A method for automatic keyword extraction based on scientific literature graph network

Abstract

The present invention discloses a method for automatically extracting keywords based on a scientific and technological literature graph network, the method comprising: establishing a scientific and technological literature graph network for a set set of scientific and technological literature according to the literature citation relationship and co-author information; establishing a data organization model based on the scientific and technological literature graph network; extracting self-information from the scientific and technological literature to be tested, and combining it with the scientific and technological literature graph network information obtained based on the data organization model to complete keyword extraction; the self-information includes publication time, title, abstract, text, references and authors. The present invention constructs a scientific and technological literature graph network based on co-authors and citation relationships, and proposes an automatic extraction method based on the scientific and technological literature graph network, further improving the automatic extraction effect of scientific and technological literature keywords; it also proposes a scientific and technological literature graph network construction method and a scientific and technological literature graph network data organization model, which can make full use of scientific and technological literature graph network information and solve the problem of how to use graph network data.

Description

Technical Field

[0001] The present invention relates to the technical fields of computer application, natural language processing and automatic keyword extraction, and in particular to an automatic keyword extraction method based on a scientific literature graph network. Background Art

[0002] Traditional scientific literature keyword extraction methods use the information of the literature itself to extract keywords, ignoring the network relationship between scientific literature and failing to apply the semantic information of the scientific literature graph network to the field of key extraction. Usually, a scientific literature has multiple authors, and multiple authors publish other scientific literature. A scientific literature can be associated with the cited scientific literature through the citation relationship, and a scientific literature can be cited by other scientific literature. Scientific literature forms a complex graph network through common authors and citation relationships. Summary of the invention

[0003] The purpose of the present invention is to overcome the defects of the prior art and propose a method for automatic keyword extraction based on a scientific literature graph network.

[0004] In order to achieve the above object, the present invention proposes a method for automatically extracting keywords based on a scientific literature graph network, the method comprising:

[0005] Step 1) Establish a scientific literature graph network for a set of scientific literature collection based on literature citation relationships and co-author information;

[0006] Step 2) Establish a data organization model based on the scientific literature graph network;

[0007] Step 3) Extract the self-information of the scientific and technological literature to be tested, and combine it with the scientific and technological literature graph network information obtained based on the data organization model to complete keyword extraction; the self-information includes publication time, title, abstract, text, references and authors.

[0008] As an improvement of the above method, the step 1) specifically includes:

[0009] Step 1-1) Set each document in the set scientific and technological document collection as a node;

[0010] Step 1-2) traverse each node, repeat steps 1-3) and 1-4), when each node is traversed, go to step 1-5);

[0011] Step 1-3) According to the document citation relationship, an edge is established from the node corresponding to the citing document to the node corresponding to the cited document; the category of the node corresponding to the citing document is set as the citing node, and the category of the node corresponding to the cited document is set as the cited node;

[0012] Step 1-4) Based on the co-author information, establish an edge between the corresponding nodes of the papers with co-authors, and set the node category to co-author node;

[0013] Steps 1-5) obtain the scientific literature graph network.

[0014] As an improvement of the above method, the step 2) specifically includes:

[0015] Step 2-1) Set the key node information according to the category of each node;

[0016] Step 2-2) Calculate the weight of the node key information according to the node category and the node key information.

[0017] As an improvement of the above method, the step 2-1) specifically includes:

[0018] For a node type of a reference node, the key node information is set to include: title, abstract, time, keywords, reference level, and reference fragment; wherein the time indicates the publication time of the document, and the reference level indicates the distance between citations of scientific and technological documents;

[0019] For a node type of cited node, the key node information is set to include: title, abstract, time, keywords, cited level and cited fragment; the cited level indicates the distance of citation of scientific and technological literature;

[0020] For a node whose type is a co-author node, the key node information is set to include: title, abstract, time, keywords and co-author level, where the co-author level represents the correlation distance between co-authors of the document.

[0021] As an improvement of the above method, the step 2-2) specifically includes:

[0022] For a node type of reference node, the information weight of the title is 1.5 Q1, the information weight of the abstract is Q1, the information weight of the keyword is 2 Q1, and the information weight of the reference fragment is 1.2 Q1; among them, Q1 is the benchmark weight of citing scientific literature, which satisfies the following formula:

[0023] Q1=A×(1-first time difference / 10) ×(1-reference level / 5)

[0024] Among them, A represents the weight base of citing scientific literature, and the first time difference represents the difference between the publication time of the citing scientific literature and the cited scientific literature;

[0025] For nodes of the cited node category, the information weight of the title is 1.5 Q2, the information weight of the abstract is Q2, the information weight of the keyword is 2 Q2, and the information weight of the cited fragment is 1.2 Q2; among them, Q2 is the benchmark weight of the cited scientific literature, which satisfies the following formula:

[0026] Q2=B×(1-first time difference / 10) ×(1-referenced level / 5)

[0027] Among them, B represents the weight base of the cited scientific literature, and the first time difference represents the difference between the publication time of the cited scientific literature and the cited scientific literature;

[0028] For nodes with the co-author category, the information weight of the title is 1.5 Q3, the information weight of the abstract is Q3, and the information weight of the keyword is 2 Q3, where Q3 is the benchmark weight Q3 of the co-author scientific literature, satisfying the following formula:

[0029] Q3=C×(1-second time difference / 10) ×(1-co-author level / 5)

[0030] Among them, C represents the weight base of the co-author scientific documents, and the second time difference represents the difference in the publication time of the co-author scientific documents.

[0031] As an improvement of the above method, the step 3) specifically includes:

[0032] Extract the information of the scientific and technological literature to be tested, and obtain the time, title, abstract, text, references and author;

[0033] Construct a scientific literature graph network based on co-authorship and citation relationships;

[0034] Organize the scientific literature graph network data based on the scientific literature citation network data organization model;

[0035] Integrate the information of the scientific literature to be tested with the network information of the scientific literature graph to extract keywords

[0036] Compared with the prior art, the advantages of the present invention are:

[0037] 1. The present invention constructs a complex scientific literature graph network based on co-authors and citation relationships, and proposes an automatic extraction method based on the scientific literature graph network to further improve the automatic extraction effect of scientific literature keywords;

[0038] 2. The present invention proposes a scientific and technological literature graph network construction method and a scientific and technological literature graph network data organization model, which can make full use of scientific and technological literature graph network information and solve the problem of how to use graph network data. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a flow chart of the automatic keyword extraction method based on the scientific literature graph network of the present invention;

[0040] Figure 2 It is the scientific and technological document node category and related node key information of the present invention;

[0041] Figure 3 It is the scientific and technological document node category and weight setting of the present invention;

[0042] Figure 4 This is an example of keyword extraction using the method of the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0044] Example 1

[0045] like Figure 1 As shown, Embodiment 1 of the present invention proposes a method for automatically extracting keywords based on a scientific literature graph network, which specifically includes the following steps:

[0046] (1) Method for constructing scientific literature graph network

[0047] For a collection of papers, we treat each paper as a node. The connection between different nodes is constructed through citation relationships and co-author information. For example, if document A cites document B, then node A will generate an edge pointing to node B. With the author information of scientific document A, we can link it to scientific document C published by the co-author, and then we can generate an edge from node C to node A. Through document citation relationships and co-author information, we can easily build a scientific document graph network.

[0048] (2) Scientific and technological literature graph network data organization model

[0049] For the scientific literature graph network data, we built a data organization model. We divided the scientific literature graph network nodes into three categories: cited scientific literature nodes, cited scientific literature nodes, and scientific literature nodes published by co-authors. The specific categories and key information of related nodes are as follows: Figure 2 shown.

[0050] (3) Keyword extraction method of scientific literature graph network

[0051] For a scientific document, with the help of the scientific document graph network data organization model, we can obtain relevant information about citing scientific documents, cited scientific documents, and co-authored scientific documents. When extracting keywords, these relevant information can be used as supplementary information for scientific documents. We assign different weights to supplementary information based on the time difference, hierarchical relationship, and key information type of the scientific documents. The smaller the time difference between scientific documents, the higher the correlation and the higher the weight. The more levels there are between scientific documents, the worse the correlation and the lower the weight. Title information is more concise than abstract information, and will be assigned a higher weight than the abstract. The weight setting is as follows: Figure 3 Specifically:

[0052] For a node type of reference node, the information weight of the title is 1.5 Q1, the information weight of the abstract is Q1, the information weight of the keyword is 2 Q1, and the information weight of the reference fragment is 1.2 Q1; among them, Q1 is the benchmark weight of citing scientific literature, which satisfies the following formula:

[0053] Q1=A×(1-first time difference / 10) ×(1-reference level / 5)

[0054] Among them, A represents the weight base of citing scientific literature, and the first time difference represents the difference between the publication time of the citing scientific literature and the cited scientific literature;

[0055] For nodes of the cited node category, the information weight of the title is 1.5 Q2, the information weight of the abstract is Q2, the information weight of the keyword is 2 Q2, and the information weight of the cited fragment is 1.2 Q2; among them, Q2 is the benchmark weight of the cited scientific literature, which satisfies the following formula:

[0056] Q2=B×(1-first time difference / 10) ×(1-referenced level / 5)

[0057] Among them, B represents the weight base of the cited scientific literature, and the first time difference represents the difference between the publication time of the cited scientific literature and the cited scientific literature;

[0058] For nodes with the co-author category, the information weight of the title is 1.5 Q3, the information weight of the abstract is Q3, and the information weight of the keyword is 2 Q3, where Q3 is the benchmark weight Q3 of the co-author scientific literature, satisfying the following formula:

[0059] Q3=C×(1-second time difference / 10) ×(1-co-author level / 5)

[0060] Among them, C represents the weight base of the co-author scientific documents, and the second time difference represents the difference in the publication time of the co-author scientific documents.

[0061] Finally, we combine the information of the scientific literature itself with the scientific literature graph network information obtained with the help of the scientific literature graph network data organization model, and then perform keyword extraction.

[0062] Simulation example:

[0063] This paper proposes an automatic keyword extraction method based on the scientific literature graph network. First, we need to build a scientific literature graph network based on citation relationships and co-author information. CiteSeer is an academic paper digital library built by NEC Research Institute based on the automatic citation indexing mechanism. By searching CiteSeer with the title of scientific literature, you can get the abstract, title, and citation context. The CiteSeer search results are shown in the figure. We can easily obtain author information, citation context, publication time, abstract information, title information, cited paper information, etc. Figure 4 shown.

[0064] Then we organize the graph network data based on the scientific literature citation network data organization model. Finally, we extract keywords based on the scientific literature graph network keyword extraction framework.

[0065] To verify the effect of keyword extraction, we selected 500 scientific papers from CNKI as experimental data. First, we used the existing keyword extractor to extract keywords from scientific papers, and calculated the extraction accuracy (P), recall rate (R), and F1 value; then we used the method proposed in this paper to extract keywords, and calculated the extraction accuracy (P), recall rate (R), and F1 value; finally, we compared the test results. The calculation method of P, R, and F1 indicators is as follows:

[0066] P = the number of accurate results automatically extracted / the total number of results automatically extracted

[0067] R = number of accurate results automatically extracted / keywords of the document itself

[0068] F1=2PR / (P+R).

[0069] Experimental results show that the method proposed in this paper can improve the keyword extraction effect by 5% to 15% on the original basis.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention and should be included in the scope of the claims of the present invention.

Claims

1. A method for automatically extracting keywords based on a scientific literature graph network, the method comprising: Step 1) Establish a scientific literature graph network for a set of scientific literature collection based on literature citation relationships and co-author information; Step 2) Establish a data organization model based on the scientific literature graph network; Step 3) extracting self-information from the scientific and technological literature to be tested, and combining it with the scientific and technological literature graph network information obtained based on the data organization model to complete keyword extraction; the self-information includes publication time, title, abstract, text, references and author; The step 1) specifically includes: Step 1-1) Set each document in the set scientific and technological document collection as a node; Step 1-2) traverse each node, repeat steps 1-3) and 1-4), when each node is traversed, go to step 1-5); Step 1-3) According to the document citation relationship, an edge is established from the node corresponding to the citing document to the node corresponding to the cited document; the category of the node corresponding to the citing document is set as the citing node, and the category of the node corresponding to the cited document is set as the cited node; Step 1-4) Based on the co-author information, establish an edge between the corresponding nodes of the papers with co-authors, and set the node category to co-author node; Steps 1-5) obtain the scientific literature graph network; The step 2) specifically includes: Step 2-1) Set the key node information according to the category of each node; Step 2-2) Calculate the weight of the node key information according to the node category and node key information; The step 2-1) specifically includes: For a node type of a reference node, the key node information is set to include: title, abstract, time, keywords, reference level, and reference fragment; wherein the time indicates the publication time of the document, and the reference level indicates the distance between citations of scientific and technological documents; For a node type of cited node, the key node information is set to include: title, abstract, time, keywords, cited level and cited fragment; the cited level indicates the distance of citation of scientific and technological literature; For a node type of a co-author node, the key node information is set to include: title, abstract, time, keywords and co-author level, where the co-author level indicates the correlation distance between co-authors of the document; The step 2-2) specifically includes: For a node type of reference node, the information weight of the title is 1.5 Q1, the information weight of the abstract is Q1, the information weight of the keyword is 2 Q1, and the information weight of the reference fragment is 1.2 Q1; among them, Q1 is the benchmark weight of citing scientific literature, which satisfies the following formula: Q1=A×(1-first time difference / 10) ×(1-reference level / 5) Among them, A represents the weight base of citing scientific literature, and the first time difference represents the difference between the publication time of the citing scientific literature and the cited scientific literature; For nodes of the cited node category, the information weight of the title is 1.5 Q2, the information weight of the abstract is Q2, the information weight of the keyword is 2 Q2, and the information weight of the cited fragment is 1.2 Q2; among them, Q2 is the benchmark weight of the cited scientific literature, which satisfies the following formula: Q2=B×(1-first time difference / 10) ×(1-referenced level / 5) Among them, B represents the weight base of the cited scientific literature, and the first time difference represents the difference between the publication time of the cited scientific literature and the cited scientific literature; For nodes with the co-author category, the information weight of the title is 1.5 Q3, the information weight of the abstract is Q3, and the information weight of the keyword is 2 Q3, where Q3 is the benchmark weight Q3 of the co-author scientific literature, satisfying the following formula: Q3=C×(1-second time difference / 10) ×(1-co-author level / 5) Among them, C represents the weight base of the co-author scientific documents, and the second time difference represents the difference in the publication time of the co-author scientific documents.

2. The method for automatic keyword extraction based on scientific literature graph network according to claim 1 is characterized in that: The step 3) specifically includes: Extract the information of the scientific and technological literature to be tested, and obtain the time, title, abstract, text, references and author; Construct a scientific literature graph network based on co-authorship and citation relationships; Organize the scientific literature graph network data based on the scientific literature citation network data organization model; The information of the scientific literature to be tested is integrated with the network information of the scientific literature graph to extract keywords.

Images