Analyzing Bangladeshi Political News Sentiment with Machine Learning

Presentation for Research Project Titled "Analyzing Bangladeshi Political News Sentiment with Machine Learning"

In this research endeavor, we extensively explore sentiment analysis within the realm of Bangladeshi political news, utilizing a diverse dataset sourced from prominent English news outlets. The dataset, meticulously categorized into positive, negative, and neutral sentiments, undergoes rigorous training across various machine learning algorithms, such as Support Vector Machines (SVM), Naive Bayes, Random Forest, and Logistic Regression. Manual annotation via the Doccano Tool enriches the dataset with nuanced sentiment variations, fostering a profound understanding of the intricate political landscape.

1. Research Final Presentation Fall 2023
Analysing Sentiment of
Bangladeshi Political News with
Machine Learning
Decoding Bangladeshi Political Sentiment in National Daily
Newspaper in a Collaborative Thesis Project
1

2. Contributors
2
Supervised By:
Mr. Rifat Ahommed
Lecturer, Dept of Computer Science and
Engineering
Southeast University,
252, Tejgaon Industrial Area, Dhaka-1208
Research Students:
Anik Kumar Chakrabortty
2017000000037
Md. Nishad Khan
20170000000150
Shafi Md. Rawfur Raad
2017000000242

3. “By far, the greatest danger of Artificial
Intelligence is that people conclude
too early that they understand it.”
― Eliezer Yudkowsky
3

4. Introduction
Brief overview of the study
4
•This study focuses on analysing a set of news
articles from Bangladeshi English News outlet.
•We categorised the data set with certain labels
such as positive, negative and neutral.
•This data set has been trained on several
Machine Learning algorithms to predict
sentiment of news articles.

5. Objectives
Our goals and targets for the study
•Accurate Sentiment analysis of Political news
article with different Machine Learning algorithms.
•Utilise manual annotation to enhance the model's
understanding of nuanced sentiment variations.
•Curate a diverse and representative dataset of
Bangladeshi political news articles, ensuring
inclusivity of various sources and perspectives.
•Implement insights from sentiment analysis to
build an efficient recommendation system for
Bangladeshi political news.
5

6. Literature Review
Related thesis work on this domain
6
# Research Topic Author Outcome
1
Sentiment analysis of political communication:
combining a dictionary approach with crowdcoding
Martin Haselmayer, Marcelo
Jenny
The results show that the crowdbased
dictionary provides efficient and valid
measurement of sentiment.
2 Political sentiment analysis of press freedomKrzysztof Rybiński
Impact of Sentiment analysis on press
freedom ranking
3
POLITICAL SENTIMENT ANALYSIS ON DELHI
USING MACHINE LEARNING
Dharminder Yadav,
Sharik Ahmad
This paper tried to show the opinion of Twitter
user’s tweets in English regarding Delhi Election
4
Sentiment Analysis of Political Tweets for Israel
Using Machine Learning
Amisha Gangwar, Tanvi
Mehta
A comparative analysis is done
based on experimental results
from different models
5 Sentiment Analysis of Turkish Political NewsMesut Kaya; Güven Fidan; Ismail H. Toroslu
Using different classifiers, all the approaches
reached accuracies of 65% to 77%.
6Understanding bag-of-words model: a statistical framework Y. Zhang
Basics of Bag of Words approach for
ML Models creation.
7An annotated corpus for sentiment analysis in political newsG.D. de Arruda
Manual curated data from Brazilian
newspaper sources for analysis

7. Dataset Analysis
Curation of Training data for the study
•Collected 3000 English Paragraph from
Bangladeshi Online English Newspaper
Politics Section
•Manually labeled the dataset with sentiment
flag and organised them in required format
using Doccano Tool.
•Additional dataset of common English
sentences to increase overall accuracy
7

8. Methodology
Working principles for the study
•Data Collection
•Manual Labelling
•Machine Learning Analysis
•Evaluation Metrics
8

9. Methodology
Supervised Learning Algorithms
•SVM
•Rain Forest
•Naive Bayes
•Logistic Regression.
9
We have used the following Machine Learning algorithms for analysis

10. Result and Discussion
Model Training
10
•Bag of Words Approach has been
used.
•SciKit Python has been used for
model building.
Algorithmic Steps for Each Model:

11. Result and Discussion
Dataset against the SLAs
11
We have trained the model using four SLAs
and calculated accuracy for both train and test
set:

12. Result and Discussion
Performance Analysis
12
•Naive Bayes algorithm exhibited the
quickest training time, showcasing
efficiency in model building.
•However, the SVM algorithm, despite
requiring the longest training time,
achieved the highest test accuracy

13. Outcome
What is the use of these analysis
•The insights gained from the
analysis can guide decisions
regarding the scalability of the
recommendation system.
•Regular updates to
recommendation models are
common to adapt to changing user
preferences, and understanding the
training time and resource required
helps in planning these
maintenance activities.
13

14. Limitation
•Data Bias and Representativeness
•Context dependency and Cultural Differences
•Dataset Size
•Ethical Considerations
14

15. Future Work
•Increase the Dataset size for more
accuracy.
•In-depth analysis of the accuracy
calculation for the applied methods.
•Building a recommendation system for
online newspaper for increased
engagement.
15

16. References
•Gangwar, A., & Mehta, T. (2023, January 1). Sentiment Analysis of Political Tweets for Israel Using Machine Learning. Springer
Proceedings in Mathematics & Statistics. https://doi.org/10.1007/978-3-031-15175-0_15
•Rybiński. (n.d.). Political sentiment analysis of press freedom. bibliotekanauki.pl.
•Yadav, D., Sharma, A., Ahmad, S., & Chandra, U. (2020, June 21). POLITICAL SENTIMENT ANALYSIS ON DELHI USING MACHINE
LEARNING. Advances in Mathematics. https://doi.org/10.37418/amsj.9.3.50
•Haselmayer, M., & Jenny, M. (2016, September 21). Sentiment analysis of political communication: combining a dictionary approach
with crowdcoding. Quality & Quantity. https://doi.org/10.1007/s11135-016-0412-4
•Sentiment Analysis of Turkish Political News. (2012, December 1). IEEE Conference Publication | IEEE Xplore. https://
ieeexplore.ieee.org/document/6511881
16

17. Thank you for your kind attention.
We invite your questions and insights.
17