CN106227835A - Team's research direction method for digging based on two subnetwork figure hierarchical clusterings - Google Patents

Abstract

The invention discloses a kind of team's research direction method for digging based on two subnetwork figure hierarchical clusterings, comprise the following steps: step 1: set up author investigation interest representation mode based on author's key word two subnetwork；Step 2: author investigation interest representation mode is carried out figure cluster: author little for the degree of concern difference to each key word is attributed to same author group；Obtain author's cluster set；Step 3: general levels clusters, obtain the research interest of each author group: author's cluster set will only comprise the group of an author, it is merged in other author group that research interest is similar, make the author's number comprised in each author group more than 2, calculate and update the research interest of each author group, i.e. team's research direction.The present invention can excavate the academic research direction of team effectively, provides advantage for analyzing and evaluate the development of team's research direction.

Description

Team's research direction method for digging based on two subnetwork figure hierarchical clusterings

Technical field

The present invention relates to a kind of team's research direction method for digging based on two subnetwork figure hierarchical clusterings.

Background technology

In the trend of globalization day by day significantly today, team unity is a phenomenon the most universal.Along with multi-disciplinary Mixing together, the interpenetrating of science technology and society, have higher requirement to research and development management and scientific research organization.Science and technology wound New team is helped each other by resource-sharing, cooperative cooperating and the division of labor so that S&T innovation efficiency is greatly improved, as science and technology research and development One effectively organizes form, and scientific and technical innovation team progressively becomes the important model of scientific research and innovation activity^[1].For ensureing Persistently the carrying out and turn out individual and the team with scientific and technical innovation spirit, National Natural Science base of science frontier research activities Gold committee started from 2000 tentative to set up " innovation colony science fund ", in order to help domestic outstanding innovative research team The basic scientific research that a certain Important Academic research direction is carried out and applied basic research^[2]。

The research direction of team is to evaluate the important symbol of whole team development." National Natural Science Foundation of China creates Recent studies on colony science fund trial method "^[3]Explicitly pointing out, creation & research team must be the entirety that Long-term Collaboration is formed, and has The research direction of Relatively centralized, and can persistently be active in the forward position of its research field.Scientific paper is led generally around science A certain problem in territory study after science record or Scientific summarization, if the opinion that relevant academic problem is studied Literary composition, their similarity is the highest, but the paper that different problems are answered, their difference is the biggest.Root in the past According to this characteristic between paper, it is possible to analyze the dependency of the research direction of an Academic Teams.And this mode is not There are the relation between member, academic relationship network in consideration team^[4]There is complex structure.Academic direction to team For, on the one hand need to consider the research interest place of member in team；On the other hand the respectively side of research is remained a need for considering in team Relation between to.

Therefore, it is necessary to the team's research direction designing a kind of combination Team Member research interest and relational network feature is dug Pick method.

List of references

[1] Wang Xinxin. the structure of scientific and technical innovation team and Study on Development Tactics [J]. science and technology and economy, 2014,27 (3): 66-69.

[2] Feng Changgen. National Natural Science Foundation of China's creation & research team [J]. science and technology Leader, 2010,28 (7): 125.

[3] " National Natural Science Foundation of China's creation & research team science fund trial method ", 2001,2.

[4]Fang Huang,Jing Liu,Xinmin Liu,et al.Academic Relation Classification Rules Extraction with Correlation Feature Weight Selection[C] .the 3rd Global Congress on Intelligent Systems(GCIS2012),Nov.6-8,2012:160- 165.

[5]Tian Y,Hankins R A,Patel J M.Efficient aggregation for graph summarization[C].AcmSigmod International Conference on Management of Data.2008:567-580.

[6] Chen Kehan, Han Panpan, Wu Jian. isomery social networks proposed algorithm [J] based on user clustering. computer Report, 2013,38 (2): 349-359.

Summary of the invention

Solved by the invention technical problem is that, for the deficiencies in the prior art, it is provided that a kind of based on two subnetwork figure layers Team's research direction method for digging of secondary cluster, research based on figure clustering method, the research direction of team is excavated, for The development analyzing and evaluating team's research direction provides advantage.

Technical scheme provided by the present invention is:

A kind of team's research direction method for digging based on two subnetwork figure hierarchical clusterings, comprises the following steps:

Step 1: set up author investigation interest representation mode based on author's key word two subnetwork；

Step 2: author investigation interest representation mode is carried out figure cluster:

Author little for degree of concern difference to each key word is attributed to same author group；Obtain author group collection Close；

Step 3: general levels clusters, and obtains the research interest of each author group:

Author's cluster set will only comprise the group of an author, be merged into other author group that research interest is similar In so that the author's number comprised in each author group is more than 2, calculates and update the research interest of each author group, i.e. Team's research direction.

Described step 1 particularly as follows:

The scientific paper collection of author from team, extraction author information and key word information, obtain preprocessed data, Wherein, author's collection is designated as V_A={ A₁,A₂,…,A_N, keyword set is designated as V_K=K={k₁,k₂,…,k_M, by author A_iScience In collection of thesis, lists of keywords and keyword set K compare, therefore, for each author A_n, this author A of obtaining_nResearch Interest representation mode is A_n={ (k₁,w_n1),(k₂,w_n2),…,(k_M,w_nM)}；

Research interest representation mode based on author, constructs the author interests matrix m of N × M, wherein, collects for author In each author, define this author A_nResearch interest vector be v_n=(w_n1,w_n2,…,w_nM)；

Author investigation interest representation mode is expressed as G=G (V, E)；

The set that wherein V is formed by author node and key word node, i.e. V={V_AUV_K, wherein V_AGather for author V_A={ A₁,A₂,…,A_n,…,A_N, V_KFor keyword set V_K=K={k₁,k₂,…,k_j,…,k_M, N and M is respectively in team Author sum and team in all authors scientific paper concentrate key word sum；E is author node and key word node Between the set that constituted of company limit, i.e. E={e (A_n,k_j)|A_n∈V_A,k_j∈K,w_nj>0}；If author is A_nScientific paper in Lists of keywords comprises certain key word k in keyword set_j, then weight w_nj> 0, at author A_nWith key word k_jBetween exist Even limit e (A_n,k_j), otherwise w_nj=0, at author A_nWith key word k_jBetween there is not even limit.

Described step 2, carries out figure cluster to author investigation interest representation mode G=G (V, E)；Comprise the following steps:

2.1) author's cluster set Groups={G is initialized₀, G₀It is one and comprises the contributors group of all authors in team Group；

2.2) forDefinition author group G_iTo key word k_jConcern collection be:

Wherein, A is author group G_iIn author.

2.3) author group G is calculated by formula (2)_iTo each key word k_j(k_j∈ K) concern situation focus_ij:

${\mathrm{focus}}_{ij}=\frac{\left|{\mathrm{Focus}}_{k_j}\left(G_i\right)\right|}{\left|G_i\right|}---\left(2\right)$

Wherein,Represent author group G_iMiddle concern key word k_jAuthor's quantity, | G_i| represent contributors group Group G_iIn comprise author sum；If attention rate focus_ij>=α, Ze Cheng author group G_i" pay close attention to by force " in key word k_j, otherwise Claim author group G_i" weak concern " is in key word k_j；Wherein α > 0, for paying close attention to intensity threshold；

In an author group, the concern situation of key word is more concentrated by authors, and the degree of polymerization of this group is more High.Ambiguity in definition degree describes the difference degree paid close attention between each author of author's group internal for key word.

2.4) each author group G is calculated by formula (3)_iAt each key word k_j(k_j∈ K) on fuzziness fuzzy_ij:

${\mathrm{fuzzy}}_{ij}={\mathrm{\&delta;}}_{k_j}\left(G_i\right)=\left\{\begin{array}{ccc}\left|{\mathrm{Focus}}_{k_j}\left(G_i\right)\right|& if& {\mathrm{focus}}_{ij}<\mathrm{\&alpha;}\\ |\left|G_i\right|-\left|{\mathrm{Focus}}_{k_j}\left(G_i\right)\right||& if& {\mathrm{focus}}_{ij}\&GreaterEqual;\mathrm{\&alpha;}\end{array}\right.---\left(3\right)$

In formula (3), as author group G_i" pay close attention to by force " in key word k_jTime, fuzzy_ijEqual to author group G_iIn do not have Pay close attention to key word k_jAuthor's number；Author group G_i" weak concern " is in key word k_jTime, fuzzy_ijEqual to author group G_iMiddle concern Key word k_jAuthor's number.

2.5) according to fuzzy_ijCalculate each author group G_iFuzziness fuzzy to keyword set K_i:

${\mathrm{fuzzy}}_i=\underset{j=1}{\overset{\left|K\right|}{\&Sigma;}}{\mathrm{fuzzy}}_{ij}---\left(4\right)$

Wherein | K | is the key word sum in keyword set K, i.e. M；

2.6) overall fuzziness Fuzzy of this Groups is calculated:

$Fuzzy=\underset{G_i\&Element;Groups}{\&Sigma;}{\mathrm{fuzzy}}_i---\left(5\right)$

2.7) fuzzy is found_ijMaximum, by the key word k of its correspondence_jAs locking word k_j′；

Find fuzzy_iMaximum, by the author group G of its correspondence_iAs group G to be divided_i′；

To treat that splitting group group is split into two new author group G_i1And G_i2, update author's cluster set Groups；

G_i1={ A_n∈G_i′||w_nj′>0}

G_i2=G_i-G_i1；

2.8) repeated execution of steps 2.2)～2.7), until the author group number in author's cluster set Groups is k；

This cluster result is designated as Groups={G₁,G₂,…,G_k, k is the classification number in cluster result, meets:

(1)And

(2)G_j∈ Groups, and i ≠ j,

2.9) relatively each stage etch 2.6) in overall fuzziness Fuzzy of cluster result Groups that obtains, will Groups corresponding to Fuzzy minima, as final cluster result, is designated as summaryGroups；

It is as follows that the algorithm of described step 2 correspondence performs process:

This algorithm originates in a group comprising all authors, then during each iteration, to original group Group divides, until having obtained k group.This algorithm is not to be randomly chosen a group to divide, but based on right The concern situation of key word selects group to be divided.The definition of contact fuzziness, closes certain for an author group " weak concern " relation of keyword, it is intended that isolate the concern collection to this key word from which；And relative to " paying close attention to by force " Situation, we then wish to isolate those non-interesting collection, and both operations Dou Huishi group has higher pass to this key word Note degree.Therefore, we select group to be divided, and it is carried out splitting operation.Each iterative computation when, to often The cluster result in individual stage preserves.Finally, an optimum cluster result is selected.

The pseudo-code of algorithm is described as follows:

The most openness, through clustering based on author's key word two subnetwork figure due to author's key word two subnetwork data After result, have indivedual author to be individually classified as a class, these discrete author node may just this Academic Teams expand new Research direction.It is therefore desirable to discrete author node is processed, the convenient handle to whole team's academic research direction Hold.

Described step 3 specifically includes following steps:

3.1) the contributors group component in the cluster result summaryGroups that will obtain in step 2 is discrete contributors group Group and discrete author group；Discrete author group refers to only comprise the author group of an author；Discrete author group is made For initial cluster；

3.2) each discrete author group G is calculated_iClass research interest vector GMI in keyword set K_i (Group Major Interests) is as the center of each initial cluster；

GMI_i=(GW_i1,GW_i2,…,GW_ij,…,GW_iM) (6)

Wherein, GW_ij(j=1,2 ..., M) represent G_iTo key word k_jConcern situation, quantitative description is:

${\mathrm{GW}}_{ij}=\frac{\underset{A_n\&Element;G_i}{\&Sigma;}{w}_{nj}}{\left|G_i\right|}---\left(7\right)$

3.3) each author A in discrete author group is traveled through_n, calculate its center with each initial cluster European away from From；Computational methods are:

If author is A_nResearch interest vector be v_n=(w_n1,w_n2,…,w_nj,…,w_nM)

$d_{ni}=\sqrt{\underset{k=1}{\overset{M}{\&Sigma;}}{({\mathrm{GW}}_{ij}-w_{nj})}^2};$

3.4) A is compared_nWith the Euclidean distance at the center of each initial cluster, select the discrete that Euclidean distance minima is corresponding Author group, by A_nDistribute to this discrete author group, will only comprise author A_nDiscrete author group and this discrete make Person group merges, and forms a new author group；

3.5) iteration carries out above-mentioned steps 3.1)～3.4), until the author group produced no longer changes；

3.6) calculate and update the class research interest vector of each author group.

The pseudo-code of the algorithm of described step 3 correspondence is described as follows:

Beneficial effect:

The present invention is according to Team Member and the scientific paper information thereof excavating certain team: document name, participate in list of authors, Lists of keywords etc., has carried out pretreatment, and has utilized Authors of Science Articles and paper in Academic Teams crucial the data set obtained Word information characterizes and quantifies the research interest of author；Author and key in the topological structure unique for two subnetworks and team Between word two are intrinsic, construct author investigation interest representation mode based on author's key word two subnetwork.Then making Carry out figure cluster on person's key word two subnetwork, excavated the body feature of team, this network is had individual Preliminary study.Finally In the body feature of author's key word two subnetwork, carry out the overall situation of the cluster result Chu Liao Team Member of network general levels Academic research direction, lays a good foundation for carrying out the analysis in team's science direction from now on.

The present invention combines figure digest algorithm, hierarchical clustering algorithm and k-means algorithm, it is proposed that based on author's key word two Subnetwork figure clustering algorithm and network general levels clustering algorithm, poor to the concern of different key words according to each author in team DRS degree, clusters the author in team as k author group；In an author group, author's concern feelings to key word Condition Relatively centralized.The present invention can excavate the academic research direction of team effectively, for analyzing and evaluating team's research direction Development provides advantage.

Accompanying drawing explanation

Fig. 1 is team's research direction method for digging flow process；

Fig. 2 is author's key word two subnetwork；

The result of author's key word two subnetwork figure cluster when Fig. 3 is k=5；

Author's key word two subnetwork figure cluster result when Fig. 4 is k=7；

Fig. 5 is 4 the author groups produced after network general levels clusters；

Fig. 6 is 5 the author groups produced after network general levels clusters.

Detailed description of the invention

Below in conjunction with the drawings and specific embodiments, the present invention is described in more detail.

The invention provides a kind of team's research direction method for digging based on two subnetwork figure hierarchical clusterings, including following Step:

Step 1: set up author investigation interest representation mode G=G (V, E) based on author's key word two subnetwork；

The set that wherein V is formed by author node and key word node, i.e. V={V_AUV_K, wherein V_AGather for author V_A={ A₁,A₂,…,A_n,…,A_N, V_KFor keyword set V_K=K={k₁,k₂,…,k_j,…,k_M, N and M is respectively in team Author sum and team in all authors scientific paper concentrate key word sum；E is author node and key word node Between the set that constituted of company limit, i.e. E={e (A_n,k_j)|A_n∈V_A,k_j∈K,w_nj>0}；If author is A_nScientific paper in Lists of keywords comprises certain key word k in keyword set_j, then weight w_nj> 0, at author A_nWith key word k_jBetween exist Even limit e (A_n,k_j), otherwise w_nj=0, at author A_nWith key word k_jBetween there is not even limit.

Step 2: author investigation interest representation mode is carried out figure cluster: by little to the degree of concern difference of each key word Author be attributed to same author group；Obtain author's cluster set；

2.1) author's cluster set Groups={G is initialized₀, G₀It is one and comprises the contributors group of all authors in team Group；

2.2) forDefinition author group G_iTo key word k_jConcern collection be:

Wherein, A is author group G_iIn author.

2.3) author group G is calculated by formula (2)_iTo each key word k_j(k_j∈ K) concern situation focus_ij:

${\mathrm{focus}}_{ij}=\frac{\left|{\mathrm{Focus}}_{k_j}\left(G_i\right)\right|}{\left|G_i\right|}---\left(2\right)$

Wherein,Represent author group G_iMiddle concern key word k_jAuthor's quantity, | G_i| represent contributors group Group G_iIn comprise author sum；If attention rate focus_ij>=α, Ze Cheng author group G_i" pay close attention to by force " in key word k_j, otherwise Claim author group G_i" weak concern " is in key word k_j；Wherein α > 0, for paying close attention to intensity threshold；

2.4) each author group G is calculated by formula (3)_iAt each key word k_j(k_j∈ K) on fuzziness fuzzy_ij:

${\mathrm{fuzzy}}_{ij}={\mathrm{\&delta;}}_{k_j}\left(G_i\right)=\left\{\begin{array}{ccc}\left|{\mathrm{Focus}}_{k_j}\left(G_i\right)\right|& if& {\mathrm{focus}}_{ij}<\mathrm{\&alpha;}\\ |\left|G_i\right|-\left|{\mathrm{Focus}}_{k_j}\left(G_i\right)\right||& if& {\mathrm{focus}}_{ij}\&GreaterEqual;\mathrm{\&alpha;}\end{array}\right.---\left(3\right)$

2.5) according to fuzzy_ijCalculate each author group G_iFuzziness fuzzy to keyword set K_i:

${\mathrm{fuzzy}}_i=\underset{j=1}{\overset{\left|K\right|}{\&Sigma;}}{\mathrm{fuzzy}}_{ij}---\left(4\right)$

Wherein | K | is the key word sum in keyword set K, i.e. M；

2.6) overall fuzziness Fuzzy of this Groups is calculated:

$Fuzzy=\underset{G_i\&Element;Groups}{\&Sigma;}{\mathrm{fuzzy}}_i---\left(5\right)$

2.7) fuzzy is found_ijMaximum, by the key word k of its correspondence_jAs locking word k_j′；

Find fuzzy_iMaximum, by the author group G of its correspondence_iAs group G to be divided_i′；

To treat that splitting group group is split into two new author group G_i1And G_i2, update author's cluster set Groups；

G_i1={ A_n∈G_i′|w_nj′>0}

G_i2=G_i-G_i1；

2.8) repeated execution of steps 2.2)～2.7), until the author group number in author's cluster set Groups is k；

2.9) relatively each stage etch 2.6) in overall fuzziness Fuzzy of cluster result Groups that obtains, will Groups corresponding to Fuzzy minima, as final cluster result, is designated as summaryGroups.

Step 3: general levels clusters, and obtains the research interest of each author group:

Author's cluster set will only comprise the group of an author, be merged into other author group that research interest is similar In so that the author's number comprised in each author group is more than 2, calculates and update the research interest of each author group, i.e. Team's research direction；

3.1) the contributors group component in the cluster result summaryGroups that will obtain in step 2 is discrete contributors group Group and discrete author group；Discrete author group refers to only comprise the author group of an author；Discrete author group is made For initial cluster；

3.2) each discrete author group G is calculated_iClass research interest vector GMI in keyword set K_iAs The center of each initial cluster；

GMI_i=(GW_i1,GW_i2,…,GW_ij,…,GW_iM) (6)

Wherein, GW_ij(j=1,2 ..., M) represent G_iTo key word k_jConcern situation, quantitative description is:

${\mathrm{GW}}_{ij}=\frac{\underset{A_n\&Element;G_i}{\&Sigma;}{w}_{nj}}{\left|G_i\right|}---\left(7\right)$

3.3) each author A in discrete author group is traveled through_n, calculate its center with each initial cluster European away from From；Computational methods are:

If author is A_nResearch interest vector be v_n=(w_n1,w_n2,…,w_nj,…,w_nM)

$d_{ni}=\sqrt{\underset{k=1}{\overset{M}{\&Sigma;}}{({\mathrm{GW}}_{ij}-w_{nj})}^2};$

3.4) A is compared_nWith the Euclidean distance at the center of each initial cluster, select the discrete that Euclidean distance minima is corresponding Author group, by A_nDistribute to this discrete author group, will only comprise author A_nDiscrete author group and this discrete make Person group merges, and forms a new author group；

3.5) iteration carries out above-mentioned steps 3.1)～3.4), until the author group produced no longer changes；

3.6) calculate and update the class research interest vector of each author group, i.e. team's research direction vector.

Main flow of the present invention is as shown in Figure 1:

Fig. 1 is team's research direction method for digging flow process, from team's scientific paper data, builds based on author crucial The author investigation interest representation mode of word two subnetwork；Then author investigation interest representation mode is carried out figure cluster, rolled into a ball The body feature of team's research interest；Finally the body feature excavated is analyzed, carries out the cluster of network general levels, Research direction to this team's overall situation.

Experimental analysis

The source of 1 data

These part data come from certain computer science and technology research team as object of study, by the paper of this team Data set has carried out experimental verification analysis, and the most in visual form and the form of form is opened up by experimental result Show.The author investigation interest representation mode of this part, author's cohort studies interest representation mode all by means of Gephi software to be carried out Signal is shown.

2 author's key word two subnetworks

Form and scientific paper data set by analyzing the team of this team, obtain in the research interest model of this team Comprising 23 Team Members, 547 paper key words, initial author's key word two subnetwork of foundation is as shown in Figure 2.At Fig. 2 Illustrate only the name of author in this team, the research interest worlds of its correspondence are dispersed in around corresponding author, due to joint Point is numerous, does not the most demonstrate keyword attribute.

3 team's research interest body feature

Based on author's key word two subnetwork figure cluster result during 1.k=5

When Fig. 3 describes k=5, the result of author's key word two subnetwork figure cluster: (made containing after cluster by 552 nodes Person group and key word), 714 limits are constituted.It can be seen that each group nodes is not of uniform size, wherein in group 1, number is relatively Many, group 4 contains only a Team Member.To above-mentioned information, it is possible to use form is shown.Such as table 1, list 5 author groups concern situation to Partial key word set.

Table 1 team research interest body feature partial information example shows (during k=5)

Based on author's key word two subnetwork figure cluster result during 2.k=7

When that Fig. 4 shows is k=7, author's key word two subnetwork figure cluster result: 554 nodes, 721 limits.And figure 3 compare, and owing to the number of cluster increases, occur in that more groupuscule, such as group 4, group 5 and group 7 in Fig. 4.Similarly, Table 2 lists the part research field that 7 author groups pay close attention to:

Table 2 team research interest body feature partial information example shows (during k=7)

Original team based on two subnetworks author investigation interest representation mode, by using based on author's key word two After subnetwork figure clustering algorithm excavates, the information of archetype can be simplified, focal point is placed on the main body of this model In structure, contribute to the assurance to team's main direction of studying.

The interpretation of result of 4 network general levels clusters

1. process team's research interest body feature during k=5

First, the method using network general levels cluster, process team's research interest body feature during k=5, Arriving: 551 nodes, 693 limits, as it is shown in figure 5, now, cluster number is 4 to obtained team's overall situation research interest.Can To find out, comparing with Fig. 3, the research interest of Team Member becomes apparent from, more concentrates.

Table 3 lists now 4 author groups concern situation to Partial key word set.Compare with table 1, group 3 and group Group 4 there occurs change to the concern situation of Partial key word (such as: lens distortions, Hanzi component).

Table 3 team overall situation research direction certain embodiments

2. process team's research interest body feature during k=7

Again, interest body feature is studied by team during k=7 and use the cluster of network general levels, due in Fig. 4 In team's body feature information, discrete author node has two, and therefore after global clustering, author's group number is 5.Global clustering Result as shown in Figure 6, wherein comprises 552 nodes, 679 limits.

Table 4 lists now 5 author groups and the part research field of concern thereof, compares with table 2, due to discrete work The classification of person's node, the research interest worlds causing network general levels to cluster the concern of Hou Ge group have occurred that change.

Table 4 team overall situation research direction certain embodiments

The present invention has initially set up author investigation interest representation mode based on two subnetworks.Emerging then in conjunction with author investigation Interest represents the feature of model, introduces figure clustering algorithm based on author's key word two subnetwork, excavates the body feature of network, And then obtain the main direction of studying of team.Finally by the basic thought of k-means algorithm, on the basis of network principal feature On, carry out the excavation of network general levels, obtain the overall situation research interest worlds of team.The present invention can excavate team effectively Academic research direction, provide advantage for analyzing and evaluate the development of team research direction.

Claims (4)

1. team's research direction method for digging based on two subnetwork figure hierarchical clusterings, it is characterised in that include following step Rapid:

Step 1: set up author investigation interest representation mode based on author's key word two subnetwork；

Step 2: author investigation interest representation mode is carried out figure cluster:

Author little for degree of concern difference to each key word is attributed to same author group；Obtain author's cluster set；

Step 3: general levels clusters, and obtains the research interest of each author group:

Author's cluster set will only comprise the group of an author, be merged in other author group that research interest is similar, Make the author's number comprised in each author group more than 2, calculate and update the research interest of each author group, i.e. team Research direction.

Team's research direction method for digging based on two subnetwork figure hierarchical clusterings the most according to claim 1, its feature Being, in described step 1, author investigation interest representation mode is expressed as G=G (V, E)；

The set that wherein V is formed by author node and key word node, i.e. V={V_AUV_K, wherein V_AV is gathered for author_A= {A₁,A₂,…,A_n,…,A_N, V_KFor keyword set V_K=K={k₁,k₂,…,k_j,…,k_M, N and M is respectively in team The key word sum that in author's sum and team, the scientific paper of all authors is concentrated；E be author node and key word node it Between the set that constituted of company limit, i.e. E={e (A_n,k_j)|A_n∈V_A,k_j∈K,w_nj>0}；If author is A_nScientific paper in close Keyword list comprises certain key word k in keyword set_j, then weight w_nj> 0, at author A_nWith key word k_jBetween exist even Limit e (A_n,k_j), otherwise w_nj=0, at author A_nWith key word k_jBetween there is not even limit.

Team's research direction method for digging based on two subnetwork figure hierarchical clusterings the most according to claim 1, its feature Being, described step 2 comprises the following steps:

2.1) author's cluster set Groups={G is initialized₀, G₀It it is an author group comprising all authors in team；

2.2) forDefinition author group G_iTo key word k_jConcern collection be:

Wherein, A is author group G_iIn author；

2.3) author group G is calculated by formula (2)_iTo each key word k_j(k_j∈ K) concern situation focus_ij:

${\mathrm{focus}}_{ij}=\frac{\left|{\mathrm{Focus}}_{k_j}\left(G_i\right)\right|}{\left|G_i\right|}---\left(2\right)$

Wherein,Represent author group G_iMiddle concern key word k_jAuthor's quantity, | G_i| represent author group G_iIn The author's sum comprised；If attention rate focus_ij>=α, Ze Cheng author group G_i" pay close attention to by force " in key word k_j, otherwise it is referred to as Person group G_i" weak concern " is in key word k_j；Wherein α > 0, for paying close attention to intensity threshold；

2.4) each author group G is calculated by formula (3)_iAt each key word k_j(k_j∈ K) on fuzziness fuzzy_ij:

${\mathrm{fuzzy}}_{ij}={\mathrm{\&delta;}}_{k_j}\left(G_i\right)=\left\{\begin{array}{ccc}\left|{\mathrm{Focus}}_{k_j}\left(G_i\right)\right|& if& {\mathrm{focus}}_{ij}<\mathrm{\&alpha;}\\ \left|\right|G_i|-|{\mathrm{Focus}}_{k_j}\left(G_i\right)\left|\right|& if& {\mathrm{focus}}_{ij}\&GreaterEqual;\mathrm{\&alpha;}\end{array}\right.---\left(3\right)$

2.5) according to fuzzy_ijCalculate each author group G_iFuzziness fuzzy to keyword set K_i:

${\mathrm{fuzzy}}_i=\underset{j=1}{\overset{\left|K\right|}{\&Sigma;}}{\mathrm{fuzzy}}_{ij}---\left(4\right)$

Wherein | K | is the key word sum in keyword set K, i.e. M；

2.6) overall fuzziness Fuzzy of this Groups is calculated:

$Fuzzy=\underset{G_i\&Element;Groups}{\&Sigma;}{\mathrm{fuzzy}}_i---\left(5\right)$

2.7) fuzzyi is found_jMaximum, by the key word k of its correspondence_jAs locking word k_j′；

Find fuzzy_iMaximum, by the author group G of its correspondence_iAs group G to be divided_i′；

To treat that splitting group group is split into two new author group G_i1And G_i2, update author's cluster set Groups；

G_i1={ A_n∈G_i′|w_nj′>0}

G_i2=G_i-G_i1；

2.8) repeated execution of steps 2.2)～2.7), until the author group number in author's cluster set Groups is k；

2.9) relatively each stage etch 2.6) in overall fuzziness Fuzzy of cluster result Groups that obtains, by Fuzzy The Groups of little value correspondence, as final cluster result, is designated as summaryGroups.

Team's research direction method for digging based on two subnetwork figure hierarchical clusterings the most according to claim 3, its feature Being, described step 3 comprises the following steps:

3.1) the contributors group component in the cluster result summaryGroups that will obtain in step 2 be discrete author group and Discrete author group；Discrete author group refers to only comprise the author group of an author；Using discrete author group as just Beginning bunch；

3.2) each discrete author group G is calculated_iClass research interest vector GMI in keyword set K_iAs each The center of initial cluster；

GMI_i=(GW_i1,GW_i2,…,GW_ij,…,GW_iM) (6)

Wherein, GW_ij(j=1,2 ..., M) represent G_iTo key word k_jConcern situation, quantitative description is:

${\mathrm{GW}}_{ij}=\frac{\underset{A_n\&Element;G_i}{\&Sigma;}{w}_{nj}}{\left|G_i\right|}---\left(7\right)$

3.3) each author A in discrete author group is traveled through_n, calculate the Euclidean distance of itself and the center of each initial cluster；Calculate Method is:

If author is A_nResearch interest vector be v_n=(w_n1,w_n2,…,w_nj,…,w_nM)

$d_{ni}=\sqrt{\underset{k=1}{\overset{M}{\&Sigma;}}{({\mathrm{GW}}_{ij}-w_{nj})}^2};$

3.4) A is compared_nWith the Euclidean distance at the center of each initial cluster, select the discrete author that Euclidean distance minima is corresponding Group, by A_nDistribute to this discrete author group, will only comprise author A_nDiscrete author group and this discrete contributors group Combination also, forms a new author group；

3.5) iteration carries out above-mentioned steps 3.1)～3.4), until the author group produced no longer changes；

3.6) calculate and update the class research interest vector of each author group.