Structural Linguistic Feature-based Sentiment Analysis On Social Media Messages

With the advent of the Web 2.0 era,the interactive and instantaneity of interpersonal communication through social networks has been strengthened.All kinds of social media have gradually become popular platforms for the public to express their feelings,view and behavior of mass text information for scientific research purposes.Sentiment analysis on social media messages becomes one of the hottest topic of study,it is an important natural language processing(NLP)task and applied to a wide range of scenarios: economics,politics,sociology,psychology etc.Considering supervised sentiment analysis method needs large annotated datasets during training stage and supervised classifier is hard to interpret and extend,in this article,we present a sentiment lexicon-based unsupervised method,particularly by extracting and utilizing linguistic features from social media messages in a comprehensive manner to improve sentiment calculation.Fist we manually created five auxiliary dictionaries: standard English word dictionary,intensifier dictionary,downtoner dictionary,negation dictionary and slang dictionary.Then we design a sentiment-aware preprocessor,including tokenizer,normalizer and POS tagger.After that we extract word-level feature,phrase-level feature and sentence-level feature in a structural manner.Finally with sentiment lexicons we can calculate sentiment score for social media messages.It's necessary to mention that the sentiment-aware data preprocessor,structural linguistic feature extraction and sentiment calculation being separate components,which allows for easy modification and extension of each component.Our system has been evaluated on five commonly used public tweet datasets.Results show that our system outperforms existing state-of-the-art lexicon-based sentiment analysis solutions.