P-005 · 2026 · Team project
Knowledge graph question answering with LLMs
A team-built system that turns natural-language questions into SPARQL queries over DBpedia and returns answers grounded in query results.
Contribution and context
I contributed to this team project for the Advanced Machine Learning course at Leuphana University. The work combines natural-language processing, knowledge-graph queries and LLM orchestration in a Streamlit application.
Method
The pipeline links entities, probes the graph schema and decomposes multi-hop questions before generating SPARQL. It executes the query against DBpedia, applies repair and retry steps, and turns the returned results into an answer.
Public outputs
The repository includes the application, architecture documentation and an LC-QuAD evaluation workflow. Evaluation compares answer sets using macro-averaged F1 and records failure categories. These materials make the system’s approach and evaluation inspectable.
Scope
Entity resolution, predicate selection and changes to the live DBpedia graph affect answer quality. The evaluation documentation explains how answerable benchmark questions are selected.