Aspect Based Opinion Mining In Social Media

Abstract

Social media provides a big pool of information and this information can be useful for many purposes. It can be seen that people post a lot of updates on the social media regarding their likes and dislikes, which adds the element of their personalization and interests. Twitter is one of the popular social media sites. Fetching relevant information and opinions from twitter is quite challenging as tweets contain short content, i.e., 140 characters and there is a lot of noise in twitter data. Existing literature addresses this issue by calculating the sentiments of tweets and giving the overall opinion to a topic. In the existing work calculating sentiments at aspect level using the twitter data have not been found. Aspects are the certain important features of the topic. Mining these aspects manually is not really feasible and scalable. The proposed methodology calculates the sentiment at the aspect level of a topic to get a clear picture of the opinion. Preprocessing tweets is done first in order to bring the text in a consistent form. After preprocessing, the second step is to find opinionated tweets. In the third step aspects from the opinionated tweets are extracted. Latent Dirichlet Allocation (LDA) has been used to extract the aspects from a dataset. The sentiments on the aspects are computed in the fourth step by using two different Application Programming Interface (API's). The final step is to generate the reports based upon sentiments on the extracted aspects. These reports give the clear idea about the sentiments of people on extracted aspects. In this research Hurricane Maria 2017 data is used. Experimental results show that overall precision (94%) of the proposed system is really high i.e., system can classify the sentiments with good precision. It will help the humanitarian organizations to deal with the situation in a much better way.