Torii: Attribute-based Polarity Analysis with Big Datasets

Polarity analysis has become a key aspect of market analysis. The number of companies that are interested in the general opinion of the crowd regarding the items that they sell is increasing everyday. Attribute-based polarity analysis is a fine-grained approach that computes if the opinion about an attribute of (a component of) an item is positive, negative, or neutral. The existing techniques have a number of problems, namely: they do not take into account the conditions expressed in the opinions (e.g., when they hold and when they do not), they do not generally use any contextual information (e.g., past user opinions on the same attribute), and they are not validated on big datasets (e.g., billions of messages). In this paper, we present Torii, which is an attribute-based polarity analysis technique that takes both conditions and contextual information into account; we also present our approach to validate it on big datasets.