Akademik Veri Yönetim Sistemi
8TH International Conference on Advanced Technologies , Sarajevo, Bosna-Hersek, 26 - 30 Ağustos 2019
The main goal of group recommender
systems is to provide appropriate products/services for members of a group.
Various aggregation techniques have been proposed for combining individual
preferences and estimating recommendations based on counts of ratings,
rankings, deviation, and rating averages. Also, the size of the groups and the
recommendation list are essential elements that directly influence the
performance of the aggregation techniques; hence they should be considered when
selecting the aggregation technique to be used in a group recommender system. In
this study, the effectiveness of 11 baseline aggregation techniques regarding
both different group sizes and recommendation lists is deeply analyzed. During
the experimental processes, the groups are constructed using the k-means
clustering algorithm, which helps clustering people with similar interests
together. Experimental studies demonstrate that three aggregation techniques
are successful, mostly when recommending a small number of items. However, they
get reasonably hurt with a large number of recommendations. Also, the
specificity of the groups is a crucial factor for aggregation techniques to
better model group profiles, and they are more successful in small groups of
people.