Identifying Networks of Semantically-Similar Individuals from Public Discussion Forums

In identifying communities in the online environment most approaches consider as the basic tie that connects social actors together some form of direct contact, such as through communication. Other approaches use surrogates for direct ties including copresense, cooccurrence, or structural equivalence. In contrast, this paper focuses on semantic equivalence among social actors, regardless of their direct contact. In particular, to index semantic similarity, it measures the entire semantic network across the body of messages an individual produces and compares that network to another person's to index how similar they are. Then it uses this similarity coefficient as the social network tie for network analysis to identify communities of semantic practice. Semantic similarity has some unique value for theory and practice in automated social network analysis. To illustrate this approach, this research extracted all 10,001 posts from a public discussion forum authored by 3,272 individuals and represented each author's semantic network based on cooccurrences of all word pairs within three word positions. Pearson correlation coefficients were computed for 5.36 million pairs of individuals using Quadratic Assignment Procedures (QAP). Authors sharing approximately 50% of their semantic networks numbered 22. Subsequent network analysis found that they constituted a single group in terms of a community of linguistic practice. A different forum was analyzed as a contrast. Applications of such a procedure can test hypotheses about semantic network similarity in relation to variations in communication frequency and modality More practical purposes would include finding persons of interest to add to a watch list.