Electronic Thesis and Dissertation Repository


Master of Engineering Science


Electrical and Computer Engineering


Dr. Miriam A. M. Capretz


Recommender systems have dramatically changed the way we consume content. Internet applications rely on these systems to help users navigate among the ever-increasing number of choices available. However, most current systems ignore that user preferences can change according to context, resulting in recommendations that do not fit user interests. Context-aware models have been proposed to address this issue, but these models have problems of their own. The ever-increasing speed at which data are generated presents a scalability challenge for single-model approaches. Moreover, the complexity of these models prevents small players from adapting and implementing contextual models that meet their needs.

This thesis addresses these issues by proposing the (CF)2 architecture, which uses local learning techniques to embed contextual awareness into collaborative filtering (CF) models. CF has been available for decades, and its methods and benefits have been extensively discussed and implemented. Moreover, the use of context as filtering criteria for local learning addresses the scalability issues caused by the use of large datasets. Therefore, the proposed architecture enables the creation of contextual recommendations using several models instead of one, with each model representing a context. In addition, the architecture is implemented and evaluated in two case studies. Results show that contextual models trained with a small fraction of the data resulted in similar or better accuracy compared to CF models trained with the total dataset. Moreover, experiments indicate that local learning using contextual information outperforms random selection in accuracy and in training time.