Clustering of Text Streams via Facility Location and Spherical K-means

Spherical k-means is a fast and effective method for clustering text documents in their directional representation over a unit hypersphere. Current needs of text clustering are related to clustering of streams. Due to memory restrictions, fast and effective methods are required that incur less space complexity. Few research works exist that have adapted spherical k-means to streaming text data, but recorded performance is not satisfactory for novelty detection imbalanced cluster structure. This paper presents streaming spherical k-means with associated facility location costs. Arriving documents are detected as new topic or join an existing depending on these costs.