Food Trend.AI
Thesis & Authors
Undergraduate thesis ยท BRAC University

Sentiment Analysis of Customer Reviews on Food Ordering Portals of Bangladesh using Natural Language Processing

Foodpanda BangladeshHungryNaki6 ML models3 deep neural networks
Read the published paper
IEEE ยท ICCCMLA 2024 ยท DOI 10.1109/ICCCMLA63077.2024.10871506

Abstract

In recent years, the popularity of online food ordering services has surged, offering consumers a convenient way to order food from restaurants and have it delivered to their doorstep. During this period, HungryNaki and Foodpanda Bangladesh have been identified as key contributors to the growth and advancement of the online food delivery market. This study aims to anticipate the sentiments of Bangladeshi customers towards online meal ordering services, with a specific focus on Foodpanda Bangladesh and HungryNaki. We created a new dataset for this research, subjected it to preprocessing and employed six machine learning models and three deep neural network models. In the machine learning approach, the Random Forest Classifier demonstrated excellence in accuracy, precision, and recall, achieving an accuracy rate of 80.31%. On the other hand, the BERT Classifier performed effectively in the deep learning approach, boasting an accuracy of 89%. Despite challenges such as data ambiguity and imbalances in the dataset, our findings underscore the potential of the BERT model in sentiment analysis, offering valuable insights for future research in this field.

Model results

Best model from each approach.

BERT Classifier
89%
accuracy
Random Forest
80.31%
accuracy

Authors

  • PDPriom Deb
  • ARAsibur Rahman Bhuiyan
  • FAFarhan Ahmed
  • RHRakib Hassan
  • HMHabiba Mahrin
  • FAFaisal Ahmed
  • DZDewan Ziaul Karim

Affiliation

Dept. of Computer Science and Engineering

BRAC University, Dhaka, Bangladesh

About the data

The dataset used in our platform, Food Trend.AI, consists of customer food reviews collected from online food ordering portals in Bangladesh. These reviews were publicly accessible and viewable by both users and visitors of the websites, even without logging in. Prior to initiating the scraping process, we thoroughly reviewed the terms and conditions of these websites to ensure compliance with their policies regarding the scraping and usage of publicly viewable customer reviews. We found no restrictions or prohibitions against scraping and using this data.

The dataset has been utilized for our thesis on sentiment analysis. We have not altered any data within the dataset; all reviews are genuine. For text processing, we only removed unwanted punctuations to clean the text for our model training.

Disclaimer

Food Trend.AI is a platform developed solely to showcase our thesis. We do not support or condone any actions to promote or demote any restaurants through this platform. Our statistics are based on real data scraped from online food ordering portals. The top restaurant rankings presented are derived purely from actual customer ratings and reviews.

Our intention is not to promote or damage the reputation of any restaurants or online food ordering websites. For any clarifications or objections, please contact.