Predicting links in a social network based on recognised personalities

Abstract

The problem of link prediction concerns the inference of network edges. Its adaptation for the task of friendship recommendation has been gaining considerable attention from computer scientists and psychologists alike, especially within the context of online social networks (OSNs). Although one’s personality is known to influence social relationships, its impact within OSNs is oftentimes overlooked. The main components in this research are the personality recognition tool and the link prediction boosting component, referred to as PALP-boost. The former extracts a user’s personality traits from their set of micro-blog postings, while the latter incorporates personality preferences towards their followee recommendations. The best performing personality recogniser model was found to be a Support Vector Machine with a Pearson VII function kernel, attaining a Mean Absolute Error of 0.105. The model achieved competitive results to previous state-of-the-art approaches. A real-world Twitter network was utilised to evaluate PALP-boost. Using the developed personality recogniser, user personalities were extracted from their collected tweets. Results suggested that the PALP-boost component enhanced link prediction models by 10% (in terms of Average Precision) in the best case.