Comparison of measures of collaborative filtering recommender systems: rating prediction accuracy versus usage prediction accuracy
Rohit Rohit, Anil Kumar Singh · 2017
Recommender Systems (RSs) help users to find items of their choices. There are many types of RSs that produce good recommendations for the user. Every system designer tries to choose the best RS for their application. There is a basic question after each recommendation computation. Is the recommended item liked by the user? The whole process to compute recommendation is meaningful, if the recommended items suit the choices of the user. The selection of best RS for an application is a bigger challenge for the system designer. RSs are compared on their evaluation matrices. This paper compares the evaluation measures for collaborative filtering RSs. The accuracy measures are Measuring Ratings Predicting Accuracy (MRPA) and Measuring Usage Prediction Accuracy (MUPA). Furthermore, the advantage and disadvantage of both accuracy measures are also examined through an experiment. The machine learning libraries of Mahout is used for the setup of an experiment. The experiment is conducted with the MovieLense dataset.