15.3.5 Residual inverse document frequency

• RIDF $\rightarrow$ $B(Tabel 15.3 $B$N(B Overestimate $B$_$?$$$J$b$N(B)
  
  $$\begin{eqnarray*}RIDF &=& IDF - \log_2(1-p(0;\lambda_i)) \\ &=& \log_2 \frac{N... ...(0;\lambda_i)) \\ &=& \log_2 \frac{N}{df_i(1-p(0;\lambda_i))} \end{eqnarray*}$$
  
  
  • content word $\rightarrow$ $1-p(0;\lambda_i)$ $B>.(B
  • non-content word $\rightarrow$ $1-p(0;\lambda_i)$ $BBg(B
• content word $B$H(B non-content word $B$K$h$kJ,I[$N0c$$$r5U