Bio-SODA UX: enabling natural language question answering over knowledge graphs with user disambiguation.

Area of Science:

• Computer Science
• Bioinformatics
• Data Science

Background:

• Natural Language Processing (NLP) over structured data is a growing research area.
• Existing Knowledge Graph Question Answering (KGQA) systems often require large training datasets or are domain-specific.
• These limitations hinder their application to complex scientific datasets.

Purpose of the Study:

• To introduce Bio-SODA, an NLP engine for scientific knowledge graphs that does not require training data.
• To enable efficient querying of complex scientific data using natural language.
• To improve the performance of KGQA systems in scientific domains.

Main Methods:

• A generic graph-based approach for translating natural language questions into SPARQL queries.
• A novel ranking algorithm incorporating node centrality for query relevance.
• Implementation of Bio-SODA UX, a graphical user interface for knowledge graph exploration and question disambiguation.

Main Results:

• Bio-SODA successfully generates SPARQL queries without question-answer pair training data.
• Experiments on scientific datasets, including bioinformatics and CORDIS, show significant performance improvements.
• Outperformed publicly available KGQA systems by at least 20% F1-score, with higher gains on complex bioinformatics data.

Conclusions:

• Bio-SODA offers a data-efficient and effective solution for NLP over scientific knowledge graphs.
• The system demonstrates superior performance compared to existing KGQA approaches.
• Bio-SODA and its accompanying UX facilitate broader access and exploration of scientific data.