Learning User Behaviour in a Pervasive Social Networking System

Social networking systems and pervasive computing are two essential paradigms for systems of the future. There has been an increasing amount of research and development done on combining location awareness with social networking. Our current research is aimed at taking this a step further and combining more general pervasive system behaviour with social networking in a fully integrated way. In order to achieve this, one of the key functionalities on which the system is based, is that of context aware personalization. However, one of the major problems with personalization lies in dealing with the changeability of user preferences, and this needs to be taken into account when choosing a strategy to handle learning of user preferences. This paper presents an approach that we have been developing, which uses two different strategies in tandem – one based on a rule-based approach, the other on a neural network with which a user can interact. The paper briefly outlines these and then describes an experiment conducted to evaluate the time required by the neural network to adapt to changes in user preferences. This is used when the two approaches produce different results, to determine which results to use. It also provides input to help determine the frequency of execution of the learning algorithm used in the rulebased approach.