DBpedia is a collaborative initiative aimed at extracting structured information from Wikipedia and presenting it as Linked Open Data. While DBpedia has dedicated chapters for resource-rich languages like English and German, there is a noticeable gap in representation for low-resourced languages such as Hindi.

This project continues that work for the DBpedia Hindi Chapter, building a neural pipeline that extracts structured facts directly from Hindi Wikipedia sentences and links them to DBpedia’s ontology. The system combines a fine-tuned language model for relation extraction, a predicate-linking model for matching extracted facts to DBpedia properties, and a human-in-the-loop review interface to ensure the final knowledge graph is trustworthy.

Contributor: Nitin Singh

Mentors: Sanju Tiwari, Debarghya Datta, Aditya Venkatesh, Ronak Panchal

Tech Stack

Extraction & Fine-Tuning

• Python
• PyTorch
• Transformers
• QLoRA / PEFT
• Gemma 3
• F2LLM

Human Review Interface

• Streamlit
• GitHub API

Ontology Alignment

• Sentence embeddings
• DBpedia ontology (2,891 properties)

Topics

• Semantic Web
• Knowledge Graphs
• Low-Resource Language
• Natural Language Processing

---

Project on GitHub · Email