A Television Recommender System Learning a User’s Time-Aware Watching Patterns Using Quadratic Programming

Abstract

In this paper, a novel television (TV) program recommendation method is proposed by merging multiple preferences. We use channels and genres of programs, which is available information in standalone TVs, as features for the recommendation. The proposed method performs multi-time contextual profiling and constructs multiple-time contextual preference matrices of channels and genres. Since multiple preference models are constructed with different time contexts, there can be conflicts among them. In order to effectively merge the preferences with the minimum number of conflicts, we develop a quadratic programming model. The optimization problem is formulated with a minimum number of constraints so that the optimization process is scalable and fast even in a standalone TV with low computational power. Experiments with a real-world dataset prove that the proposed method is more efficient and accurate than other TV recommendation methods. Our method improves recommendation performance by 5-50% compared to the baselines.

Television Rating Control in the Multichannel Environment Using Trend Fuzzy Knowledge Bases and Monitoring Results

“…The selective filtering recommends items on the grounds of a comparison between the items content and the viewers' demand. The aim of program selection is to reduce the system complexity by detecting similar items and merging multiple preferences in the candidate rule set [4,6]. At the collaborative filtering stage, the dimensionality reduction technique, singular value decomposition, is used to find the most similar items in each time and item cluster [7,8].…”

“…Given the primary fuzzy model (6) and (7) and the output classes X kj and Y kj , j = 1, 7, the problem of tuning the composite fuzzy model for content management is formulated as follows [25][26][27]. For each output class and Y kj , j = 1, 7, the solution set (23), (24) should be found which provides the least distance between observed and model fuzzy effects vectors in correlations (6) and 7:…”

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

“…For example, Cremonesi et al [9] divided a day into eight time slots to build a time-channel rating matrix for each user, used the cumulative viewing duration of each time slots as the user viewing preference in that slot, and realized preference prediction by applying Tucker decomposition. Turrin et al [7], Wu et al [25] and Kim et al [29] divided a week into several time slots to capture user preferences in combination with TV program metadata. However, the time-division strategies of the above methods rely on experience and have poor universality, as they only focus on the channel and ignore the features of the current programs at the time of recommendation.…”

“…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.…”

Learning Heterogeneous Temporal Patterns of User Preference for Timely Recommendation

“…Then CoNCARS models the non-linear interaction of the user and the item via convolutional neural networks. [14] uses day of the week and hour information to recommend television programs to users, considering that users tend to watch television programs in periodic patterns in terms of day of the week and hour. However, they may not be able to use their knowledge when the user is interested in the item at a time slightly deviated from the regular routine in which the user normally consumes the item.…”