Characterisation of User-Contexts for Context-Aware Recommendation Systems

Abstract

AbstractThere is a growing interest in context-aware recommendation systems (CARS) due to their ability to enable novelty, dynamism and serendipity in resulting recommendations. The incorporation of different user-contextual parameters is important in CARS. However, the problem of identifying the most optimal subset of user-contextual parameters that add real value to resulting recommendations from CARS remain unresolved. This paper investigates an approach to characterise user-contextual parameters and identify the optimal subset of parameters that enables best results from CARS algorithms. In this paper, two data sets were collected from different people in Africa, each data set had 21 classified user-contexts and corresponding likes. An association rule mining (apriori) algorithm was used to generate rules from the user profiles (user contextual-parameters) from both data sets, to identify user-contextual parameters which are strongly associated with certain user preferences. Five(5)[Hobbies, Mood, Family setup, Religion and Gender] user-contextual parameters out of the 21 were identified as the optimal subset of parameters for the domain and the data sets in this study. This optimal subset was found promising in helping CARS as the parameters which can estimate best the preferences of the users and hence led to best recommendations. This paper contributes a new method of identifying the optimal subset of user-contextual parameters in any domain, which enables CARS lgorithms to produce the most appropriate recommendations for each user.KeywordsContexts (user-contextual parameters)Recommender systemsAprioriSocialCultural orientationPsychological profileEconomic statusUser profile