Context Similarity Measurement Based on Genetic Algorithm for Improved Recommendations

Abstract

Recommender Systems (RSs) are new types of internet-based software tools, used to provide personalized recommendations to users by handling information overload problem on the Web. Collaborative Filtering (CF), the most known and commonly used recommendation technique in the domain of RSs, generates recommendations toward items which were preferred by other like-minded users in the past. Computing similarity among users efficiently in case of sparse data is the major concern of CF technique. A recent study observed that Context Awareness in CF (CACF) is the next generation of the traditional user–item RSs which provides more accurate and relevant situational recommendations by incorporating contextual ratings given by the user. In this work, we first extend two-dimensional RSs by incorporating contextual information into fuzzy CF user profile (CA-FCF) through contextual rating count approach. Second, we employ genetic algorithm into CACF (GA-CA-FCF) to learn user preferences on individual hybrid user features. By learning the weights on each feature, the user similarity is computed efficiently. We evaluate our approach with Mean Absolute Error (MAE) and coverage performance measures using LDOS-CoMoDa dataset. Experimental results show that our approach has an acceptable improvement in the accuracy with comparisons to classical CF approaches.