diff --git a/src/year2/natural-language-processing/sections/_rag.tex b/src/year2/natural-language-processing/sections/_rag.tex index dcd37b2..e14242e 100644 --- a/src/year2/natural-language-processing/sections/_rag.tex +++ b/src/year2/natural-language-processing/sections/_rag.tex @@ -101,7 +101,7 @@ \end{itemize} It is defined as follows: \[ - \texttt{BM25}(t, d) = + \texttt{BM25}(q, d) = \sum_{t \in q} \left( \texttt{idf}(t) \frac{\texttt{tf}(t, d) \cdot (k+1)}{k \cdot \left( 1 - b + b \frac{|d|}{|d_\text{avg}|} \right) + \texttt{tf}(t, d)} \right) \] where $|d_\text{avg}|$ is the average document length, and typically $k \in [1.2, 2]$ and $b = 0.75$. @@ -260,7 +260,7 @@ \begin{description} \item[Exact match] \marginnote{Exact match} - Ratio of matches between predicted answer and the ground-truth computed considering the characters at each position. + Ratio of match between predicted answer and the ground-truth computed considering the characters at each position. \item[F1 score] \marginnote{F1 score} Macro F1 score computed by considering predictions and ground-truth as bag of tokens (i.e., average token overlap).