Prototype to analyze customer online sentiment in Kenya

Abstract

The advent of Web 2.0 has meant that an average user can be able to generate content on the Internet simply by making a post on a social media site or a blog entry in their personal blog or filling out a customer feedback form. Micro-blogging sites like Twitter make it possible for people to quickly comment on things around them or how they feel about something without having to publish a whole article on the specific issue. These opinions are important both at an individual and organization level in that they help guide the decision making process. Given that the number of comments posted about various topics is enormous, it is not feasible for human beings to sift through the constant flow of comments and make a valid assessment. This big data problem would require, opinion mining, an automated way of extracting and analyzing sentiments about various topics. This research looks into the components of opinion mining and explores how it can be applied in real life. We develop a prototype that reads through various twitter posts (tweets) and determines which statements express positive, negative or neutral sentiment based on the available training data. The Naïve Bayes and Maximum Entropy algorithms were used in this implementation. The research concludes by testing the prototype and assesses its accuracy in determining opinion orientation. This research study provides companies with a basis for a tool to quickly establish public perception about their goods and services. The prototype tool was able to obtain an accuracy of 81%. It will be useful in helping companies gain knowledge of consumer sentiment in a real-time basis.