Customer satisfaction monitoring with sentiment analysis based on twitter feeds in telecom domain

Abstract

With this increased competition among telecom service providers, it has become more difficult to retain the existing customers, but when the number of customers reaches its peak, finding and securing new customers become increasingly difficult and costly. Therefore, it would be better to prioritize the retention of the existing customers, than trying to win new ones. Customer reviews can be recognized as fruitful information sources for monitoring and enhancing customer satisfaction levels as they convey the real voices of actual customers expressing relatively unambiguous opinions. This research is aimed at mining and measure customer satisfaction toward Telecom Service based on reviews and feedbacks from Twitter. This research is mainly focus on one of the largest mobile operator in Sri Lanka and the analysis has been done only for English language. Tweets were classified into three classes as Positive, Negative and Neutral with the use of four dictionaries (Lexicon, SentiWordNet, Slangs& Emoticons). The framework was built based on six steps and it shows that Lexicon performs well on the dataset better than SentiWordNet. After fine-tuning lexicon and stop words dictionary and integrating with Slangs dictionary, positive classification shows 91.98% accuracy without Emoticon dictionary while for negative classification, the accuracy is 82.27% with Emoticons dictionary