Sentiment Analysis of Bangla News Comments Using Machine Learning: Project Overview

• Developed a machine learning model that can classify the sentimental category (positive, negative and neutral) of a news comment written in Bangla Text.
• For the implementation a publicly available dataset of 12k news comments have been used.
• To create the system TF-idf feature extraction technique with n-gram features have been used.
• Analysed the performance of different machine learning algorithms for n-gram feature by using various evaluation metrics such as accuracy, precision, recall and f1-score.

Dataset Distribution

The dataset consists of 12K news comments of five sentiment categories. For the ease of implementation converted this five categories into 3 categories.

Dataset Summary- includes total number of words and unique words in each class.

Model Evaluation

Differnet Machine learning classifers are taken to train and evaluate the system efficacy. The experiment is done for N-gram features and measuers the performance using various evaluation metrics.

Performance on Unigram feature:

Performance on Bigram feature:

Performance on Tri-gram feature:

From the above analysis, it is observed that for trigram feature Multinomial Naive Bayes shows good performance in all evaluation metrices.

Accuracy and F1-score Plot:

References:

1. Dataset Link

Resources Used

• Python Version: 3.7
• Packages: Scikit Learn, Numpy, Pandas, Matplotlib, Seaborn