CN103020302A - Academic core author excavation and related information extraction method and system based on complex network - Google Patents

Abstract

The invention belongs to the field of data mining. Aiming at the problem of mining core authors in a certain academic field and intelligently extracting their relevant information, the invention proposes an improved academic core author mining and information mining algorithm based on the core node discovery algorithm in social network analysis technology. Extraction method and system. This method integrates vertical search technology, social network analysis technology and text analysis technology, and can find core authors or groups in a certain academic field in massive information, and then obtain their relevant personal information. The present invention adopts vertical search technology to collect open source document data. Analyze the importance of various social entities present in the data using bibliometric techniques and complex network analysis techniques. And use the community discovery algorithm to cluster entities based on the closeness of the relationship between entities to discover academic communities. Users sort according to the importance of entities, find the core authors or institutions, and find the leadership team according to the distribution of the number of publications of the cooperative groups.

Description

Method and system for academic core author mining and related information extraction based on complex network

technical field

The invention relates to the field of data mining, in particular to a complex network-based academic core author mining and related information extraction method and system.

Background technique

Many practical networks have a common property, that is, they are all networks connected by various communities through public nodes. The connection between the nodes within the community is relatively tight, and the connection between the communities is relatively sparse. For example, the World Wide Web can be regarded as composed of a large number of website communities, and many sites within the same community often discuss some topics of common interest. Similarly, in the author cooperation network or circuit network, each node can also be divided into different communities according to their different properties. Therefore, the number of communities in the network and the belonging community and number of each node are of great significance to the study of complex networks.

For the definition of community structure in the network, there is no generally accepted standard. Therefore, there are many forms of community structure definitions in the network, but they are generally divided into two categories:

1. Use the relative density of edges between node pairs to measure the community structure. Under the definition of this method, the connection between the node pairs within each community is relatively tight, but the connection between each community is relatively sparse.

2. Use precise quantitative indicators in graph theory to define the community structure. These community structures are all derived from the definition of clique in graph theory. In the definition of this type of structure, it is generally required that every point in the community is adjacent, or at most it can be non-adjacent to the number of points, or how many jumps are the farthest between any two points, and so on.

At present, expert identification and recommendation in the field usually adopts the method of constructing fuzzy text classifiers, fuzzy text classification of documents uploaded by experts to the knowledge base, and establishing expert knowledge models in combination with factors such as quantity and time. This method has incomplete text databases. The coverage is low, and it is difficult to comprehensively analyze the specific contributions and related personal information of experts in the field in multiple fields, which has great limitations. Based on this, the present invention uses the complex network construction, parameter analysis and community discovery algorithm in the complex network analysis technology, which can be effectively used for the discovery of core figures or core groups in the subject field and the acquisition of related information.

Contents of the invention

Aiming at the problem of digging core figures in a certain academic field and intelligently extracting their relevant information, the present invention proposes an improved academic core author mining and information extraction algorithm and system based on the core node discovery method in social network analysis technology. The method and system aim at literature data in a specific field, and use complex network construction, parameter analysis, and community discovery algorithms in complex network analysis techniques to efficiently find core groups or key figures in the field.

A kind of academic core author digging and related information extraction method based on complex network proposed by the present invention, it comprises:

Step 1. Use vertical search technology to collect literature data in the designated field, and organize and analyze the literature data to obtain author-related information;

Step 2. Extract the author cooperation network according to the obtained author-related information, and count the author-related parameters, and obtain different author ranking information according to the different related parameters that are counted;

Step 3. Divide the extracted cooperative network into communities, and the divided communities are regarded as a scientific research group;

Step 4: Display the different author ranking information and scientific research groups to the user, and recommend core authors and leadership teams to the user according to the author ranking information and scientific research groups selected by the user.

The present invention also proposes a complex network-based academic core author mining and related information extraction system, which includes:

Data collection and collation device: used to collect literature data in a designated field using vertical search technology, and organize and analyze the literature data to obtain author-related information;

Parameter analysis and statistics device: extract the author cooperation network according to the obtained author-related information, and count the author-related parameters, and obtain different author ranking information according to the different related parameters that are counted;

Community division device: divide the extracted cooperative network into communities, and the divided communities are regarded as a scientific research group;

Result display device: display the different author ranking information and scientific research groups to the user, and recommend core authors and leadership teams to the user according to the author ranking information and scientific research groups selected by the user.

Description of drawings

Fig. 1 is a schematic diagram of the application system of the present invention;

Fig. 2 is a simple flow chart of the application system of the present invention;

Fig. 3 is the flow chart of the academic core author digging and related information extraction method based on complex network in the present invention;

Fig. 4 is a sub-flow chart of data collection in the present invention;

Fig. 5 is a sub-flow chart of data acquisition configuration in the present invention;

Fig. 6 is a sub-flow chart of data analysis and arrangement in the present invention;

Fig. 7 is a screenshot of the application system realized by the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be described in further detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

The complex network-based academic core author mining and related information extraction method and system proposed by the present invention are invented for information retrieval of core expert groups in the field. The principle of the application system is shown in Figure 1.

The technology used in the present invention is introduced below:

1. Collection technology

1.1 Vertical search

This method uses vertical search technology to obtain relevant metadata such as authors, institutions, conferences, etc. from commonly used literature search engines such as CNKI and SpringerLink according to the fields and conferences that users are concerned about, and automatically downloads and parses the full text of the literature to obtain the literature. Correspondence details of the author or institution.

Vertical search is a professional search engine for a certain field. It is a subdivision and extension of search engines. It is an integration of a certain type of specialized information in the webpage library, and the required data is extracted by directional sub-fields for processing and then processed by a certain field. form returned to the user. Compared with the general search engine, which has a large amount of information, inaccurate queries, and insufficient depth, etc., a new search engine service model is proposed, through providing valuable information and Related Services. It is characterized by "specialization, precision, and depth" and has an industry color. Compared with the disordered massive information of general search engines, vertical search engines are more focused, specific, and in-depth.

The most important technology of vertical search is search engine crawler. Search engine crawler technology is a technology that automatically crawls network information according to certain rules. The design of relevant search engine crawlers in this system is based on common crawlers, and its functions are effectively expanded, mainly including field-related initial URL seed sets, page crawling modules, topic correlation analysis modules, URL duplicate checking and page downloading, etc. module. This design can ensure good topic relevance of the system, so as to improve the hit rate of crawled topic-related pages and meet the needs of users.

1.2 Web page collection

Web page collection in this technology is mainly divided into deep web collection and dynamic web collection. The deep web is characterized by the concealment of its pages, and generally requires users to submit data request forms to obtain returned results. The main feature of the pages of the dynamic network is "dynamic existence", that is, the pages that are dynamically generated temporarily by the program when the user calls the page. According to the distribution of information items, dynamic web pages are mainly divided into two types: one is multi-record item dynamic web pages; the other is single-record item dynamic web pages. An exact representation of the request.

2. Analysis technology

2.1 Complex network technology

2.1.1 Basic concepts

A specific network can be abstracted as a graph G composed of a point set V and an edge set E, the number of nodes is denoted as N=|V(G)|, and the number of edges is denoted as M=|E(G)|. Each edge in E has a corresponding node pair in V. If any pair of vertices corresponds to the same edge, the network is undirected, otherwise it is directed. If the network contains only one type of nodes and edges, the network is said to be homogeneous, otherwise the network is a heterogeneous network.

2.1.2 Betweenness centrality

Betweenness centrality is defined based on the ability of nodes to control network communication. It believes that if a node exists on the only way for communication between other node pairs in the network, it must have an important position in the network.

2.1.3 Aggregation coefficient

Clustering coefficient (clusteringcoefficient) is often used to describe the transitivity of the network. For example, in a social network, your friend's friend is likely to be your friend; two of your friends are likely to be friends with each other. The aggregation coefficient is used to measure this property of the network.

2.2 Other statistical indicators

2.2.1 H-index

An important measure to evaluate the influence of scientists is the H-index measure. The value of H-index is based on the number of scientists' articles and the number of citations. For example, if a certain scholar has at least h articles cited h times, the H-index value of this scholar is h. From the above description, the greater the H-index value of a scholar, the greater his influence in his research field. The H-index metric comprehensively takes into consideration the quantity and quality of scientific research achievements published by scholars.

2.2.2 APS value (average output score)

The APS value is defined as: For a paper with n authors, APS gives each author a score of 1/n. An author's APS is the sum of the scores of all his papers. It describes the author's contribution to the articles he publishes.

The present invention proposes a complex network-based academic core author mining and information extraction method. The simple use process of the application system is shown in Figure 2, and the process of the academic core author mining and related information extraction method is shown in Figure 3. Specific steps are as follows:

Step 1: Data collection and collation. This method adopts the vertical search technology to collect the paper literature data of the specified conference. The collection process is shown in Figure 4. This step consists of three stages:

Phase 1: Basic data acquisition, specifically includes: Step a) Determine the acquisition conditions, see Figure 5 for the determination of the acquisition conditions. First, you need to determine the type of retrieval, including three types of retrieval: periodicals, conferences, and keywords. Then, according to different types, determine search terms, time and other search conditions, such as meeting configuration conditions (meeting-related search terms, etc.), literature search source and search year and other configuration conditions. Then select data sources, including different databases at home and abroad. Thus, a collection of retrieval conditions is formed. Among them, the conference configuration conditions need to be input by the user, and the other configuration conditions are adjusted by the system itself; step b), dynamically configure the collection information according to the collection conditions, and configure the collection information for each determined data source site, such as CNKI, SpringerLink, etc., such as retrieval If the type is periodical, the configured collection information is periodical, etc.; step c) basic document data collection. Here, using vertical search technology, according to the fields and conferences that users are concerned about, through the initial URL seed set, page crawling module, topic correlation analysis module, URL duplicate checking and page downloading modules, from commonly used documents such as CNKI, SpringerLink, etc. Obtain relevant metadata such as authors, institutions, conferences, etc. from the search engine, and automatically download and analyze the full text of the literature.

Stage 2: Data collation, specifically includes: Step d) Data cleaning, mainly to standardize the author’s name, remove redundant characters, such as spaces, etc., and merge institutions, such as the second-level institution unit is replaced by its first-level unit name, etc. ; Step e) obtains specified information, the main research object in the present invention is the author, so in this step, simple author information can be obtained, namely the author's name and the unique identification ID assigned by the system.

Stage 3: information storage, specifically includes: step f) display the results to the user, and let the user judge whether they are satisfied with the results, if satisfied, proceed to step g), otherwise return to step a) to reconfigure; step g) transfer basic document information and The author's information is stored in the designated database; step h) the system judges whether to collect data cyclically, if so, wait for a period of time and then collect again, otherwise end the collection step.

Step 2: Statistical analysis of parameters. See Figure 6 for the sub-process of data analysis and arrangement. The research object of this method is the core authors and groups related to the specified field. Therefore, it is necessary to analyze the statistical parameters of the authors' literature, and to identify the core authors in this field by comprehensively ranking the values of each parameter. Statistical parameters include the distribution of the author's published papers and the distribution of the author's APS (average output score), and use the collaborator relationship to extract the author's cooperation network, and analyze the author's node intermediary centrality, degree distribution, network aggregation coefficient and H in the cooperation network. -index measure, where node betweenness centrality is used to measure the degree to which an author can control the communication between others, if a node is on the shortest path of many other point pairs, it has a high betweenness centrality. It can be considered that the author is in an important position, and the degree distribution indicates how many people an author has had a cooperative relationship with. The network aggregation coefficient refers to the proportion of adjacent points of nodes in the network that are also adjacent points of each other, that is, the degree of perfection of the small cluster structure. A parameter to measure the author's clustering situation in the network nodes; H-index measurement indicates that h articles of an author have been cited h times, and the value of the scholar's H-index is h, which is used to measure its influence in the research field force. Save the author ranking information obtained by different parameters, that is, according to the distribution of the author's publication volume, the author's APS (average output score) distribution, the author's node intermediary centrality in the cooperation network, degree distribution, network clustering coefficient and H-index Parameters such as metrics get different author ranking information.

Step 3: Carry out group analysis according to the community division algorithm. This method divides the community of the author's cooperation network, and each community after division is equivalent to a scientific research group. The distribution of published papers is counted for all scientific research groups.

Step 4: Display of author ranking information and research group information. Display the different author ranking information saved in step 2 and the scientific research groups found in step 3 to the user, and recommend important authors as scientific research leaders and important groups as the core team according to the author ranking information and scientific research group rankings selected by the user.

Step 5: Extract and display core author information. Users select scholars in major fields as core authors according to their needs, and the system automatically extracts their personal profile information through document information and presents them to users for related business or research use.

Among them, in stage 1 of step 1, the literature collection method is a combination of deep web collection and dynamic web collection.

The working process of deep web collection can be divided into 3 steps: 1) analyze the page and find the form; 2) learn to fill in the form; 3) identify and retrieve the result page. Among them, the deep web crawler crawls form pages from the homepage of the site in the first step. This process uses a set of heuristic rules to remove non-research forms; the second step extracts labels from the form, and cooperates with the domain rule knowledge base and website feature identification ( Username, password or captcha), the crawler tries its best to learn how to fill out the form correctly; the last step is to submit the form, and then get back the result page identification record. In addition, in the process of deep web collection, web crawlers need to intelligently identify specific application domain knowledge based on the domain knowledge base to ensure the relevance and accuracy of the collected information.

In the process of dynamic web collection, when extracting the information of multi-record dynamic web pages, it is necessary to use the tree edit distance model and tree merge model algorithm to locate and extract web page information. Use the tree edit distance to accurately locate the extraction structure of the webpage, convert the dynamic webpage into a tag tree and locate the data items in the separated webpage, and generate a separate data item tree for a single data item; apply the tree merge model to the pattern extraction of multiple data items above, control repeated data items and optional data items, and generate a wrapper tree for extraction, which is the final extractor. When extracting the information of a dynamic web page with a single record item, the user needs to customize the extracted data item through an optional module, and the system will generate an extraction template according to the data item selected by the user. In the extraction process, the webpage is first converted into a tag tree, and the webpage information is extracted and saved through user-defined extraction template matching.

In step c of stage 1, the sources of the document engine mainly include CNKI and SpringerLink, and the collected content includes the title of the document, the original text of the document, the author of the document, the keywords of the document, the author’s institution, the publication of the document, and the publication time of the document.

In step 2, the expression of betweenness centrality is defined as:

${\mathrm{<span\; class="notranslate">\; BC</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; <</span>\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\frac{{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; jk</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; jk</span>}}}<span\; class="notranslate">\; ,</span>$

where g _jk (i) represents the number of shortest paths between nodes j and k passing through node i, and g _jk represents the number of shortest paths between node j and node k. For a directed network, the directionality of the path needs to be considered. The concept of betweenness centrality is very important in social network analysis. In addition, the concept of betweenness centrality can not only define the betweenness of nodes, but also define the betweenness of edges to measure the importance of edges in the network.

The network aggregation coefficient indicates whether the adjacent points of the nodes are connected, and is a measure of the network transitivity. In layman's terms, the neighbors of the node's network neighbors may also be the neighbors of the node, defined as:

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; 3</span>}\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; \&Delta;</span>}}}{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; 3</span>}}}<span\; class="notranslate">\; .</span>$

where N△ refers to the number of triangles in the cooperative network, and N3 refers to the number of connected triplets in the cooperative network. A connected triplet refers to a triplet consisting of three nodes including a given node, and at least two edges from the given node to two other nodes.

In step three, the community partition algorithm uses a fast community partition algorithm for directed networks. The fast community partition algorithm is an improved algorithm based on the modularity concept proposed in the GN algorithm.

First introduce the GN algorithm:

A simple way to divide communities is to remove the edges connecting different communities, which is the central idea of the split method. The community discovery algorithm proposed by Grivan and Newman-GN algorithm is the most famous splitting algorithm for community discovery. The algorithm uses the betweenness centrality of the edges introduced above, and then gradually deletes the edges that do not belong to the community according to the extent to which the edges do not belong to the community, until all the edges are deleted. According to the definition of community, the edge between communities has a larger edge betweenness than the edge within the community. By gradually removing edges with high betweenness, the network can be divided into communities.

However, in the worst case, the GN algorithm needs to recalculate the betweenness of all edges every time an edge is removed, which is only suitable for medium-scale social networks. Aiming at this defect, many studies have improved it from different angles. In addition, scholars have proposed that the GN algorithm does not have a quantitative definition for the community structure of the network. Therefore, it cannot be judged directly from the topology of the network whether the obtained community structure has practical significance. In addition, the GN algorithm does not know this decomposition without knowing the number of communities.

$\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; ij</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; \&sigma;</span>}}_{{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}$

To which step to terminate. To solve this problem, Newman et al. introduced a standard to measure the quality of network community division - modularity (Q), which is defined as:

Among them, A is the adjacency matrix of the graph, A _ij represents the edge weight, ki and k _j are the degrees of nodes i and _j respectively, the degree refers to the number of edges associated with the node (vertex), and m is the graph The total number of edges, C _i and C _j represent the community numbers of nodes i and j respectively. If the base node i and node j belong to the same community, the delta function takes 1, otherwise it takes 0.

The fast community partition algorithm is introduced as follows:

The GN algorithm is a very important milestone in the field of community discovery algorithms, but due to its relatively large algorithm complexity, it is only limited to the study of medium-scale social networks. For this reason, a fast community partition algorithm is proposed based on the GN algorithm. This fast algorithm is actually an agglomerative algorithm based on the idea of greedy algorithm. The algorithm steps are as follows: 1. Initialize the network as n communities, that is, each node is an independent community; 2. Merge the pairs of communities connected by edges in turn, and calculate the modularity (Q) increment after the merger; 3. Repeat execution Step 2, constantly merging communities until the entire network is merged into one community. Each operation of merging communities in the algorithm corresponds to a modularity value, and when it corresponds to the local maximum modularity value, it is the best community division.

However, due to the high complexity of the algorithm, at present, most community discovery algorithms are still focused on the research of undirected networks, but in fact, most of the networks we are interested in are directed networks, such as the World Wide Web, telecom communication networks, etc. , Email communication network, biological network and so on. Neglecting the direction of network connection for community discovery means discarding important information in the network structure, which makes the results of community discovery biased.

For directed networks, a modified modularity formula (modularity function) is used

$\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; m</span>}}\underset{\mathrm{<span\; class="notranslate">\; ij</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; A</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; in</span>}}{\mathrm{<span\; class="notranslate">\; k</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; out</span>}}}{\mathrm{<span\; class="notranslate">\; m</span>}}<span\; class="notranslate">\; ]</span>{\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; ,</span>$

${\mathrm{<span\; class="notranslate">\; \&delta;</span>}}_{{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}\\ \mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>& {\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; \&NotEqual;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}\end{array}\right.$

为节点i的入度，

为节点j的出度；节点的入度是指进入该节点的边的条数；节点的出度是指从该节点出发的边的条数。m为网络的总边数。Among them, A is the adjacency matrix of the graph, A _ij represents the edge weight, and δ is the Kroenekef delta symbol. If the base node i and node j belong to the same community, the δ function takes 1, otherwise it takes 0.

is the in-degree of node i,

is the out-degree of node j; the in-degree of a node refers to the number of edges entering the node; the out-degree of a node refers to the number of edges starting from the node. m is the total number of edges in the network.

In Step 5, the contact information of the contact may refer to the Email of the contact. Email information is obtained from the original text of the documents extracted in step 1. The system automatically parses the original document, uses regular expressions to match the text information in the Email format, and extracts all Email information contained in the original document. The system uses three regular expressions to match the target Email:

Regex1=

″\w+([-+.]\w+)*\w+([-.]\w+)*\.\w+([-.]\w+)*″

Regex2＝

″\{?(\w*([-+.]\w+)*, (\s)*)*\w*([-+.]\w+)*\}?

\w+([-.]\w+)*\.\w*([-.]\w*)*″

Regex3＝

″(\s)*e-mail: (\s)*\w+([-+.]\w+)*\w+([-.]\w+)*\.\w+([-.]\w+) *(\s)*″

The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present invention in detail. The screenshot of the application system realized by the present invention is shown in Figure 7. It should be understood that the above descriptions are only specific embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention are Should be included within the protection scope of the present invention.

Claims (10)

1. the academic Core Authors based on complex network excavates and the relevant information abstracting method, and it comprises:

Step 1, employing vertical search technology gather the data in literature of designated field, and described data in literature is carried out finishing analysis, to obtain author's relevant information;

Step 2, extract author's cooperative network according to author's relevant information of obtaining, and the statistics author parameter of being correlated with, different author's ranking informations obtained according to the different correlation parameters of adding up;

Step 3, the cooperative network that extracts is carried out corporations divide, the corporations after the division are as a scientific research colony;

Step 4, show described different author's ranking information and scientific research colony to the user, and recommend Core Authors and leader team according to user-selected author's ranking information and scientific research colony for the user.

2. the method for claim 1 is characterized in that, the data in literature that gathers designated field in the step 1 specifically comprises:

Step 11, determine acquisition condition, comprise the deterministic retrieval type, according to difference retrieval type deterministic retrieval condition;

Step 12, according to the Information Monitoring of acquisition condition dynamic-configuration;

Step 13, obtain data in literature according to acquisition condition and Information Monitoring.

3. the method for claim 1 is characterized in that, in the step 1 data is carried out finishing analysis and specifically comprises to obtain author's relevant information:

Step 14, carry out data cleansing;

Step 15, obtain author's relevant information of appointment.

4. the method for claim 1, it is characterized in that step 1 comprises that also the author's relative information displaying that will obtain to the user, need to determine whether the Resurvey data by the user, if need to would reconfigure acquisition condition, and carry out image data according to the acquisition condition that reconfigures.

5. the method for claim 1, it is characterized in that correlation parameter described in the step 2 comprises author's the distribution of dispatch amount, author's average output score, node Betweenness Centrality, degree distribution, network convergence factor and the H-index tolerance of author in cooperative network.

6. method as claimed in claim 5 is characterized in that, described node Betweenness Centrality calculates according to following formula and obtains:

${\mathrm{BC}}_i=\underset{j<k}{\mathrm{\&Sigma;}}\frac{g_{\mathrm{jk}}\left(i\right)}{g_{\mathrm{jk}}},$

Wherein, g _Jk(i) number of the shortest path by node i between expression node j and the k, g _JkThe number of shortest path between expression node j and the node k;

Described network convergence factor obtains according to following formula:

$C=\frac{{3N}_{\mathrm{\&Delta;}}}{N_3}.$

Wherein, N _△The number that refers to cooperative network intermediate cam shape, N ₃Refer to be communicated with in the cooperative network quantity of tlv triple.

7. the method for claim 1 is characterized in that, corporations described in the step 3 divide the quick group dividing method that adopts for directed networks, specifically comprise:

Step 31, the described cooperative network of initialization are n corporations, and namely each node is independent corporations;

Step 32, be associated with the corporations that the limit links to each other successively, and calculate the modularity value after merging;

Step 33, repeated execution of steps 32, until whole cooperative network all is merged into corporations, wherein, when the modularity value was maximum, corresponding corporations were the corporations after final the division after merging.

8. method as claimed in claim 7, wherein said modularity value is calculated according to following formula:

$Q=\frac{1}{m}\underset{\mathrm{ij}}{\mathrm{\&Sigma;}[}{A}_{\mathrm{ij}}-\frac{k_i^{\mathrm{in}}{k}_j^{\mathrm{out}}}{m}]{\mathrm{\&delta;}}_{c_i{c}_j},$

${\mathrm{\&delta;}}_{C_i{C}_j}=\left\{\begin{array}{cc}1,& C_i=C_j\\ 0,& C_i\&NotEqual;C_j\end{array}\right.$

Wherein, Q is the modularity value, and A is adjacency matrix, A _IjExpression limit power,

Be the in-degree of node i,

Out-degree for node j; M is total limit number of cooperative network.

9. the method for claim 1 is characterized in that, the method also comprises:

Step 5, analysis data in literature, the personal information that extracts Core Authors also offers the user.

10. the academic Core Authors based on complex network excavates and the relevant information extraction system, and it comprises:

Data acquisition and collating unit: be used for adopting the vertical search technology to gather the data in literature of designated field, and described data in literature is carried out finishing analysis, to obtain author's relevant information;

Parameter analytic statistics device: extract author's cooperative network according to author's relevant information of obtaining, and add up the parameter that the author is correlated with, obtain different author's ranking informations according to the different correlation parameters of adding up;

Corporations divide device: the cooperative network that extracts is carried out corporations divide, the corporations after the division are as a scientific research colony;

Exhibiting device as a result: show described different author's ranking information and scientific research colony to the user, and recommend Core Authors and leader team according to user-selected author's ranking information and scientific research colony for the user.

Images