next up previous
次へ: (2) Semantic Attribute System 上へ: Semantic-Vector Space Model 戻る: Semantic-VSM

(1) Semantic Vector

Here, we propose a new method in which, instead of literal words, meanings of words are used as the vector elements. In this method, the meanings of all of the Japanese words are classified into k categories and they are used as the bases of the specific vector.

Here, let $S_i$ represent the weight of all the words in document $D_j$ which have the meaning of $\char93 i$, the specific vector $V_j$ for the document $D_j$, is written as;

\begin{displaymath}
V_j=(S_1,S_2,\cdot \cdot \cdot S_i,\cdot \cdot \cdot S_k)
\end{displaymath} (3)

Similarly to Word-VSM, there are many ways to determine the values of weight $Wi$. We use the values of $tf \cdot idf$.

Hereafter, we call the specific vector given by (3) as "Semantic-Vector" and the VSM which uses this type of specific vector as "S-VSM" (Semantic Vector Space Model).



Jin'ichi Murakami 平成13年10月5日