Truncated singular value decomposition for semantic-based data retrieval

This paper addresses the increasingly encountered challenge of knowledge indexation. In the past decade, research on numerical schemes on knowledge indexation has been quite intensive. Vector space model is only based on the information contained in term weighting and does therefore not process the semantic contained in the sequence in which the words appear in a bag-of-words. This representation provides an abstraction of semantic relations between different linguistic units. A novel semantic-based method for knowledge indexation, which can provide improvement in both indexing and retrieval, is described. Despite a huge dimension in vector space model size, retrieval accuracies are seen to improve significantly when the proposed system is applied for indexing Reuters-21578 corpus.