Dr. C. Selvi

Collaborative Filtering (CF) is a prominent approach to ensure personalized recommendations to active online users. An efficient CF is the memory-based strategy that finds nearest neighbours to an active user using conventional similarity measures. Most such measures deal with a co-rated item rated by a pair of users and hence they are not appropriate to provide an effective recommendation to a sparse dataset having less co-rated items. This study proposes a novel similarity measure, Matusita coefficient in CF (MCF), which considers all ratings given by a user to estimate nearest neighbours. MCF considers local and global rating information provided by users on different rating scales. The performance of the proposed measure is examined and checked by comparing it to conventional measures using popular benchmark datasets like MovieLens and Netflix. The recommendation results demonstrate that the proposed measure outperforms conventional similarity measures on various performance metrics like Mean Absolute Error, Root Mean Squared Error, accuracy, precision, recall and coverage.

The enormous growth of Internet facilities, the user finds difficulties in choosing the music based on their current mindset. The context-aware recommendation has turned out to be well-established technique, which recommends a music based on the mindset of the user in various contexts. To enhance the potential of music recommendation, the emotion and time interval are considered as the most important context. Emotion context is not explored due to the difficulty in acquisition of emotions from user's microblogs on the particular music. This paper proposes an algorithm to extract the emotions of a user from microblog during a different time interval and represented at different granularity levels. Each music piece crawled from online YouTube repository is represented in a triplet format: (User{\_}id, Emotion{\_}vector, Music{\_}id). These triplet associations are considered for developing several emotion-aware techniques to provide music recommendations. Several trial of experimentation demonstrates that the proposed method with user emotional context enhances the recommendation performance in terms of hit rate, precision, recall, and F1-measure.