Sentiment Analysis on Bangla and Phonetic Bangla Reviews: A Product Rating Procedure using NLP and Machine Learning

Online shopping has become very popular all over the world, although in Bangladesh it is at its early stages but people of Bangladesh are becoming familiar with this. Day by day online market is becoming more challenging due to the open platform, where product quality and customer satisfaction are the most important factors for building a reliable stand in this very challenging and competitive market. In Bangladesh an extensive number of populations are not aware of online shopping, so in many cases they aren't able to identify the reliable online platform for shopping. In recent times many online companies are making frauds with the people of Bangladesh “Evaly“ ”Alesha Mart“ are two of them. In this research we employed a customer review system, which will be a probable solution for overthrowing the problem. Our aim for this research is to monitor reviews given by the customer after buying some product from an online market, where we can easily find out what are the companies or shops are making frauds with the people. So, we created our own “PR” (Product Review) dataset containing 4,000 reviews both in Bangla and Phonetic Bangla from various e-commerce websites, online store pages, social media platforms, and YouTube videos. We manually labeled the collected data into “positive” and “negative”. Using TF-IDF we extract features from the data and train our model using supervised machine learning algorithms. The results showed that SVM had the highest accuracy in both the Bangla and Phonetic Bangla datasets, with 82% and 94% respectively.