|Thesis abstract: |
Massive amount of multimedia content is b eing generated at an unprecedented rate everyday so much so that the users to day need more than a traditional ¿search¿ for the multimedia information but rather a ¿discovery¿ of the multimedia content. Alb eit robust in retrieving textual information, to day¿s search engines like Go ogle have had limited capabilities in
retrieving information that are based on multimedia. We aim to investigate the impact
of Multimedia (MM) content on Recommender Systems (RS) along two directions: (1)
From content point-of-view (2) From user p erspective. As for the content side, we aim to study a variety of feature descriptors that can help us in bridging the gap b etween low-level information in a video and human high-level understanding of its content. This requires studying dozens of low-level and high-level features that can represent and describe the structure and content of a video from multiple of persp ectives. As for the user side, we focus our attention on recommender technologies integrated with smart interactive systems
(e.g. smart phones, large displays). Our goal is to develop techniques able to capture
contextual information of the user (e.g. mo o d, emotions) from the real-time analysis of live feeds from video input sources (e.g. web cams). The idea here is to study the user behavior, emotion and affective resp onses in a multimo dal interactive environment. Such multimedia signals need to extract variety of information hidden in user speech signals, the user facial
features and group preferences and incorp orate this rich source of information into the RS for more precise understanding of users behavior and thus providing more meaningful recommendations.