Sentiment analysis on mobile operators’ customer review

Abstract

Mobile is a great invention for the global village. Now, people from different corners of the world are interconnected over this device. Therefore, in Bangladesh, there are different telecommunication companies have emerged. They are competing among them for providing a better experience for their users. However, over the past few years, for the significant development of technology and software, we have got many advantages from this mobile phone using the internet. We live in a time when emotions are high because they’re publicly shared, and displayed, in social media [1]. Among all of the social network sites, Facebook is very popular where people pass their time most. Researchers said on average a person spends 38 minutes every day on Facebook [2]. Telecommunication companies have opened their Facebook pages for providing a better experience to the customers. They post different types of services, packages and offers on their pages. On these Facebook pages, customers share their suggestions, complains and reactions on every post. Our research purpose is to analyze the sentiment of customers from these comments. We are analyzing whether a comment can be positive, negative or neutral. We have scrapped and labeled around 7 thousand comments ourselves from different verified pages of top telecommunication companies in Bangladesh. We use these as our dataset. In addition to that, we labeled each comment by positive, negative and neutral. Then, we have implemented Support Vector Machine (SVM), Na¨ıve Bayes, Decision Tree and K-Nearest Neighbor (KNN) algorithms for getting our results. After implementing those algorithms, when we keep all language’s comments in one dataset, we got 73% accuracy for SVM, 57.3% accuracy for KNN, 63.61% accuracy for Decision Tree and 64.11% accuracy for Na¨ıve Bayes algorithm. In the same way, when we have separated Bangla language dataset, we got 70.65%, 65.58%, 55.07% and 56.70% accuracy respectively for SVM, Na¨ıve Bayes, Decision Tree and KNN algorithms. In addition to that, we also have separated our Banglish (Bangla words typed using English letters) dataset and got 74%, 60.29%, 71.23% and 61.30% accuracy respectively for SVM, Na¨ıve Bayes, Decision Tree and KNN algorithms. We do believe that our research has a great impact on the business of telecommunication companies. Through our research, a telecommunication company can understand customer sentiment on different new products, projects, and offers which is very important in the era of business competition. As a result, they can achieve the highest success through the highest client satisfaction.