Web of Science科研社区挖掘算法研究

Research on Web of Science Academic Community Mining Algorithm

本文以1900-2019年"Web of Science(WOS)"核心合集中的中国科学院(中科院)部分论文数据为面板数据对热点学科、科研社区及相关权威专家进行了分析研究.首先对艺术与人文、生命科学与生物医学、自然科学、社会科学、应用科学五大学科数据进行分析,发现应用科学(Technology)发表论文年增速最快,且研究热点为计算机科学(Computer Science);其次针对研究热点应用Neo4j图数据库构建论文语义网络图,对实体关系进行优化,提升了社区内部关联度;并基于Louvain社区发现算法进行了相关优化和数据挖掘,分析了其背后的优秀科研团队;最后针对挖掘出的社区,利用PageRank算法筛选出高产出的权威科研人员,为科研合作和人才发现甚至国家学科布局提供参考.实验表明,通过Neo4j图数据库中实体数据索引设计,查询性能提升高达16倍;通过对Louvain算法关系属性weight添加机构影响维度,社区模块度提升了84%.

Based on the’Web of Science(WOS)’core collection papers of the Chinese Academy of Sciences(CAS)from 1900 to2019 being regarded as the panel data,this study analyzes the issues concerning about the popular subjects,scientific research communities,and related authoritative experts.Firstly,after the data researching in the five subjects of Art and Humanities,Life Sciences and Biomedicine,Natural Sciences,Social Sciences,and Applied Sciences,it is found that the fastest annual grow th of papers published are articles related to Technology Applied Sciences and the popular research topic is the Computer Science.Secondly,for the research of the popular application,network diagram about papers semantics is constructed based on the Neo4 j graph database,and entity relationship is optimized w hich improves the internal relevance of the community.Then,according to the Louvain community discovery algorithm,we perform the related optimization and data mining and analyze the outstanding scientific research team making efforts for the achievements.Finally,targeting the mined communities,we screen out the high-output authoritative scientific researchers for providing the reference for scientific research cooperation,talent discovery,and even the national discipline layout.According to the experiments,the results show that search performance is improved by up to 16 times due to the design of the entity data index based on the Neo4 j graph database.Besides,since adding the dimension of institutional influence into the weight of Louvain algorithm relationship attributes,the community modularity is increased by 84%.