AFS Graph: Multidimensional Axiomatic Fuzzy Set Knowledge Graph for Open-Domain Question Answering.

Area of Science:

• Artificial Intelligence
• Natural Language Processing

Background:

• Open-domain question answering (QA) requires models to retrieve inference chains from extensive document collections.
• Effective QA models depend critically on robust information filtering and reasoning capabilities.

Purpose of the Study:

• To propose a semantic knowledge reasoning graph model using multidimensional axiomatic fuzzy sets (AFS) for unsupervised knowledge graph (KG) generation and reasoning path construction.
• To enhance the interpretability and reasoning abilities of QA models.

Main Methods:

• Developed a semantic knowledge reasoning graph model based on multidimensional axiomatic fuzzy sets (AFS).
• Utilized the AFS framework for unsupervised KG generation and reasoning path creation.
• Calculated paragraph similarity using AFS descriptions to construct the graph.

Main Results:

• The proposed AFS Graph model demonstrates improved interpretability and reasoning over existing methods.
• Achieved state-of-the-art performance on HotpotQA, SQuAD, and Natural Questions Open datasets.
• The AFS framework provides concise and flexible semantic descriptions for documents.

Conclusions:

• The AFS Graph model offers a more interpretable and capable approach to open-domain question answering.
• The model's ability to learn and analyze semantic relationships is a key advantage.
• This approach advances the field of reading comprehension and QA systems.