Evaluation of context awareness algorithms for recommending mobile content
Dong, Hong Liang
Date of Issue2014
Wee Kim Wee School of Communication and Information
Context-aware recommender systems (CARSs) gradually play a crucial role in modern information systems. Previous studies showed that they had great influence to users’ behaviors of information retrieval and decision making process. With the fast growth of mobile systems (e.g., smart phones), context-aware recommendation on mobile content became a hot trend in the academic studies and industrial applications. These systems are also known as mobile context-aware recommender systems (MCARSs). The various approaches in context-aware recommender systems were categorized into three approaches: pre-filtering, post-filtering, and contextual modelling. There are numbers of algorithms in each approach. These approaches and algorithms extended their usage from web based to mobile context-aware recommender systems. However, most of previous studies focused on the web based context-aware recommender systems, and there were few evaluations on the context-awareness recommendation algorithms for mobile content. The richness and dynamic of contextual information on mobile devices makes mobile context-aware recommender systems specific to web based ones. This study purposes to fill the gap of evaluating context-awareness algorithms for recommending mobile content. This dissertation aims to address a problem: which context-awareness recommendation algorithm is appropriate for the mobile context-aware applications, between the two targeted algorithms: generalized pre-filtering and weight post-filtering. The objectives of this study are to summarize the ways of systematic evaluation on context-awareness algorithms for mobile applications, and to provide advices on algorithm utilization for mobile content recommendations. In this study, I reviewed the definitions and concepts related to mobile context-aware recommendation: context, context-awareness, mobile context-awareness, recommender systems, context-aware recommender systems, mobile context-aware recommender systems, and approaches used in context-aware recommendation processes. And I also discussed the methodologies and guidelines of evaluating mobile context-aware recommender systems. I summarized the three types of experiments and the three measures of prediction accuracy for recommender systems. After modelling contextual information of the dataset, an offline experiment was implemented for the two selected algorithms. The experiment results indicated that the location-based generalized pre-filtering algorithm has higher accuracy than weight post-filtering algorithm; and for generalized pre-filtering, accuracy of usage increases when the range of generalized location increases reasonably. The findings and the way of systematic evaluation on mobile context-aware recommender system would benefit the selection process of context-awareness recommendation algorithms for mobile content.
DRNTU::Engineering::Computer science and engineering::Information systems::Information storage and retrieval
Nanyang Technological University