Text Classification on Social Media using Bidirectional Encoder Representations from Transformers (BERT) for Zakat Sentiment Analysis

Abstract

Social media has impacted social communities in many aspects. They provide a platform to share information, express opinions and discuss common interests. They have become important part of everyday life. Companies used them for marketing and engage with customers for their services and products. The usage of social media has great influence on company image reputation. A viral positive message is good for a company but negative viral message may affect image reputation of the company. Zakat institution manage zakat payment and distribution of zakat to the zakat receivers. There are zakat payers and receivers that did not satisfy with the current mechanism of zakat and they use social media to communicate not only on personal matters but also their satisfaction with services or products that they received. Therefore, sentiment analysis on zakat is important to resolve the customer dissatisfaction or problems. BERT model is used to analyze the sentiment analysis since it is a powerful Transformer-based machine learning model for Natural Language Processing (NLP). Data is obtained from social media based on keywords of zakat. Then a text pre-processing and word-embedding features of BERT are used to build a text classification model. This classification model is used to analyze the sentiment analysis on the zakat institution. The result from this model can be used by zakat institutions to resolve zakat payer and receiver matters in a more personalized customer service systematically and efficiently.