Evaluation of Hybrid Collaborative Filtering Approach with Context-Sensitive Recommendation System

Abstract

This study aims to evaluate the recommendation system we have implemented. The test was carried out using student profile data with (1) adding contextual information and (2) without contextual information. The evaluation was carried out using three predictive methods, user collaborative filtering, item collaborative filtering, and hybrid. We use the correlation coefficient to determine which method has the best correlation coefficient between the predicted and actual values. In this experiment, the best method for predicting the actual value is to use a user-based collaborative filtering method. A low correlation coefficient indicates that a machine learning model is needed to learn the predictive formula of the predictor features and their actual values. The test results using these three methods show that the correlation coefficient between the actual value and the predicted value using user-based collaborative filtering is the highest. Meanwhile, the lowest correlation coefficient between the actual value and predicted values using item collaborative filtering is the lowest. The results of this study prove that contextual information as an additional feature of student profiles increases the correlation coefficient between actual and predicted scores using a user collaborative filter.