Social Network Science Approaches for Disease Named Entity Recognition and Extraction

Conventional machine learning approaches adopted for large-scale social media analysis have encountered significant limitations in capturing the underlying dynamics, evolution, and semantic nuances of user posts, hindering comprehensive analysis for tasks related to population health analytics. In this article, the integration of network science-based techniques for node importance/influence analysis, and, Transformer models for Named Entity Recognition are proposed, to facilitate the extraction of structured knowledge from social network posts for population health analytics applications. Standard datasets comprising user account details and postsare considered for the experiments, which are first transformed into graph representations suitable for both structural and behavioral analytics. To evaluate the node importance/influence, different centrality measures were employed and compared. Additionally, a comparative study to assess the impact of varying network sizes by manipulating the number of nodes within the network is conducted. Large-scale mining of disease mentions as a named entity recognition task is also attempted, using neural language models. The proposed approach achieved promising results, outperforming state-of-theart works by 14.7% in terms of f1-score.