Browse files

use term in five-fold; sum up results chapter

  • Loading branch information...
1 parent 5b1386b commit e2d09619f26ec25159c627b7c60dee92e9525e93 @olav committed Jun 8, 2011
Showing with 12 additions and 2 deletions.
  1. +2 −2 thesis/src/results.prediction.agg.tex
  2. +10 −0 thesis/src/results.rank.agg.tex
@@ -17,7 +17,7 @@ \section{Prediction Aggregation}
In order to verify these hypotheses, we performed adaptive prediction aggregation on the two datasets previously described.
-5-fold cross validation was performed to further verify each result.
+Five-fold cross validation was performed to further verify each result.
Table \ref{table:results:e1} gives the results from Experiment 1 (MovieLens).
Table \ref{table:results:e2} gives the results from Experiment 2 (Jester).
@@ -74,7 +74,7 @@ \section{Prediction Aggregation}
While we can not generalize too much on this basis,
the fact that this dataset is a common testing ground for recommender systems,
that RMSE is the de facto measure for determining performance,
-and because of our 5-fold cross-validation, the results allow us
+and because of our five-fold cross-validation, the results allow us
to confirm hypothesis H1 in these conditions, and likely for other, similar scenarios.
We shall discuss this result in Chapter \ref{chap:discussion}.
@@ -274,6 +274,16 @@ \subsection{Adaptive IR Weights}
can be used to provide personalized search}.
This positive result for Experiment 3 confirms hypothesis H3,
at least for this dataset, this IR system and our chosen recommender algorithms.
+By confirming H3, we have shown that adaptive recommenders can be used for personalized search.
+This results in a search engine results page that inherits the strenghts of adaptive recommenders.
+Each item on the result list is ranked not just based on query matching,
+but based on a number of signals, represented by recommender systems.
+Each signal is adaptively used based on how well it suits the current user,
+and how well it has worked in the past for each item.
+This adaptive results page will hopefully help mitigate the latent subjectivity problem,
+by ranking each element based on the current context.
We will discuss this further in the next chapter.

0 comments on commit e2d0961

Please sign in to comment.