Triangulated Sentiment Analysis of Tweets for Social CRM

High resolution data from social media platforms like Twitter presents an unprecedented opportunity to organisations for social customer relationship management (Social CRM) by analysing the ongoing discussion about business events such as a service outage. Text based sentiment analysis has been widely researched utilising mainly lexicon-based and machine learning approaches to uncover customers' opinions. They are similar in the sense that the machine learning approach relies on an initial lexical model on which the learning is based. Both methods view sentiment as either positive, neutral, or negative. This is not the case for the psycholinguistic approach following which text sentiment is more continuous. We compare these three approaches with a Twitter dataset collected during a service outage. Contrary to our expectation, we find that the language used in tweets is not very negative or emotionally intense. This research therefore contributes to the sentiment analysis discussion by dissecting three methods and illustrating how and why they arrive at differing results. The selected research context provides an illuminating case about service failure and recovery.