Projects
WASSA Shared Task on Multi-Label Emotion Detection
- Developed a multi-label emotion classification model, achieving high performance in classifying SMS messages into one or more
of 11 categories leading to 10% gain on F-score through extensive model evaluation and hyperparameter tuning
- Designed and implemented a custom deep learning pipeline using Pytorch, including data preprocessing, tokenization, and model
fine-tuning for code-mixed text (Romanized Urdu + English)
Recommender System for E-Commerce Independent Study supervised by Prof. Christoph Eick
- Designed and implemented an ontology-based recommender system for an e-commerce website, leveraging Graph Neural Networks (GNNs) to enhance the accuracy of product recommendations
- Constructed a comprehensive ontology that captured domain-specific knowledge and relationships between products, enabling an understanding of user preferences and item characteristics
- Employed GNNs to model and learn complex interactions within the product ontology, achieving a accuracy rate of 72% in generating personalized recommendations for users, thereby improving customer engagement
Modeling Patient Pathways
- Performed data science and unsupervised learning on a dataset of 65,000 patients’ electronic health insurance claims and structured electronic medical records to deliver insights for optimization of ovarian and prostate cancer treatment processes
- Engineered a framework using network analysis tools such as DAGs[DP1] (Directed Acyclic Graphs), predictive link analysis, and community detection to model patient pathways [DP2] across healthcare providers to visualize treatment narratives
- Designed and optimized data pipelines and workflows for data collection, preprocessing, and analysis for over 1 million data points
Page Design