Debayan Banerjee

I am employed as 'Akademischer Rat' at the Leuphana University at Lüneburg, Germany, where I work in the group of Prof. Dr. Ricardo Usbeck. Akademischer Rat is a German term which losely translates to 'Academic Advisor'. My tasks are to conduct teaching and research.

I completed my doctorate from the Language Technology Group of the University of Hamburg, Germany. I was co-supervised by Prof. Dr. Chris Biemann and Prof. Dr. Ricardo Usbeck. My area of interest lies broadly in the field of Natural Language Processing, and currently, I focus on Question Answering over Knowledge Graphs.

I completed my Bachelors in Information Technology in 2009 at the National Institute of Technology, Durgapur, India. I spent the next 7 years in the software industry, of which 4 years I spent as a co-founder of Gazemetrix Inc. After Gazemetrix, I worked at paytm.com in the DevOps team.

In 2017 I took a break from the industry and resumed my academic journey. I started my M.Sc. in Computer Science at the University of Bonn, Germany where I worked under Prof. Dr. Jens Lehmann on Knowledge Graph Question Answering.

Google Scholar  /  Twitter  /  Github

We evaluate SPARQL-MCP-powered intelligent agents on a federated Knowledge Graph Question Answering benchmark.

Book Chapter from "Frontiers in Artificial Intelligence and Applications", "Volume 400: Handbook on Neurosymbolic AI and Knowledge Graphs"

We investigate whether state-of-the-art Large Language Models can automatically translate SPARQL queries between different Knowledge Graph schemas.

A question answering system for the DBLP scholarly knowledge graph.

A new KGQA dataset derived from SPARQL query logs over the DBLP scholarly knowledge graph, providing a more comprehensive benchmark for evaluating KGQA systems.

We present best practices for evaluating AI and data science models.

We describe the results of the KGQA shared task co-located with the TextGraphs 2024 workshop, where participating systems answer questions using the interaction between large language models and knowledge graphs.

A hybrid scholarly question answering dataset combining factual and analytical questions over scholarly knowledge graphs.

We report and analyse the environmental impact of language models on the example of commonsense question answering with external knowledge.

An analysis of large language models for semantic parsing in the context of knowledge graph question answering.

A web application that performs entity linking over the DBLP KG.

We show that certain vocabularies are better than others for the task of semantic parsing.

We show that T5 can generate logical forms and also learn a simple embedding space for entities.

A dataset consisting of 10,000 question/answer pairs and corresponding SPARQL query over the DBLP scholarly knowledge graph.

We present a system that enables human-AI collaboration for online customer support.

We present a web interface for exploring the web of scholarly data.

We evaluate the performance of contemporary Text-2-Text pre-trained language models against a Pointer Generator Network in the task of SPARQL Semantic Parsing.

Hybrid Intelligence Systems help overcome current pitfalls in Customer Support Services by combining the complementary strengths of artificial and human intelligence.

A Pointer Network is employed in the task of Entity Linking.

A large question answering dataset over Wikidata and DBpedia Knowledge Graphs is developed and shared with the community.

Joint Entity and Relation Linking problem is framed as a Travelling Salesman Problem and solved approximately.

Images are segmented based on their text captions.

Website Template taken from John Barron.