Abstract

Recommender systems are software tools and techniques which aim at suggesting new items that may be of interest to a user. This dissertation is focused on four problems in recommender systems domain. The first one is context-awareness, i.e. how to obtain relevant contextual information, how to model user preferences in a context and use them to make predictions. The second one is multi-domain recommendation, which aim at suggesting items from many domains using user ratings from all of them. The third one is new user cold-start problem which occurs when a new user registers into a recommender system. He will not receive interesting recommendations just because the system does not know his preferences yet. The last problem is the need of generating explanations on how recommender system works. Two contextual user models and appropriate recommendation algorithms were proposed. First model is based on existing SIM method. The ontology-based contextual pre-filtering technique associated with the model allows dynamic generalization of contextual parameters values and generation of multi-domain recommendations. The second model consists of contextual conditional preferences. The post-filtering method associated with it, which is called re-rankCCP, allows to generate explanations for users and contextual recommendations in new user cold-start situations as well as in typical scenarios.

Author (1)