Document Ranking Method for High Precision Rate

Document Ranking Method for High Precision Rate Mee-Sun Jeon and Se-Young Park Natural Language Processing Section Electronics and Telecommunications Research Institute TaeJon, Korea {msjun,sypark}@com.etri.re.kr Abstract Many information retrieval(IR) systems retrieve relevant documents based on exact matching of keywords between a query and documents. This method degrades precision rate. In order to solve the problem, we collected semantically related words and assigned semantic relationships used in general thesaurus and a special relationship called keyfact term(FT) manually. In addition to the semantic knowledge, we automatically constructed statistic knowledge based on the concept of mutual information. Keyfact is an extended concept of keyword represented by noun and compound noun. Keyfact can be a verb and an adjective including subject or object term. We first retrieved relevant documents with original query using tf * idf weighting formula and then an expanded query including keyfacts is used in both second document ranking and word sense disambiguating. So we made an improvement in precision rate using keyfact network. 1 Introduction The general users are more interested in concepts rather than words itself. But many commercial IR systems retrieve relevant documents based on keyword string matching between a query and documents. There are two problems in using the method. The first problem is that words are ambiguous, and this ambiguity is causative of retrieving irrelevant document semantically. Therefore lexical ambiguity has to be resolved. The second problem is that a document is treated as a irrelevant document in spite of a relevant document, for the document does not include the same words as query terms. So an original query has to be expanded to semantically related words. In order to solve the problems, we consider keyfacts as well as keywords. A keyfact is an extended concept of a keyword and can be defined as a verb and an adjective which are every probability that occurs in several times based on threshold value. We collect semantically related keyfacts from an encyclopedia and assigned semantic relationships using in general thesaurus and a special relationship of FT. We use the semantic informations for document ranking and word sense disambiguation. The remainder of this paper is organized as follows. Section 2 gives a definition of the KN and the KN construction method. Section 3 describes word sense disambiguation. The results of performance comparison are presented in section 4. The concluding remarks are described in section 5. 2 Keyfact Network 2.1 The Definition of KN In this paper, we collected words and their semantically related words from an encyclopedia(ENCY). Collected informations are used to understand user's request. The KN consists of nodes and edges. Nodes are defined as words rather than word sense and the edges represent binary relationships, such as BT(Broader Term), NT(Narrow Term), RT(Related Term), HP(Has Part), UF(Used For), and FT(keyFact Term). FT is defined as relationship 255 between noun and verb or adjective. Most IR systems have used to the relationships except FT relationship in their thesaurus. FT is effectively used to resolve lexical ambiguity and compute query-document similarity. …

Content truncated to 500 words