语句层共被引关系内容抽取与分类及其应用研究——以Athar引用语料库为例

Research on Relation Content Extraction and Classification in Sentence-level Co-citation and Its Application:An Example of Athar’s Citation Sentiment Corpus

[目的/意义]语句层共被引的关系内容抽取与分类有助于揭示共被引论文间的主题关联。[方法/过程]文章从共被引主题的相似性和相关性出发,利用引用标注位置、作者、语义角色分析、句法分析等信息,将语句层共被引关系划分为同系列、同主题、发展关联、运用关联、并列关联,然后抽取相应的引用主题,构建<被引论文及主题,关系类型,共被引论文及主题>双层三元组,实现共被引关系内容结构化表达,并在Neo4j图数据库中呈现。[结果/结论]实验采用Athar引用语料库;结果表明,本文研究方法可提高语句层共被引网络中关系的可读性和共被引论文的语义搜索、问答与推荐的效率。[局限]实验方法针对英文文献而设计,未来将在更多领域的英文语料上进行验证,并从名词性关系识别、术语选择等方面完善关系内容抽取与分类。

[Purpose/significance]The relation content extraction and classification of sentence-level co-citation networks is helpful to explain the topic relations between co-cited papers.[Method/process]From the perspective of similarity and relevance of co-cited topics,this paper uses the information of citation positions,authors,semantic role analysis,and syntactic analysis to classify the sentence-level co-citation relations into same series,same topic,development association,application association and juxtaposition association;and then extracts the corresponding cited topics to construct the double-layer triplets.The structured expressions of co-citation relation content are presented in the Neo4j graph database.[Result/conclusion]This experiment used Athar’s citation sentiment corpus.The result shows that this approach improves the readability of relations in sentence-level co-citation networks and the efficiency of semantic search,question answering and recommendation about co-cited papers.[Limitations]The experimental method in this paper is designed for English documents.In the future,it will be verified on English corpus from more fields,and improve relation content extraction and classification in nominal relation recognition,term selection and so on.