An intelligent similarity measure for effective text document clustering
Document Clustering is an important tool for many Information Retrieval (IR) tasks. The huge increase in amount of information present on Web poses new challenges in clustering regarding to underlying data model and nature of clustering algorithm. Document clustering techniques mostly rely on single term analysis of document data set. To achieve more accurate document clustering, more informative feature such as phrases are important in this scenario. Hence first part of the paper presents phrase-based model, Document Index Graph (DIG), which allows incremental phrase-based encoding of documents and efficient phrase matching. It emphasizes on effectiveness of phrase-based similarity measure over traditional single term based similarities. In the second part, a Document Index Graph based Clustering (DIGBC) algorithm is proposed to enhance the DIG model for incremental and soft clustering. This algorithm incrementally clusters documents based on proposed cluster-document similarity measure. It allows assignment of a document to more than one cluster. The DIGBC algorithm is more efficient as compared to existing clustering algorithms such as single pass, K-NN and Hierarchical Agglomerative Clustering (HAC) algorithm.