Multi-Label Classification of Social Networks’ Users Based on Services

Abstract

Most classification problems assign a single class to each example or instance. However, there exist other classification tasks where each instance can be associated with one or more classes. The latter represents a task known as Multi-label classification. Our work consists in using this notion in the recommender systems field. For this purpose, we have designed a mechanism that extracts the interests of users based on the web services they have accessed to, in order to use that in a multi-label classification architecture. Once the classification model is built, It allows to predict web services that interest new users. Our study was conducted on several datasets of different sizes concerning the number of labels and instances, where the instances represent users’ profiles and the labels are the web services that interest these users. We applied several multi-label classification techniques and compared the obtained results using a variety of evaluation measures, in order to extract the factors that would allow us to obtain better results, as well as the best technique for our case study.