An Automatic Semantic Term-Network
Soon Cheol Park, Lim Cheon Choi · 2009
An automatic semantic term-network construction system using the singular value decomposition (SVD) is implemented in this research. The term-network construction is to compute the similarities between rows of the large term-by-document matrix generated from a document corpus to get the relationships between terms. A reduced matrix, U of SVD, is decomposed from the term-by-document matrix to improve the speed and provide the latent semantic structure. The SSTRESS criterion is used for the numerical measure of closeness between original term by document corpus matrix and the decomposition matrix with different ranks. In order to measure the performance of our system, a standard data collection set, Reuters-21578, is used and about 2000 terms are extracted from the set automatically. This term-network construction could be expected to easily apply to constructing the ontology system and to supporting the semantic retrieval system in the near future.