Bio-SODA

Bio-SODA answers natural-language questions over domain-specific scientific knowledge graphs by translating them into SPARQL queries to retrieve structured scientific data.

Key Features:

• Training Data Independence: Operates without question-answer pair training data, enabling use on scientific datasets with scarce or no annotated examples.
• Graph-Based Approach: Employs a generic graph-based methodology to translate natural-language questions into SPARQL and generate a ranked list of candidate queries.
• Novel Ranking Algorithm: Ranks candidate SPARQL queries using a node centrality–based relevance measure to improve selection accuracy.
• Performance: Demonstrates at least a 20% F1-score improvement relative to publicly available knowledge graph question-answering (KGQA) systems, with pronounced gains on complex bioinformatics datasets.

Scientific Applications:

• Knowledge Graph Question Answering: Enables QA over domain-specific scientific knowledge graphs to retrieve structured biological and scientific information.
• Bioinformatics Datasets: Applied to complex bioinformatics datasets where annotated training examples are limited, improving information retrieval accuracy.
• Benchmarking and Evaluation: Used in evaluations and challenges such as the bioinformatics QALD challenge and the CORDIS dataset of European projects.

Methodology:

Translates natural-language questions into candidate SPARQL queries via a graph-based method and ranks candidates using a node centrality–based algorithm; does not require question-answer pair training data.