Classification of Social Media Users by Interests and Sentiments using Text Mining Techniques

Abstract

Social media sites like Twitter, Facebook, YouTube etc. have emerged as powerful platforms of communication
where people share all kinds of information about topics they are interested in such as their opinion on real world
events, personal experiences, product reviews and many more. The problem with this information is that it is
unorganised and unstructured, therefore, it is difficult to assess automatically and in bulk. Studies on Twitter data
have demonstrated that aggregating millions of messages can provide valuable insights into the interests of a
population and opinions about said population. This study is aimed at gaining insights from the ever-growing
Nigerian data generated from twitter by profiling the user to determine their interests and opinions. A framework
for topic extraction and opinion mining is developed. The study used datasets across 5 popular and verified users
in Nigeria to evaluate the proposed framework for its reliability and validity. Topic modelling was used to extract
the topics of interest to the user while sentiment analysis was used to detect their opinions in each of their tweets
which was further aggregated over each topic to get their total sentiscore about each topic. Topics of interests and
overall interest level were detected within the timeframe of the datasets for the users. The interest of the users was
obtained and compared among the last 6 months under observation to determine how users’ opinions and interests
changes over time. The findings, therefore shows that even though opinions and interests do change over time,
the changes are generally minimal in subsequent months.