摘要 |
PURPOSE:To detect plural most closely related documents from many combined categories and to eliminate the retrieval of categories not supposed to include the most closely related documents. CONSTITUTION:A result of all-text document retrieval by a character string searching processor (SSP) 16 is processed as a vector pattern, i.e., a filed pattern, whose elements become a term coincidence class by the use of a membership function for term coincidence frequency stored in a membership memory MFM 24. A pattern most closely related to an inquired pattern can be detected by similarity between the inquired pattern and each sample of the filed pattern. The similarity is calculated by a similarity calculator PSC 26 based upon fuzzy logic. Since the similarity is ranked in the descending order of similarity, retrieving time can be shortened. The retrieving time can also be shortened by sorting the filed pattern by a set of terms and similarity to a cluster center pattern.
|