Graduate Research Assistant at UT Austin
NLP Research (Computational Linguistics and CS)
Aug 2022 - Present
Currently advised by Jessy Li, I am working on two major projects. One of them involves improving the explainability of models that are used for emotion detection in the context to disaster-related tweets The other one aspires to compare NLP models against political science baselines while measuring affective and ideological polarization.
NLP Research Intern
Jan 2022 - July 2022
Under the supervision of Dr. Scott Hale, I developed a system that can identify questions about misinformation with respect to COVID-19 on social media. I also helped perform a large-scale analysis of the geo-demographic variation of these questions using neural embeddings and K-Means Clustering.
MITACS Global Research Intern 2021
A data driven analysis of user anxiety in social media:
June 2021 - Present
Under the supervision of Dr. Morteza Zihayat and Dr. Rhonda McEwen, I aim to provide an answer to the question, how can machine learning research be employed to provide useful and actionable information for mental health professionals in the identification and monitoring of people suffering from social anxiety? This project seeks to determine the extent to which data patterns that emerge within social media platforms are causally related to information seeking, use, and sharing for those dealing with anxiety. One specific objective of this research is to develop a framework that can reveal behavioral patterns within and among social media to model social anxiety.
Research Intern at the Centre of Artificial and Machine Intelligence
EMMA-HD: An explainable ensemble model to distinguish threats from hate-speech on Twitter
December 2020 - Present
As the lead researcher on this project, I have developed a dataset of threat tweets and proposed an ensemble model of machine learning classifiers that can distinguish threats from sexist tweets. The research question that motivated this project is, how can we leverage natural language processing to make cyberspace safer for women?
The proposed model, EMMA-HD, outperforms the existing state-of-the-art hate speech detection models with an accuracy of 97.9%.
Deployed in real-time, this product can be used to detect threatening tweets and report the offender's account for suspension. We have also proposed a mechanism to keep our dataset growing over time. The technology stack we have used includes Python, Heroku, and Firebase.
This paper is currently under review.
Machine Learning Intern at MyWays
May 2020 - July 2020
Using a combination of NLP, text analytics and ML, I developed a fully functional resume parser capable of automatically extracting relevant details from resumes made using MS Word, Google Docs or Overleaf/
This resume parser was built using a python libraries like Stanford NLP's Stanza, Spacy, NLTK and Wordnet with an estimated accuracy of 80%.