Automatic trend detection: Time-biased document clustering

Abstract Identifying the trending topics in journals and conferences is valuable for understanding role of authors, institutions, funding agencies progression knowledge produced field. However, many available clustering methods do not accommodate a desire temporally clustered results that are typical trends, part because time publication often neglected as feature. As demonstration how can be emphasized trend detection, we use novel approach introducing weighted temporal feature to bias topic toward articles similar frame; this performed over set finance journal abstracts from 1974 2020. Latent Dirichlet Allocation (LDA) used parameterize each abstract, followed by dimensionality reduction using Singular Value Decomposition (SVD). We detect identifiable when standard with no bias. To identify topics, utilize metric silhouette score divided deviation clusters time. then isolate identified validate them expert judgment. Our strategy readily utilized other fields discovering rise fall trends.