Personalized and Context-Aware TV Program Recommendations Based on Implicit Feedback

Abstract

The current explosion of the number of available channels is making the choice of the program to watch an experience more and more difficult for TV viewers. Such a huge amount obliges the users to spend a lot of time in consulting TV guides and reading synopsis, with a heavy risk of even missing what really would have interested them. In this paper we confront this problem by developing a recommender system for TV programs. Recommender systems have been widely studied in the video-on-demand field, but the TV domain poses its own challenges which make the traditional video-on-demand techniques not suitable. In more detail, we propose recommendation algorithms relying exclusively on implicit feedback and leveraging context information. An extensive evaluation on a real TV dataset proves the effectiveness of our approach, and in particular the importance of the context in providing TV program recommendations

The Importance of Context When Recommending TV Content: Dataset and Algorithms

“…They smooth temporal context and use distance between contextual settings to recommend TV programs. [30] presents a comparable dataset, but includes familiar context, that is, the additional users watching. Their results suggest that temporal context cancels the effect of social context when using both to recommend TV content.…”

Efficiently using contextual influence to recommend new items to ephemeral groups

“…Then, when needed, the preferences for the items can be easily computed on the fly exploiting the preferences for the item attributes. A state-of-the-art content-based technique conceived for the TV domain is that proposed in [15], which will be employed in the experimental evaluation of Section 5. This technique builds a multidimensional tensor where the dimensions are represented by users, context dimensions and item attributes.…”

“…The preference functions adopted to determine the contextual and non-contextual scores of the items for the single users in the TV dataset are those recently proposed in [15], a state-of-the-art technique to generate TV program recommendations for individual users relying on implicit feedback. It is a contentbased method, thus allows to compute users' preference values also for new items.…”

Facing Cold-Start: A Live TV Recommender System Based on Neural Networks

“…RSs cannot recommend new programs due to the cold-start [8] problem. Second, families constitute the main audience of live TV, and multiple family members share a terminal [9], [10]; thus, an RS needs to identify the preferences of multiple people. Third, an audience shows its preference for a program via implicit feedback (e.g., viewing duration) rather than explicit feedback (e.g., rating).…”

“…Content-based (CB) algorithms [16]- [19], context-based algorithms [20], and social network-based algorithms [21], [22] can address such problems, but these algorithms require different amounts of additional information. RC methods rely on the time factor to establish a channel-time correlation, and they convert the program preferences of users to the channel preferences of users [7], [9], [11], [23]- [29]. Such methods can handle the cold-start problem and have good real-time performance.…”