Leveraging movie recommendation using fuzzy emotion features

Abstract

User generated data like reviews encapsulate bursts of emotions that are generated after reading a book or watching a movie. These emotions garnered from reviews can be used in recommending items from entertainment domain with similar emotions. Till now much work has been done using emotions as discrete features in recommender system. In this paper, we delve deeper to use fuzziness in the emotion categories. The use of emotional features such as love, joy, surprise, anger, sadness and fear has been shown to be effective in identifying items with similar features for recommendations. However, there is a certain degree of vagueness and blurring boundaries between the lexicons of these categorical emotion features that has hitherto been largely ignored. In this paper, we tackle the problem of inherent vagueness in emotional features by proposing a framework for movie recommendation using fuzzy emotion features by taking each emotion category as a linguistic variable. We develop a Mamdani Model to extract fuzzy classification rules for recommending movies from emotions extracted from their corresponding reviews. Results show that Guassian fuzzy model with 5 linguistic variables yield 68.43% F-measure which is 10.5% improvement over the SVM based crisp model for recommending movies.