Semantic Propagation on Contextonyms using SentiWordNet

Sentiment analysis and affect detection algorithms are generally based onto annotated data, structured into dictionaries, ontologies or word nets. The focus, so far, has been concentrated on manual annotation of the data, and then, in some situations, a semantic valence propagation is applied. The problem with this approach is that while it is able to build new affective labels through the propagation process, the precision of the decision decreases. Our approach disambiguates through the data, by offering a strong context using a contextonym model, for the usage of a certain term with a valence.