Improving the Keyword Co-occurrence Analysis: An Integrated Semantic Similarity Approach

Abstract

Bibliometric studies help yield useful information about the quantity and quality of research works in a particular academic domain. We used the popular bibliometric technique of co-occurrence analysis to explore emergent topical areas in the field of urban mobility. Our work contributes from a methodological perspective to improve the conventional co-occurrence analysis method. The modified co-occurrence analysis is two-fold: firstly, we used the Rapid Automatic Keyword Extraction (RAKE) algorithm to extract keywords from abstracts of documents in a corpus to generate a new co-occurrence matrix. Secondly, we produced “semantic similarity” between each keyword in matrix form, which is combined with the co-occurrence matrix of the extracted keywords from documents in a corpus to yield a weighted co-occurrence matrix. We analyzed the unweighted and weighted matrices in terms of their network structure and cluster quality. We demonstrated that the weighted matrix shows network structures with higher modularity and superior cluster quality than its unweighted counterpart. These observations are consistent in terms of more meaningful content and greater ease of exposition in the emergent themes.