Question-Answering over DBpedia with Pretrained Auto-regressive Models
- Mentors
- Tommaso Soru, Anand Panchbhai, Saurav Yogen Joshi, Dr. Sanju Tiwari
- Organization
- DBpedia
- Technologies
- python, tensorflow
- Topics
- knowledge graphs, Transformers, Large Language Models, Question-Answering
This proposal aims to contribute to addressing the challenge of generating formal SPARQL queries from natural language questions, thereby facilitating intuitive question-answering from open knowledge graphs, such as DBpedia. Among the three major approaches to this task, i.e., classification, ranking, and translation [Chakraborty et.al, 2019], this proposal will primarily focus on the latter one (translation). The key motivation behind choosing the translation approach is the significant performance improvement observed with the use of pre-trained language models such as T5 or GPT models in translation tasks.