A context-aware method for top-k recommendation in smart TV

Abstract

We discuss the video recommendation for smart TV, an increasingly popular media service that provides online videos by TV sets. We propose an effective video recommendation model for smart TV service (RSTV) based on the developed Latent Dirichlet allocation (LDA) to make personalized top-k video recommendation. In addition, we present proper solutions for some critical problems of the smart TV recommender system, such as sparsity problem and contextual computing. Our analysis is conducted using a real world dataset gathered from Hisense smart TV platform, JuHaoKan Video-on-Demand dataset(JHKVoD), which is an implicit watch-log dataset collecting sets of videos watched by each user with their corresponding timestamps. We fully portray our dataset in many respects, and provide details on the experimentation and evaluation framework. Result shows that RSTV performs better comparing to many other baselines.We analyse the influence of some of the parameters as well as the contextual granularity.