Showing 1–1 of 1 results for author: de Lima, A P
-
Limits to Surprise in Recommender Systems
Authors:
Andre Paulino de Lima,
Sarajane Marques Peres
Abstract:
In this study, we address the challenge of measuring the ability of a recommender system to make surprising recommendations. Although current evaluation methods make it possible to determine if two algorithms can make recommendations with a significant difference in their average surprise measure, it could be of interest to our community to know how competent an algorithm is at embedding surprise…
▽ More
In this study, we address the challenge of measuring the ability of a recommender system to make surprising recommendations. Although current evaluation methods make it possible to determine if two algorithms can make recommendations with a significant difference in their average surprise measure, it could be of interest to our community to know how competent an algorithm is at embedding surprise in its recommendations, without having to resort to making a direct comparison with another algorithm. We argue that a) surprise is a finite resource in a recommender system, b) there is a limit to how much surprise any algorithm can embed in a recommendation, and c) this limit can provide us with a scale against which the performance of any algorithm can be measured. By exploring these ideas, it is possible to define the concepts of maximum and minimum potential surprise and design a surprise metric called "normalised surprise" that employs these limits to potential surprise. Two experiments were conducted to test the proposed metric. The aim of the first was to validate the quality of the estimates of minimum and maximum potential surprise produced by a greedy algorithm. The purpose of the second experiment was to analyse the behaviour of the proposed metric using the MovieLens dataset. The results confirmed the behaviour that was expected, and showed that the proposed surprise metric is both effective and consistent for differing choices of recommendation algorithms, data representations and distance functions.
△ Less
Submitted 10 July, 2018;
originally announced July 2018.
