Mediknow - A Malayalam Cancer Question Answering System

Abstract

This paper introduces "MediKnow," a
pioneering Malayalam Question Answering System
designed to address the scarcity of generative answer
works in the realm of healthcare information
accessibility, specifically tailored for cancer-related
queries. The dearth of such systems in Dravidian
languages, particularly Malayalam, has motivated the
development of a robust solution. Leveraging advanced
Natural Language Processing (NLP) techniques,
including OpenAI models and FAISS for efficient vector
storage, MediKnow employs a specialized Malayalam
language model to navigate the intricacies of the
Dravidian linguistic context. The processing pipeline
encompasses document loading, text splitting, and
embeddings, enhancing the system's capacity to
comprehend and accurately respond to a diverse range of
cancer-related questions. This work underscores the
critical need for bridging the gap in generative answer
works for Dravidian languages, highlighting the specific
challenges posed by the Malayalam language due to its
complexity. Beyond providing accessible information,
MediKnow exemplifies the efficacy of employing state-ofthe-art NLP technologies to address linguistic nuances.
The paper evaluates the system's performance on a
dataset of cancer-related questions, demonstrating its
ability to deliver accurate and informative answers. The
innovative approach presented herein contributes to the
advancement of NLP capabilities in non-English
languages, particularly focusing on healthcare-related
information retrieval. The development and deployment
of "MediKnow" signify a significant stride in tackling
linguistic and domain-specific challenges in cancerrelated question answering, ultimately making critical
healthcare information more accessible to Malayalam
speakers.