Applying Latent Semantic Indexing to Information Retrieval System
LSI潜在语义标引方法在情报检索中的应用

This paper presents a new method of automatic indexing and retrieval.The approach is to take advantage of terms with documents (“latent semantic-structure”)in order to improve the detection of relevent documents on the basis of terms found in queries.A particular technique used is singular-value decomposition in which a large term-document matrix is decomposed into a set of korthogonal factors.The original matrix can be approximated by linear combination from the factors set.Documents and queries are represented as vectors for med from weighted combinations of these factors. The relevancy prediction is achieved by comput ing the similarity of query and documents.