Abstract

One of the main challenges of recent years is to create Affective-aware Human Computer Interaction (HCI)
systems and context-aware Affective Computing (AC) systems. But, what does it mean to create or advance
such systems when incorporating context features and which should be the most appropriate type of such
context features? Even though a number of studies have analyzed how different features, when incorporated
into AC systems and particular into Sentiment Analysis (SA) systems, improve their performance; a complete
picture of their effectiveness remains unexplored. So far, a wide range of context-aware features has been
independently tested by a large number of research teams, mostly in constrained settings (Beineke et al. 2004,
Pang et al. 2004, Pang et al. 2002, Turney 2002). Nevertheless, there is not a clear picture of the impact of
every feature set and there is little to no evidence regarding how the combination of such context-aware
features behaves with different in size and genre of information sources. In light of these observations, we
attempt to extend the pool of the context-aware sentence features used into context-aware SA and to further
provide the foundations for a comprehensive analysis of the relative importance of the various types of
contextual features.