Dimentional sentiment analysis by synsets and sense definitions

Abstract

This paper proposed several approaches to predict sentiment feature values by information provided in semantic resources. The rating of a sentiment feature of a target word was predicted by weighted averaging of similar words in WordNet or Cilin whose sets of senses greatly overlapped with the sense set of the target. A vector space was also built where the dimensions were sememes in eHownet. The sentiment rating of a target word was predicted by weighted averaging of the ratings of its k nearest neighbors in the space. The performance of our system ranked in the middle position among the participants.