Stream-Based Classification For Social Network Recommendation Systems

Social media proliferated increasingly over the past years. Many companies are keen to tap on its benefits through data mining the insights from the social media. One of the challenges associates with data mining social media is the ever expanding volume of such social media. Often the social data are in formats of live feeds and they are unbounded. One alternative method to data mining social media is treating the training dataset as an infinite data stream, and the model induction is done in incremental manner. A classical data stream mining called Very-Fast-Decision-Tree (VFDT) invented in 2000 has been popularly adopted. In this book chapter, an improved version of VFDT, namely Optimised VFDT or OVFDT is used to data mine social media. Aspecific case of an online recommender is considered where the online users' votes and opinions are taken as training samples. The recommender is powered by a classifier which is to be induced by OVFDT. Experiments are conducted for comparing the efficacy of OVFDT and some variants of VFDT. OVFDT shows superiority in performance indicating that it is suitable for mining social media in the case of online recommender.