Off-line Evaluation of Recommendation Functions
Full Text:
um2005.pdf
This paper proposes a novel method for assessing the performance of any Webrecommendation function (i.e., user model), M, used in a Web recommendersytem, based on an off-line computation using labeled session data. Eachlabeled session consists of a sequence of Web pages followed by a page p
(IC)that contains information the user claims is relevant. We then apply M toproduce a corresponding suggested page p
(S).In general, we say that M is goodif p
(S) has content similar to the associated p
(IC), based on the the samesession. This paper defines a number of functions for estimating this p
(S) top
(IC) similarity that can be used to evaluate any new models off-line, andprovides empirical data to demonstrate that evaluations based on thesesimilarity functions match our intuitions.
Citation
T. Zhu,
R. Greiner,
G. Haeubl,
K. Jewell,
B. Price.
"Off-line Evaluation of Recommendation Functions".
User Modeling (UM), Edinburgh, UK, July 2005.
Keywords: |
WebIC, machine learning |
Category: |
In Conference |
BibTeX
@incollection{Zhu+al:UM05,
author = {Tingshao Zhu and Russ Greiner and Gerald Haeubl and Kevin Jewell
and Bob Price},
title = {Off-line Evaluation of Recommendation Functions},
booktitle = {User Modeling (UM)},
year = 2005,
}
Last Updated: April 23, 2007
Submitted by Nelson Loyola