An effective approach of feature selection for recommender systems using fuzzy C means clustering along with ant colony optimization and neural networks

Ratings provided by user for a movie or TV program also depends on demographic information, day type, time, mood, and other factors of environment. Thus, context aware recommender systems are used for providing recommendations. As recommender systems suffers from cold start, sparse data and scalability problems. Therefore, artificial neural networks (ANN) are used to solve these problems. But, training of ANN again takes a very large time, due to high dimensional features. Hence, feature subset selection is done to solve this problem. But, it is a kind of combinatorial optimization problem, which could be solved by using Ant Colony Optimization (ACO). ACO with heuristic function as fuzzy values obtained after applying fuzzy c means clustering algorithm over movie lens data set provided optimum results. Back propagation algorithm has been used, for the training of neural network, only on those features which are obtained after applying modified ACO along with roulette wheel selection algorithm. Results are analyzed by providing ratings of those movies which are not provided in “movie lens data set” yet, as a solution to cold start problem. Finally, the accuracy of recommender systems is calculated by determining mean absolute error and comparison is provided with other previous approaches for different data sets.