An Evolutionary Based Recommendation Approach

A recommender system aims to satisfy its users by offering them relevant items (films, books, products, etc). This is done by comparing the items deemed satisfactory by the user with all of the items (content-based filtering), or by searching for similar users (collaborative filtering). We propose a genetic based approach to recommend relevant items without needing an explicit request from the user. Our approach looks for the most similar users using a genetic algorithm. Then, a recommendation space is constructed by grouping the items preferred by each similar user, and removing those preferred by the active user. After that, the system predicts a rating for each unrated item. The recommendation space will be reduced by keeping only relevant items using a threshold. An experimental study has been made in comparison with KNN algorithm. Experimental results seem interesting and show an improvement in precision, recall and F-Measure. Also Mean Absolute Error (MAE) has been reduced.