IBeST: An algorithmic framework for extending item-based collaborative filtering with social tags

Abstract

Collaborative filtering technology is one major method used in recommendation systems. Most existing collaborative filtering algorithms merely use rating data as their prediction input. Social tags have become widely used in web applications which not only reflect the user's personality but also item's properties and semantic meanings. We design an algorithmic framework by extending item-based collaborative filtering with social tags which we call IBeST. IBeST contains the whole lifecycle of the item similarity measurement based on social tags and improves item-based algorithmic results in four phases: dataset preprocessing, metadata injection, algorithm selection and optimization, and similarity weight selection. The calculated similarity is then used in item-based algorithm. MovieLens 10M ratings 100k tags dataset is used in our experiment. IBeST generates improved recommendation ratings than baseline item-based algorithms, and provides a feasible and loosely coupled solution to use social tags in item-based recommendation system.