Using learning to rank for a top-n recommendation system in TV domain

Acar, Bedia
In this thesis, a top-N recommendation system in TV domain is proposed using learning to rank. The design, development and evaluation of the proposed recommender system are described in detail. Instead of calculating rating score of items like in conventional recommender systems, the ranked recommendation item list is presented to TV users. Moreover, path-based features which are used to build ranking model is explained in detail. These features provide collaborative filtering, content-based filtering and context aware recommendation system. Furthermore, some state of the art learning to rank approaches from each category called as pointwise, pairwise and listwise have been experimented to generate a ranking model. Then a baseline which does not use any learning are compared with the one using learning to rank algorithm. It is shown that the model constructed with learning to rank algorithm gives better results.
Citation Formats
B. Acar, “Using learning to rank for a top-n recommendation system in TV domain,” M.S. - Master of Science, Middle East Technical University, 2016.