research

thesis · publications

ananya@purdue:~$ls -la ./research

total 3

thesis.md data valuation for label error detection in ML pipelines · ongoing
ccce23.md interpretable hybrid recommender · CCCE'23, Stockholm
isda22.md RePI: research-paper impact analysis · ISDA'22

ananya@purdue:~$cat thesis.md

[ m.s. thesis · in progress ]#

Data Valuation for Label Error Detection in ML PipelinesResponsible Data Science (RDS) Lab, Purdue University · Aug 2024 – PresentAdvised by Dr. Romila Pradhan

Developing Shapley-value-based data-valuation methods to detect and repair mislabeled training data in ML pipelines, with a focus on improving model fairness, reliability, and explainability.

ananya@purdue:~$cat ccce23.md

[ ccce'23 · stockholm ]#

An Interpretable Hybrid Recommender Based on Graph Convolution to Address SerendipityPublished at CCCE'23 · Stockholm, March 2023 — Springer link

Two novel contributions built on top of a 4-model hybrid graph-convolutional recommender:

A new distance-based metric for quantifying recommendation serendipity, going beyond standard diversity/novelty proxies
KNN feature-importance analysis layered on the hybrid to make its recommendations interpretable to end users

ananya@purdue:~$cat isda22.md

[ isda'22 ]#

RePI: Research Paper Impact AnalysisPublished at ISDA'22 · December 2022 — Springer link

A web application for analyzing research-paper impact, built around a novel impact-factor ratio — a new metric for publication influence that goes beyond raw citation counts.

Implemented the metric and pipeline in Python on top of the Semantic Scholar API
Built an interactive interface in Streamlit for exploring per-paper and per-author impact

ananya@purdue:~$exit