Soft Constraint Query Answering on Knowledge Graphs

Key facts

• Methods for query answering over incomplete knowledge graphs retrieve likely answers when direct graph traversal fails due to missing edges.
• Existing approaches focus on queries formalized using first-order-logic.
• Real-world queries often involve vague or context-dependent constraints such as preferences for attributes or related categories.
• The problem of query answering with soft constraints is introduced and formalized.
• Two efficient methods adjust query answer scores by incorporating soft constraints without disrupting original answers.
• The methods are lightweight, requiring tuning only two parameters or a small neural network.
• The neural network is trained to capture soft constraints while maintaining the original ranking structure.
• The paper is available on arXiv with ID 2508.13663.