Learning Sentiments using Co-occurrence Analysis

As part of natural language processing, sentiment analysis intends to investigate the author's emotions while writing a given text. This paper proposes a new method that uses a co-occurrence graph that can be automatically extended in a background reading (and learning) process, when a person's additional text source is available. The proposed method applies the concepts of PageRank to find the sentiment value of a considered node depending on the sentiment value of predecessor nodes that point to it and iteratively process those values in the same manner. The experiment results showed that 1) the methods described converge and 2) from a few, initially labelled positive (good) and negative (bad) words precise sentiment values for other words can be obtained, which corresponds to the authors emotions.