Mining contextual knowledge for context-aware recommender systems

With the rapid growth of the number of electronic transactions conducted over the Internet, recommender systems have been proposed to provide consumers with personalized product recommendations. A hybrid symbolic and quantitative approach for recommender agent systems is promising because it can improve the recommender agents' prediction effectiveness, learning autonomy, and explanatory power. However, recommender agents must be empowered with sufficient domain-specific knowledge so as to reason about specific recommendation contexts to improve their prediction accuracy. This paper illustrates a novel text mining method which is applied to automatically extract domain-specific knowledge for context-aware recommendations. According to our preliminary experiments, recommender agents empowered by the text mining mechanism outperform the agents without text mining capabilities. To our best knowledge, this is the first study of integrating text mining method into a symbolic logical framework for the development of recommender agents. © 2012 IEEE.