Opinion Mining on Emojis using Deep Learning Techniques
Valmeekam Karthik, Dheeraj S Nair, J. Anuradha · Procedia Computer Science · 2018
Opinion mining is a trending research topic right now. These techniques are used to know the status of the performance of a business product or the way in which information is spread (positive or negative). It is also being used in the domain of journalism to know the pulse of the people in a famous public action. Even with such huge applications, opinion mining has been restricted to text mining. This can lead to wrong results whenever these texts are used with other tools like emojis. On platforms such as Twitter and Facebook, there is a significant impact of emojis on the opinion that the posts or tweets convey along with the text. This paper aims to provide insight into mining opinions through emojis on tweets using different techniques such as Machine Learning Classifiers, Artificial Neural networks, and Convolutional Neural networks. Additionally, an analysis is made as to which is the best technique to get the proper polarities for the emojis. Through this analysis, the emojis and the text of a tweet are mined separately and then aggregated to generate the polarity of the opinion conveyed in the tweet. This procedure works on a datastore which contains the Unicode data of the emojis and the associated weights/polarities. The tweets contents are searched for these emojis and the emoji polarities are given. A relation between the text polarity and emoji polarity is found to detect cases of sarcasm in some tweets.