Browsing by Author "Agarwal, Harsh"

Now showing 1 - 1 of 1

Unified Framework for Pointwise Explainable Information Retrieval
(Indian Statistical Institute, Kolkata, 2024-06) Agarwal, Harsh
As Machine Learning (ML) models become increasingly sophisticated and opaque, the necessity for explainability to ensure transparency and accountability in their applications grows. Despite numerous proposed methods for explaining these complex models, there remains a lack of a unified framework that encompasses these approaches for comprehensive experimentation and analysis. This thesis introduces “ir_explain”, an integrated Python module that consolidates various explainability techniques specifically for Information Retrieval (IR). While the entire module represents a collaborative effort, my focus has been on the implementation and analysis of Pointwise explanations. By consolidating these methods into a single package, ir_explain simplifies their application and facilitates robust analysis. Through a series of experiments, this thesis showcases the module’s practicality and effectiveness, contributing to the development of more transparent and interpretable ML models in the IR domain, with a primary focus on Pointwise explanations.