Examining knowledge entities and its relationships based on citation sentences using a multi-anchor bipartite network

This paper proposes a novel entitymetrics approach by exclusively focusing on citation sentences. Since citation sentences offer authors’ research interest, knowledge entities that appear in such sentences can be considered as key entities. To characterize such key entities, we focus on citation sentences that were extracted from full-text research articles collected from PubMed Central. We used “opioid” as our search query since it is an actively studied domain, which indicates that rigorous amounts of knowledge entities and entity pairs are available for examination. After which we construct two novel citation sentence-based networks, namely the Direct Citation Sentence (DCS) network and the Indirect Citation Sentence (ICS) network. The DCS network is built upon direct entity pairs that are captured within citation sentences. The ICS network, on the other hand, utilized indirect entity cooccurrences based on cited author information and section information. To do this, we propose a multi-anchor bipartite network that uses cited author information and section headings as a multi-anchor that is related to bio-entity nodes, namely the [author/section]-entity bipartite network. To demonstrate the usefulness of the DCS and ICS network, a conventional full-text network is formed for comparison analysis. In addition, during this process, MeSH tree structure is used to examine the bio-entity level characteristics. The results show that DCS and ICS network demonstrate distinct network characteristics and provide unobserved top-ranked bio-entity pairs when compared to traditional method. This indicates that our method can expand the base of entitymetrics and provide new insights for entity level bibliometrics analysis.