A Comparative Study between Collaborative Filtering Techniques and Generate Personalized Story Recommendations for the Vixio Application

Abstract

Interactive fiction (or text-based game) is a game that consists of texts which are used to bring interactivity to a story. Interactive fiction shows the potential to improve reading behaviour and engage the player with reading materials. In continuing to explore more benefits in reading, creating, and sharing interactive fiction, a web application called Vixio is developed as a platform, where users can develop and distribute interactive fiction. To engage and to feed the users with more interactive stories, a recommender system is applied to provide recommendations of stories that would be suitable to the reader’s interest. This paper is focused on developing a recommender system which can generate personalized story recommendations for the Vixio web application. This paper also discusses determining which techniques are better to be implemented inside the recommender system by conducting a comparative study between five collaborative filtering techniques, which are: Three Matrix Factorizations (SVD, SVD++, and NMF), Slope One, and Co-clustering. To compare each technique with one another, 5-fold cross-validation and response time were measured. Based on these two evaluations, it is shown that there is no technique which has a superior accuracy over the others. However, Slope One algorithm is eminent in terms of fit time and mean response time.