A fuzzy based recommendation system with collaborative filtering

Recommendation of items is a popular utility in today's social networks and ecommerce sites. The task becomes more critical when high level of accuracy is required. This paper analyzes the improvement in recommendation of movies using Fuzzy logic. The user's choice similarity and acceptance rate is calculated with different similarity measurement approaches, e.g., Euclidean Distance, Manhattan Distance, Pearson Coefficient and Cosine Similarity. To calculate user's choice similarity we take the K most similar users and find the average rating of the target movie given by the similar users. To find acceptance rate, we need to find similar movies to the target movie. To do that, we consider the movies that have the highest number of matching genres against the genres of target movie. Then we consider K most similar movies and calculate the average rating of those similar movies given by the target user. We calculate expected rating using those two parameters and then decision making is made using Mamdani inference systems. We also report performance results of various models based on different similarity measurement techniques, and membership functions with type and number variations.