Sentiment analysis of social media content using N-Gram graphs

Sentiment Analysis over Social Media facilitates the extraction of useful conclusions about the average public opinion on a variety of topics, but poses serious technical challenges. This is because of the sparse, noisy, multilingual content that is posted on-line by Social Media users. In this paper, we introduce a novel method for capturing textual patterns that inherently supports this challenging type of content. In essence, it creates a graph whose nodes correspond to the character n-grams of a document, while its weighted edges denote the average distance between them. Multiple documents of the same polarity can be aggregated into a polarity class graph, which can be compared with individual documents in order to identify the category of their sentiment. To evaluate our approach, we conducted large scale experiments on a real-world data set stemming from a snapshot of Twitter activity. The outcomes of our evaluation indicate significant improvements over other the methods typically used in this context, not only with respect to effectiveness, but also to efficiency.