The effect of information retrieval method using dependency relationship between words
Atsushi Matsumura, Atsuhiro Takasu, Jun Adachi · 2000
It is difficult to improve retrieval effectiveness by using only keyword-based retrieval due to its high dependence on statistical word distributions. We propose a method that enhances retrieval effectiveness using dependency relationships between words in sentences of documents and queries. In our method, we create a Structured Index, represented by a binary tree, in three steps. These steps are morphological analysis, dependency analysis based on a title template, and compound nouns analysis based on word bigram statistics. In the retrieval phase, a query is structured in the form of a binary tree by the same process as indexing and is compared with a Structured Index by shrinking binary trees according to the rules of dependency relationships of Japanese. We also define the similarity between a binary tree of a query and a set of binary trees of a document. From retrieval experiments with NTCIR--1, which is a large-scale Japanese test collection for information retrieval, we show t...