Deprecated: The each() function is deprecated. This message will be suppressed on further calls in /home/zhenxiangba/zhenxiangba.com/public_html/phproxy-improved-master/index.php on line 456

Address: [go: Go Back, main page]
Remove Scripts Accept Cookies Show Images Show Referer Rotate13 Base64 Strip Meta Strip Title Session Cookies

U-PLSI,Q-PLSI

IR に適用する際, U-PLSI, Q-PLSI という2つの手法を用い比較実験

• U-PLSI
  確率ベクトル, $P(w\vert d)$ を用いて VSM を構築
  $$\begin{eqnarray*} \vec{d} = \Bigl\{ P(w_1\vert d),P(w_2\vert d),\ldots, P(w_N\vert d)) \Bigl\} \end{eqnarray*}$$
  
  
  
  
  

  実際の観測値との smoothing

  $$\begin{eqnarray*} P'(w\vert d) &=& \lambda \hat{P(w\vert d)} + (1-\lambda) P(w\... ...{P(w\vert d)} &=& \frac{n(d,w)}{n(d)},\ \ \ 0 \le \lambda \le 1 \end{eqnarray*}$$
  
  
  
  
  

• Q-LSI 圧縮された($K$次元)確率ベクトル, $P(w\vert z),P(d\vert z)$ を用いて VSM を構築
  $$\begin{eqnarray*} \vec{d} = \Bigl\{ \sum_{w\in d} P(w\vert z_1)\rm {idf}(w),\su... ...f}(w), \ldots, \sum_{w \in d} P(w\vert z_K)\rm {idf}(w) \Bigl\} \end{eqnarray*}$$