Customer Experience and The Impact Towards Customer Perception in Indonesian Telecommunication Company Using Twitter Sentiment Analysis: PT. Smartfren Telecom, Tbk. Case Study

Abstract

onsumers of telecommunications are growing, and 97% of all users are assigned to prepaid cell plans. Moreover, social media is utilized by marketers and salespeople to connect with and reach their target clients, because it helps businesses quickly and affordably reach their target customers and understand their requirements and goals. The limitations of data deep dive analysis prevent insights from being sufficient for business questions, despite Smartfren starting to analyze social media sentiment data as of 2021. To meet the commercial needs of Smartfren, a Twitter sentiment analysis was done for this study. Sentiment analysis labeling is conducted using a distinct lexicon-based method to increase accuracy. This method divided the lexicon dictionary and labeling process based on churn-related tweets and non-churnrelated tweets. Following training and testing using the Support Vector Machine (SVM) classification algorithm, which is divided into three feature extraction techniques (Count Vectorizer, Bags of Words, and TF-IDF), the results of the sentiment labeling are then used to evaluate the accuracy of the labeling using a parameter. The Bags of Words approach produces the best classification results, with 97% accuracy (98% precision, recall, and F1 score). The results of a sentiment analysis reveal that there are more tweets with negative sentiment than positive sentiment during a period of 2.5 years (70.5% of negative tweets). Product experience and network experience are the parts of the customer experience that are most frequently discussed, accounting for (37.8% and 37.5% respectively). Most of the network-related tweets have mostly discussed coverage, social media, and gaming experiences. Positive sudden sentiment changes and customer churn dominate negative ones in terms of consumer perception change. The researcher found that the number of sentiment changes and churn are higher with a higher level of average engagement rate, and vice versa.