Document Clustering with Explicit Semantic Analysis (ESA)

Document clustering recently became a vital approach as numbers of documents on web and on proprietary repositories are increased in unprecedented manner. The documents that are written in human language generally contain some context and usage of words mainly dependent upon the same context; recently researchers have attempted to enrich document representation via external knowledge base. This can facilitate the contextual information in the clustering process. An enrichment process with explicit content analysis using Wikipedia as knowledge base has been proposed. The approach is distinct in the sense that only the conceptual words from a document were used and their frequency to embed the contextual information. Hence, the approach does not over enrich the documents. A vector based representation, with cosine similarity and agglomerative hierarchical clustering is used to perform actual document clustering. The proposed method was compared with existing relevant approaches on NEWS20 dataset, with evaluation measure for clustering including FScore, Entropy and Purity.