A social media sentiment analysis model to support marketing intelligence in Kenya

Abstract

ABSTRACT Over a decade ago, what commenced as a collection of individual musings scattered across the internet has since evolved into the de facto voice of the global public. It is called social media. The use of social media has brought about the biggest shift of how we gather and respond to information since the advent of internet itself. The sentiments expressed therein have led to undeniable influence in changing the world around. What do tweets, blogs and posts about your products and services tell you about what they think and feel? These will definitely influence your sales and other KPIs. Every business needs to be able to meticulously translate these social media data for marketing guidance. This can give competitive advantage in terms of early trend detection, alarming on emerging issues and monitoring on competitors activities among others. Posting reviews online has become an increasingly popular way for people to express opinions and sentiments toward the products bought or services received. Analyzing the large volume of online reviews available would produce useful actionable knowledge that could be of economic value in terms of marketing intelligence. This study, in the literature, investigates the different available social media platforms and sentiment analysis techniques together with approaches to combine several classifiers. The main aim of this study is building a sentiment classifier to classify twitter opinions as positive, negative or neutral. The classifier is specifically used in a business setting for marketing intelligence. This involves analyzing the business products, services, brand and presence. The information yields very useful information and insights for marketing strategies. Our review of literature indicated that support vector machine (SVM) was generally the most accurate machine learning classifier for sentiment analysis and robust on large feature spaces. The study employs a combination of SVM with Naïve Bayes algorithms using the ensemble approach to enhance the overall performance. Model validation is investigated and a prototype is constructed for output presentation. iv