Sentiment Analysis Application
My MSc dissertation was a team project that developed a new approach to sentiment analysis. Sentiment analysis typically involves processing a large number of words, and then giving an overall rating on whether they had 'positive' or 'negative' meaning. The drawback here is that computers are easily confused by things like sarcasm or double-negatives, which means a loss in accuracy.
Our approach was new: instead of evaluating words alone, we also took the use of emoji into account, and found we got more accurate results. Emoji make up a huge part of online communication, and address the issue of clarification where a sentence might otherwise be ambiguous.
Using Twitter’s open API to source our data, we applied our method to the content of Twitter users’ tweets in order to assess the global opinion on a given topic, which were then visualised on a live-map.
The application is named SAVI: Sentiment Analysis Visualisation Interface.
We used colour to represent the positive/negative sentiment scale: red for negative, blue for positive.This method played a big role in our overall design language. Our landing page, for example, featured a CSS-animated colour shifting background.
The live map displayed the tweet results of 16 cities across the world, with their representative circles growing and changing colour in response to the scale of their emotional response.
Hovering on a city displays more information on the number of tweets processed, the local time, sentiment value score, and a stream of recent tweets from that city.
As further proof-of-concept we carried out 2 case studies as examples of different applications of the tool. As an example of global reaction to events in popular culture, we looked at viewer reaction to character deaths in the TV series Game of Thrones, and as a study of positivity vs negativity in political campaigns, we looked at Ireland’s 2015 same-sex marriage referendum.
