next up previous
次へ: Characteristics of S-VSM 上へ: Semantic-Vector Space Model 戻る: (1) Semantic Vector

(2) Semantic Attribute System for Japanese

In order to implement Semantic-VSM, we use the Semantic Attribute System which was recently proposed in "A-Japanese Lexicon"(Ikehara et. al, 1997). This system is shown in part in Fig.[*].

In this system, semantic usage of Japanese words is classified into 2,710 attributes and the relationships among them, namely, "is-a" relation and "has-a" relation, are represented by a tree with 12 levels. A semantic word dictionary is also given in this lexicon, where the semantic usage of 400 thousand Japanese words is defined by using their Semantic Attributes.

Thus, if the frequency of words used in documents is obtained, the values of vector elements $S_i$ in eq.(3) can easily be calculated by summing up such words that have the meaning $\char93 i$.


図 1: Portion of General Noun Semantic Attributes System
\begin{figure}\begin{center} \epsfile{file=zu1.eps,height=7cm} \end{center}\end{figure}


Jin'ichi Murakami 平成13年10月5日