Opinion Mining from Bangla and Phonetic Bangla Reviews Using Vectorization Methods

• Published in: 2019 4th International Conference on Electrical Information and Communication Technology (EICT) pp. 1 - 6
• Main Authors: Haque, Fabliha, Manik, Md. Motaleb Hossen, Hashem, M.M.A.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2019
• Subjects: CountVectorizer; HashingVectorizer; Machine Learning; Opinion Mining; Phonetic Bangla; Restaurant Reviews; TF-IDF
• DOI: 10.1109/EICT48899.2019.9068834

Opinion mining is the computational study of people's opinions, emotions and attitudes which is one of the key research field in Natural Language Processing (NLP). To cope with the competitive world, owners of business need to extract exact opinion of people about his/her business. Recently, people in Bangladesh are more interested to express their opinion in Bangla and most importantly in Phonetic Bangla rather than English. Since no specific work of Opinion mining introduced this criteria, in this paper, we have developed review analysis system on Bangla and Phonetic Bangla where we have used Restaurant reviews as case study and the dataset is created manually by us without using translator. Our approach starts by preprocessing raw data and then feature extraction with different N-gram techniques. Then vectorization is applied on that data with HashingVectorizer, CountVectorizer and TF-IDF vectorizer. Later machine learning based approaches namely Support Vector Machine (SVM), Decision Tree (DT) and Logistic Regression (LR) are applied to classify reviews. We have classified the reviews in three different classes, i.e. bad, good and excellent. Finally a comparison is shown between vectorizers in accordance with different classifiers where SVM provides better accuracy with 75.58%.