Multiclass classification on tweets about the coronavirus

Positive/negative sentiment model on cleaned text data using Distilbert NLP pre-trained model from Hugging Face

Successfully developed a fine-tuned DistilBERT transformer model which can accurately predict the overall sentiment of a piece of financial news up to an accuracy of nearly 81.5%.

Fine tune bert on a question answering dataset that is further finetuned on finance data to answer questions posed by senior leadership

The official repository for the PSYCHIC model

This paper describes Humor Analysis using Ensembles of Simple Transformers, the winning submission at the Humor Analysis based on Human Annotation (HAHA) task at IberLEF 2021.

Deep learning for Natural Language Processing

This project centers on elevating customer satisfaction by conducting sentiment analysis on customer feedback for an online classes and video conferencing app. The aim is to decipher customer sentiments in their feedback, extract insights, and improve user experience while addressing any concerns.

Thesis Project

The data and code for my master's thesis for the MA Digital Text Analysis at the University of Antwerp

Classification, ADSA and Text Summarisation based project for BridgeI2I Task at Inter IIT 2021 Competition. Silver Medalists.

This repository contains a DistilBERT model fine-tuned using the Hugging Face Transformers library on the IMDb movie review dataset. The model is trained for sentiment analysis, enabling the determination of sentiment polarity (positive or negative) within text reviews.

This is a production ready DistilBERT Sentiment Analysis model for service reviews designed to work as a low cost market research tool with the nuiance of an actual market researcher.

Using Bert and Distilbert to fill in the gaps

Finetuning the Bert-based LLM to predict whether the tweet is toxic or not

Using BERT models to perform sentiment analysis on women's clothing

Analyzes emotions in text chunks per chapter using a sentiment analysis model, visualizing scores across chunks as line graphs. Includes pie charts showing dominant emotions per chapter, enhancing understanding of emotional variations in text chunks. Developed using Transformers library.

This project involves analyzing and classifying the BoolQ dataset from the SuperGLUE benchmark. We implemented various classifiers and techniques, including rules-based logic, BERT, RNN, and GPT-3/4 data augmentation, achieving performance improvements.