RAM, operating system are Windows XP；

2 training sample of table

When training, after randomly selecting in 100 data samples, and using remaining sample as training sample, two classifiers are The tool box Libsvm, the classification of SVM bis- use gaussian kernel function；

Step 4, contribution employs decision:

The SVM classifier obtained according to step 3, determines whether the contribution employs, specifically:

Four sorting algorithms based on binary tree, first category training dataset be divided into two subset SetA (80,100) and SetB (60) assigns label+1 to the data of SetA, and the data of SetB assign label -1, with this two classes data configuration classification function SVM1 further executes identical step to the training dataset in SetA, obtains another classification function SVM2, such as Fig. 1 It is shown.

The sample new for one, is classified with SVM1: if result is 1, show that the sample may belong to 100, One of 80 classes, then classified with SVM2, if result is -1, result 60；If tied when being classified with SVM2 Fruit is 1, then shows that the sample class is 100, be otherwise 80；100 and 80 be respectively that high-quality and quality preferably evaluates magnitude； By calculating above, the manuscript of high quality is filtered out.

As shown in table 3, the information and parameter selection information of data set are listed.

Table 3:

Table 4 gives the data result of 10 groups of random tests, and the classification accuracy of 10 groups of data concentrates on as can be seen from Table 4 90% or so, it is minimum to be also not less than 87.5%, reach as high as 97.5%.Therefore, the binary tree multi-classification algorithm based on SVM is for original text Part quality evaluation problem, binary tree multi-classification algorithm are relatively effective.

Table 4:

Claims (1)

1. a kind of technical journal manuscript quality evaluation system based on SVM learning model, specifically includes the following steps:

Step 1, obtain several experts'；The expert opinion index is the evaluation obtained according to traditional expert assessment method Index, including 6 political nature, ideological content, novelty, academic nature, science and practicability evaluation contents, each evaluation index point Preferably, preferably, generally, 4 opinion ratings such as difference, assign 4 opinion ratings of 6 evaluation contents to corresponding qualitative fuzzy comment The qualitative fuzzy evaluation magnitude of valence magnitude, 4 opinion ratings is respectively 100,80,60,0；

Step 2, the evaluation index of each expert is established into following collection resultant:

A={ good, preferably, generally, poor };B={ 100,80,60,0 };C=(a₁, a₂, a₃, a₄, a₅, a_6, b)；

Wherein, set A is the fuzzy set of evaluation index, and set B is quantization set corresponding with set A, and set C is According to the vector that evaluation index obtains journal appraising, wherein a₁, a₂, a₃, a₄, a₅, a₆For six evaluation index quantized values, b is Evaluate the variable after classification quantifies comprehensive evaluation index；

Step 3, using several expert's set C as training sample, equipped in the computer of SVM learning model, computer is pressed for input It is trained, examines according to SVM learning model, obtain SVM classifier；

Step 4, contribution employs decision:

The SVM classifier obtained according to step 3, determines whether the contribution employs, specifically:

Four sorting algorithms based on binary tree, first category training dataset be divided into two subset SetA (80,100) and SetB (60) assigns label+1 to the data of SetA, and the data of SetB assign label -1, with this two classes data configuration classification function SVM1 further executes identical step to the training dataset in SetA, obtains another classification function SVM2；

The sample new for one, is classified with SVM1: if result is 1, showing that the sample may belong to 100,80 One of class is then classified with SVM2, if result is -1, result 60；If result when being classified with SVM2 It is 1, then shows that the sample class is 100, be otherwise 80；100 and 80 be respectively that high-quality and quality preferably evaluates magnitude；It is logical The above calculating is crossed, the manuscript of high quality is filtered out.