Aspect Based Sentiment Analysis of PLN Customer Complaints Data Using BERT to Improve Services

Customer service quality determines customer satisfaction with a product or service. PLN, as a state-owned company in the electricity sector with the largest number of customers, can rely on customer feedback to evaluate and follow up on problems that arise, thus preventing similar problems from arising in the future. This research analyzes customer complaint feedback using Aspect-Based Sentiment Analysis (ABSA). Weighted cross-entropy loss and stratified random sampling are used in the calculation of the BERT Transformer Algorithm which is compared with two other methods, SVM and LSTM for comparison. Experimental results on 12511 rows of data show that the results obtained by the BERT model outperform the other two methods. The results concluded that the BERT model showed a promising average F1-Score value of 0,756. The implementation of BERT in sentiment analysis based on aspects of service complaints in PLN has shown substantial capabilities. And solutions to improve PLN customer service, which is also a challenge for the development of in-depth NLP (Natural Language Processing) systems.