CN106484839A - A kind of journal impact appraisal procedure based on academic big data - Google Patents

Abstract

The invention discloses a kind of journal impact appraisal procedure based on academic big data, represent the power of influence of this periodical using the average influence power of all papers on each periodical.The method calculating paper impact factor is a kind of algorithm based on PageRank：AR PageRank, this algorithm considers the factors such as similarity between paper and the paper quoting it, author's H index, the former makes this algorithm become the algorithm of " weighting ", during algorithm iteration, every paper, in paper " ballot " that it is quoted, can give different contributions because of the similarity between paper；The latter makes " importance " of every paper to greatly increase, by the H index adding and being calculated every paper of author's H index, this index is as a label of this paper, the diversity making every paper gradually displays in calculating process, so as to relatively accurately distinguish the power of influence of every paper, finally obtain the power of influence of each periodical.

Description

A kind of journal impact appraisal procedure based on academic big data

Technical field

The present invention relates to the method based on academic big data, journal impact being estimated in sphere of learning, more particularly, to A kind of journal impact appraisal procedure that PageRank algorithm is combined with H index.

Background technology

With scientific and technological continuous development, the number being engaged in scientific research gets more and more, and their achievement in research is generally to learn The forms such as art paper, patent, books present, and the scientific research personnel of substantial amounts creates hundreds of millions of data.These data are united Referred to as academic big data.Academic big data of many uses, journal impact assessment be academic big data an important application. The power of influence of periodical is to be basic with the academic level of periodical, Academic Characteristics, is embodied with social credit worthiness by mark A kind of comprehensive effect, the quality of periodical decides the power of influence of periodical.The adopted method of journal impact assessment at present is all Analyzed, including Impact Factor (factor of influence), Eigenfactor (feature based on quoting (citation-based) The factor), PageRank scheduling algorithm.But above-mentioned algorithm there is also, and evaluation index is excessively single, cross-cutting cannot assess periodical shadow The problems such as power of sound.

Content of the invention

In place of some shortcomings mainly for above-mentioned existing research for the purpose of the present invention, proposed the phase based on academic big data Periodical power of influence appraisal procedure, by being analyzed to the reference information of paper on periodical, the PageRank proposing a kind of weighting calculates Method AR-PageRank algorithm, this algorithm considers the list of references registration of two papers, is characterized by this registration The similarity of this two papers, has been simultaneously introduced and author's H index has been considered, and author's H index of a paper is to a certain extent Represent the power of influence of this paper.

Technical scheme：

A kind of journal impact appraisal procedure based on academic big data, step is as follows：

1) the H index of Authors of Science Articles is acted on the assessment of journal article power of influence as influence factor；

2) similarity of article is calculated by comparing " registration " of the list of references of two papers；

3) pass through analyze paper citation network feature, PageRank algorithm is improved, by 1) in H index and 2) In " registration " conduct " weight " element be attached in PageRank algorithm it follows that AR-PageRank algorithm, calculate every The AR-PageRank value of piece paper；

4) the AR-PageRank value of papers all in periodical is summed up, obtain the assessed value of periodical.

Step 1)：H index is an index of author's aspect, it be a scholar or scientist the quantity of publication and The tolerance of power of influence aspect, H index can relatively accurately reflect the academic achievement of a scholar, and the H index of a scholar is got over Greatly, the paper impact factor of this scholar is bigger.

Define the H index of a paper using equation below in the present invention：

Wherein p_jRepresent paper, A (p_j) represent p_jH index, a_nIt is article p_jOn author, H (a_n) represent a_nAuthor H Index.

In order to distinguish significance level in collection of thesis for the paper, the present invention is using the paper H index tried to achieve and equation below Paper significance level in a network is defined：

Wherein P represents paper data set, Max ({ A (p_z)|p_z∈ P }) represent paper data set in paper H index maximum Value, the value of θ is the effect of 0.01, θ is to make δ (p_j) it is not 0.

Step 2)：It is obtained in that research contents and the direction of this article by the list of references of an article, for this Consider, the present invention to calculate the similarity of article by comparing " registration " of the list of references of two papers, and using as follows Formula calculates the list of references " registration " of two papers：

Wherein p_i, p_jRepresent any two papers in data set, OUT (p_i) and OUT (p_j) represent p_iAnd p_jReference literary composition Offer set,Represent p_i, p_jThe similarity of two articles, two papers are more similar,Value bigger.

Step 3)：It has been investigated that, the PageRank algorithm after either the PageRank algorithm of most original still deforms, It is all the algorithm of " no weight ".So-called " no weight " refers to that an article is " averagely main in article " ballot " quoted to it Justice ", it does not distinguish the particularity quoting article, but " making no exception ".The present invention is according to step 2) in draw " similar Degree ", modifies to PageRank algorithm as the eigenvalue quoting article, new algorithm is in original PageRank algorithm On the basis of add " weight " element.Meanwhile, by step 1) in the article significance level in a network that calculates also incorporate In PageRank algorithm, obtain AR-PageRank algorithm.

The expression formula of AR-PageRank algorithm is as follows：

Wherein d is " damped coefficient ", is typically set to 0.85, IN (p_i) represent all references p_iArticle set, PR (p_i) Represent p_iPageRank score,It is the fundamental formular of PageRank algorithm.'s Add the PageRank algorithm making this algorithm become " weighting ".Because paper p_jIn every list of references " ballot " to it This list of references can be carried out " analysis ", the different list of references of similarity can be treated with a certain discrimination.

Step 4)：According to step 3) the paper AR-PageRank value that obtains, these paper scores are entered in units of periodical Row is cumulative, and cumulative point is averaging, using this average mark as last periodical scoring.

The scoring formula of periodical is as follows：

Wherein Publish (j_i) represent be published in periodical j_iOn paper set, | Publish (j_i) | represent this collection of thesis The size closed, PR (p_k) it is paper p_kAR-PageRank score.

Beneficial effects of the present invention：The method calculating paper impact factor is a kind of algorithm based on PageRank：AR- PageRank, this algorithm considers the factors such as similarity between paper and the paper quoting it, author's H index, and the former makes This algorithm becomes the algorithm of " weighting ", and during algorithm iteration, every paper, can be because in paper " ballot " that it is quoted Similarity between for paper and give different contributions；The latter makes " importance " of every paper to greatly increase, by author H Index plus and be calculated the H index of every paper, this index as this paper a label so that the difference of every paper The opposite sex gradually displays such that it is able to relatively accurately distinguish the power of influence of every paper in calculating process, final acquisition The power of influence of each periodical.

Brief description

Fig. 1 applies obtained on DBLP data set respectively for PageRank, A-PageRank, AR-PageRank algorithm Ranking results.

Fig. 2 is flow chart of data processing Microsoft's MAG data set being carried out according to requirement of experiment.

Fig. 3, Fig. 4 and Fig. 5 apply on DBLP, DBLP (2011-2015), MAG data set respectively for four kinds of algorithms, and The correlation coefficient situation of each algorithm obtaining after being calculated using Spearman's correlation coefficient.

Fig. 6, Fig. 7, Fig. 8 and Fig. 9 be four kinds of algorithms on DBLP, DBLP (2011-2015) data set result right Than figure.

Specific embodiment

In order that the object, technical solutions and advantages of the present invention are clearer, below by the specific embodiment party to the present invention Formula is described in further detail.

Embodiments provide a kind of journal impact appraisal procedure based on academic big data, the method includes：

Step 1：Choose DBLP, MAG data set as the experimental data set of this method, DBLP and MAG data set is carried out Pretreatment.

The present invention with reference to the journal ranking that Chinese Computer Federation is recommended for 2015, and have chosen " artificial intelligence " field 42 periodicals as evaluation criterion (MAG data set have chosen 45 artificial intelligence field periodicals).According to requirement of experiment to two Individual data set carries out pretreatment, and Fig. 2 gives the flow chart that MAG data set is processed.

The information of latter two data set of pretreatment is as follows respectively：

Table 1MAG data set

Tab.1MAG dataset

Table 2DBLP data set

Tab.2DBLP dataset

In MAG, the quantity of paper will be significantly less than the quantity of DBLP paper as can be seen from the table, because extract obtaining MAG data set in Quantity of Papers very few, the present invention has crawled DBLP data set, and using this data set as " leading " data Collection, using MAG data set as " contrast " data set.

In order to calculate factors affecting periodicals (5-year IF), the present invention has individually extracted 2011- from DBLP data set 2015 data of totally 5 years, constitute the 3rd data set.

Table 3DBLP data set (2011-2015)

Tab.3DBLP dataset(2011-2015)

Step 2：The algorithm realization part of the present invention employs 3 data sets, is MAG, DBLP, DBLP (2011- respectively 2015).MAG, DBLP data set applies respectively Eigenfactor, PageRank, A-PageRank, R-PageRank, AR-PageRank algorithm.

AR-PageRank, i.e. Author Reference PageRank, Author mean that this algorithm considers H index This factor, Reference represents the relation considering paper and its list of references.Therefore, A-PageRank algorithm is exactly only Consider the factor of H index, P-PageRank algorithm is exactly the relation only considering paper and list of references.

The expression formula of AR-PageRank algorithm is：

The expression formula of A-PageRank algorithm is：

The expression formula of R-PageRank algorithm is：

The expression formula of PageRank algorithm is：

List in Fig. 1 is that PageRank, A-PageRank, AR-PageRank algorithm is applied respectively in DBLP data set Ranking results obtained by upper.

Step 3：By step 2) in the result that obtains calculated using Spearman's correlation coefficient, obtain each algorithm Correlation coefficient situation.

The present invention is calculated the similarity between list using mathematical method.Spearman's correlation coefficient is to weigh two Variable dependent nonparametric index, it evaluates the dependency of two statistical variables using dull equation.The following institute of computing formula Show：

Wherein, ρ represents Spearman's correlation coefficient, J_iRepresent one of journal list periodical, R₁、R₂It is journal list, R₁(J_i) represent periodical J_iIn list R₁In position.

Fig. 3, Fig. 4 and Fig. 5 give the correlation coefficient situation that four kinds of algorithms are concentrated in different pieces of information.It can be seen that in DBLP In data set, the correlation coefficient when periodical number is equal to 42 for the algorithm AR-PageRank is maximum, and this algorithm is in DBLP data set On behave oneself best；In DBLP data set (2011-2015), AR-PageRank algorithm is still Evaluated effect in each algorithm Preferably；In MAG data set, the correlation coefficient outline of A-PageRank algorithm is higher than AR-PageRank algorithm, R- The result of PageRank algorithm and PageRank algorithm is sufficiently close to.The reason above-mentioned two situations occur and MAG data The feature of collection is relevant, by observation it was found that the paper number of MAG data set, the paper number that is cited, author's number etc. are below it Its two datasets, " fine or not " of data set has important impact to experimental result.

Fig. 6, Fig. 7, Fig. 8 and Fig. 9 be four algorithms on DBLP, DBLP (2011-2015) data set result right Ratio figure, red broken line represents that algorithm applies the result on DBLP (2011-2015) data set, and the broken line of black represents Algorithm applies the result on DBLP data set.It is not difficult to find out process on DBLP data set for the same algorithm from figure Effect is better than DBLP (2011-2015) data set, and the latter is the former subset, and comparatively speaking, the former data scale is more Greatly, data class is more, and its experimental result is also more accurately and reliably.

It is the specific embodiment of the present invention and the know-why used described in above, if conception under this invention institute Make change, function produced by it still without departing from description and accompanying drawing covered spiritual when, must belong to the present invention's Protection domain.

Claims (1)

1. a kind of journal impact appraisal procedure based on academic big data is it is characterised in that step is as follows：

Step 1)：The assessment H index that the H index of Authors of Science Articles is acted on journal article power of influence as influence factor is author One index of aspect, it is a scholar or the quantity of publication of scientist and the tolerance of power of influence aspect, and H index compares Reflect the academic achievement of a scholar exactly, the H index of a scholar is bigger, and the paper impact factor of this scholar is bigger；

Define the H index of a paper with equation below：

$A\left(p_j\right)=\underset{a_n\∈p_j}{\mathrm{\Σ}}H\left(a_n\right)$

Wherein：p_jRepresent paper, A (p_j) represent p_jH index, a_nIt is paper p_jOn author, H (a_n) represent a_nAuthor H refer to Number；

Using the H index of the Authors of Science Articles obtaining and equation below, to paper, significance level in a network is defined：

$\mathrm{\δ}\left(p_j\right)=\frac{A\left(p_j\right)}{Max\left(\right\{A\left(p_z\right)|p_z\∈P\})}+\mathrm{\θ}$

Wherein：P represents paper data set, Max ({ A (p_z)|p_z∈ P }) represent paper data set in Authors of Science Articles H index Big be worth, the value of θ is the effect of 0.01, θ is to make δ (p_j) it is not 0；

Step 2)：To calculate the similarity of article by comparing " registration " of two references in papers

Obtain research contents and the direction of this article by the list of references of an article, by comparing two references in papers " registration " similarity to calculate article, and calculate " registration " of two references in papers using equation below：

Wherein：p_i, p_jRepresent any two papers in data set, OUT (p_i) and OUT (p_j) represent p_iAnd p_jList of references collection Close,Represent p_i, p_jThe similarity of two articles, two papers are more similar,Value bigger；

Step 3)：By analyze paper citation network feature, PageRank algorithm is improved, by step 1) in H refer to Number and step 2) in " registration " conduct " weight " element be attached in PageRank algorithm it follows that AR-PageRank Algorithm, calculates the AR-PageRank value of every paper；

The expression formula of AR-PageRank algorithm is as follows：

Wherein：D is " damped coefficient ", is set as 0.85, IN (p_i) represent all references p_iArticle set, PR (p_i) represent p_i's PageRank score,It is the fundamental formular of PageRank algorithm；Plus Enter so that this algorithm becomes the PageRank algorithm of " weighting ", as AR-PageRank algorithm；

Step 4)：The AR-PageRank value of papers all in periodical is summed up, obtains the assessed value of periodical

According to step 3) the paper AR-PageRank value that obtains, paper score is added up in units of periodical, cumulative point It is averaging, using this average mark as last periodical scoring；

The scoring formula of periodical is as follows：

$J\left(j_i\right)=\frac{\underset{p_k\∈Publish\left(j_i\right)}{\mathrm{\Σ}}PR\left(p_k\right)}{\left|Publish\left(j_i\right)\right|}$

Wherein：Publish(j_i) represent be published in periodical j_iOn paper set, | Publish (j_i) | represent this paper set Size, PR (p_k) it is paper p_kAR-PageRank score.