Named Entity Sentiment Classifications using Peripheral Words and Dependencies in Online Discussions

The sentiment classification method of posts and sentences have been proposed in several online discussion forums. However, large-town online meetings require new methods to determine the sentiment polarity of each keyword because several topics are discussed simultaneously. We propose a sentimental classification method for each named entity in various online discussion forums. We employ machine learning for the web discussion corpus and sentiment lexicon that we have developed. We define three features that focus on the peripheral words of the named entities and on the modification structures. Our experimental results exhibit that the features of the peripheral and modification structure improve the f1-score as compared with the baseline of the f1-score.