A Novel Recommendation Approach For Groups Based On Aggregating Top-k Lists

Abstract

One of the most challenging issues in group recommendation is how to consider the dispersion in members’ preferences to reach a joint recommendation satisfying all the group. The group recommendation is generated by aggregating either the users’ preferences or the individual recommendations. In this paper, we focus on the recommendations aggregation strategy which consists in generating a recommendation list for each group member, then combining these individual lists to produce a single group recommendation list. Many aggregation functions (such as average, least misery, Borda, etc.) were used to resolve the ranking aggregation problem. However, they cannot support the partial information presented over the top-k recommendation lists which may not contain the same set of alternatives. To this end, we propose a group recommendation method based on a novel aggregation function that considers the partial rankings and computes each item's relevance to the target group based on both the users’ preferences and the items’ positions. We conduct a deep experiment to study the impact of some parameters (such as the individual recommendation algorithms, the aggregation functions, and the group size) on the group recommendations quality. We have shown that the accuracy of the group recommendations does not depend on the algorithm used to generate individual recommendations. However, it is strongly affected by the method used to aggregate data and the group size. Experimental results show a considerable improvement in the group recommendation performance using the proposed solution.