Resource 2: Explaining why the Recommendation was Made

Wouldn't it be nice to know why your favorite recommender made a recommendation? This paper looks at automated collaborative filtering systems (ACF) and asks the question of how to make a recommender more trusted. If the system creators don't wish to let on what the conceptional model is, then by simply stating the historical accuracy of the recommender for that user may do the trick.

Most of the paper assumes that the recommender creators will allow the user in on the thinking process that is going on. They performed an experiment in which the recommendations were presented in 21 different manners. Some gave the "supposed" reasoning behind the recommedation while others just presented the recommendation in a different format. The best performing presentation was a simple histogram with 3 values(good, bad, neutral). This binary choice beat out a 5 bar histogram (1 - 5 stars). That was unexpected to me. Unfortunately, the link to all 21 representations that were used in the experiment no longer works. The paper is reasonably easy to read and interesting.

The application recommender was for movies (movielens). The website is: http://www.movielens.org/login

"Explaining Collaborative Filtering Recommendations", Herlocker, Konstan, Reidl, 2000 ACM

http://www.grouplens.org/papers/pdf/explain-CSCW.pdf

Resource 1: Blog on a Blog on Recommender Systems

What way to start a blog but to reference a link to another blog site
http://koranteng.blogspot.com/2005/05/on-recommendation-systems.html that deals with recommender systems?! Included in the discussion is a link to an article from the poeple who work on the Amazon recommendation system(http://hugo.csie.ntu.edu.tw/~yjhsu/courses/u2010/papers/Amazon%20Recommendations.pdf). Perhaps I can gleam how to get certain topics out of Amazon's VERY long memory!!!...