Research on the Selection of Feature Transfer Relations in Latent Semantic Indexing

Abstract The latent semantic index (LSI) has been widely used in many fields of natural language processing in which co- occurrence features can be captured by the transfer relations between the documents and in the documents. Document features with a higher frequency in the collection of the document are more likely to introduce some unreasonable feature transfer relations to the latent semantic space which affects the similarity between features and between documents in document sets in our recent study. In the paper a feature optimize technology in latent semantic indexing that uses feature transfer relation in documents and between documents is proposed. By the complete-link algorithm, the experimental results show that the method effectively improves the performance of latent semantic indexing.