Effects of Fuzzy Theoretic Inference Strategy and Similarity Measure on a Recommender System

There is a need for modeling and reasoning on the subjective, incomplete, imprecise and vague nature of item features, user behavior and their relationships in machine learning algorithms. This paper presents results of analysis of effects of fuzzy theoretic inference/recommendation strategies and fuzzy theoretic similarity measures on a recommender system's accuracy-measured in terms of precision, recall and F1-measure. The study uses the MovieLens benchmark dataset on movies that are widely used in recommender system research, and conduct simulation runs. Finally, a guideline for recommender system designer to choose a combination of inference strategy and similarity measure for personalized recommender systems is presented based on statistical analysis of the simulation results.