[enh] paper

report/recommendation-rl.tex

@@ -220,7 +220,7 @@ \subsection{Simulation}
 The simulation consists in the interaction between a recommender and a user. To simulate online learning, ratings from users are known in advance and retrieved at each epoch. 
 We ran each algorithm for several hundred epochs and averaged the results over more than 100 users. 
 
-Results can be found in section \ref{results} \todo{explain the simulation better}
+Results can be found in section \ref{results} .
 
 \subsection{Metrics} \label{metrics}
 
@@ -304,7 +304,7 @@ \subsubsection{Exact bayesian}
 \begin{center}
 \includegraphics[width=0.47\textwidth]{img/bayes_time1.png}
 \caption{Cumulative sum of the time needed to produce the recommendation}
-\label{greedy3}
+\label{bayes_time1}
 \end{center}
 \end{figure}
 
@@ -323,7 +323,7 @@ \subsection{Collaborative filtering and content based comparison}
 \begin{center}
 \includegraphics[width=0.47\textwidth]{img/collabo1.png}
 \caption{Comparison of the average cumulative regret for greedy algorithm on the content based model and collaborative filtering model}
-\label{schema}
+\label{collabo1}
 \end{center}
 \end{figure}
 
@@ -333,7 +333,7 @@ \subsection{Collaborative filtering and content based comparison}
 \begin{center}
 \includegraphics[width=0.47\textwidth]{img/collabo2.png}
 \caption{Comparison of the mean square error for thecontent based model and collaborative filtering model}
-\label{schema}
+\label{collabo2}
 \end{center}
 \end{figure}
 
@@ -343,15 +343,27 @@ \subsection{Collaborative filtering and content based comparison}
 \begin{center}
 \includegraphics[width=0.47\textwidth]{img/collabo3.png}
 \caption{Comparison of the average rating given as feedback by the user for each algorithm}
-\label{schema}
+\label{collabo3}
 \end{center}
 \end{figure}
 
 Maybe the more important graph for the comparison is the one above. It tends to show that there is a significant different on the feedback given by the user depending on the model used for the greedy algorithm. User tends to give a higher feedback to movie suggested by the collaborative filtering approach.
 
 \subsection{Overall performance}
 
-\todo{overall performace}
+To conclude the results section, let's look at the overall performance of all models and algorithms 
+
+\begin{figure}[H]
+\begin{center}
+\includegraphics[width=0.47\textwidth]{img/greedy_all0.png}
+\caption{Comparison of all model and algorithm}
+\label{overallperf}
+\end{center}
+\end{figure}
+
+Obviously, the random random choice is the worst.
+
+The best techniques seems to take a greedy that explore a lot at the beginning and reduce the exploration in profit of exploitation over time while using a collaborative filtering model.
 
 % ------ discussion -------
 \section{Discussion}
@@ -373,8 +385,6 @@ \section{Discussion}
 
 \todo{futur work, real-life data, bias evaluation, big data toussa toussa}
 
-\todo{futur work, real-life data, bias evaluation, big data toussa toussa}
-
 \section{Conclusion}
 
 We built a thorough functional movie recommender system including data mining, features engineering, dimension reduction and several models and content based as well as a collaborative filtering approaches to reinforcement learning.