Permalink
Browse files

fix language; update refs

  • Loading branch information...
1 parent bdced56 commit 1dcc1afc4968a23c8a40a5563aa153d6bfe225fe @olav committed Jun 8, 2011
Showing with 4 additions and 4 deletions.
  1. BIN paper/dist/paper.pdf
  2. +2 −2 paper/src/adaptive.recommenders.tex
  3. +2 −2 paper/src/introduction.tex
View
Binary file not shown.
@@ -87,7 +87,7 @@ \section{Adaptive Recommenders}
based on a generalized combination found by minimizing some error across all users.
With adaptive recommenders, the $Adapters$ are themselves user modeling methods.
-However, instead of modeling users, we wish to model the behavior of recommender systems.
+Instead of modeling users, we wish to model the behavior of recommender systems.
More specifically, we wish to model the \emph{accuracy} of the recommender systems.
Methods in this second layer are used to predict how accurate each of their corresponding basic recommenders will be.
It is these methods that will allow us to do adaptive aggregation based on the current user and item.
@@ -101,7 +101,7 @@ \subsection{Adaptive Aggregation}
To perform adaptive aggregation, we need the $Adapters$ to be actual recommender systems.
The simplest generalized way of prediction aggregation is to take the average of all predictions made
by the different methods (e.g. \cite[p.3]{Aslam2001}).
-However, many aggregators attempt to weigh each method differently (e.g. \cite{Claypool1999}):
+Many aggregators attempt to weigh each method differently (e.g. \cite{Claypool1999}):
\begin{eqnarray*}
\hat{r}_{u,i} = \sum_{m \in M} w_{m} \times p(m,u,i).
@@ -9,7 +9,7 @@ \section{Introduction}
Our ability to make properly informed decisions is often the first thing to go
\cite[p.1]{Davenport2001}.
-However, while people struggle with excessive information,
+While people struggle with excessive information,
many algorithms in the field of artificial intelligence
can increase their performance by accessing more information.
\cite{Halevy2009} calls this the ``unreasonable effectiveness of data''.
@@ -42,7 +42,7 @@ \section{Introduction}
If we can accurately predict how each user will react to each item,
we will have come a long way towards solving information overload.
-However, despite their apparent power, recommender systems are often confined
+Despite their apparent power, recommender systems are often confined
to simple tasks like creating small lists of recommended items
or computing similar items to the ones being considered.
Common examples are lists of recommended items based on the one being viewed,

0 comments on commit 1dcc1af

Please sign in to comment.