Query answering over incomplete knowledge graphs traditionally focuses on retrieving entities through strict logical queries, which can be limiting when dealing with missing edges or vague constraints. This work identifies a gap in handling soft constraints such as preferences for certain attributes or related categories, which are common in practical applications. To address this, the authors propose the Neural Query Reranker (NQR), a model that adjusts the scores of query answers by incorporating soft constraints without altering the original query results. NQR works interactively, allowing users to provide incremental examples of preferred and non-preferred entities, thereby refining the answer set dynamically. The approach is validated by extending existing query answering benchmarks with datasets that include soft constraints, demonstrating that NQR effectively captures these nuances. Experimental results show that NQR maintains robust performance in query answering while improving the relevance of answers under soft constraints. This method enhances the flexibility and applicability of knowledge graph query systems, particularly in real-world scenarios where user preferences and context play a significant role. The interactive nature of NQR also suggests potential for adaptive systems that learn from user feedback over time. Overall, this research advances the state-of-the-art in knowledge graph querying by bridging the gap between formal logic-based methods and the practical need for handling vague, context-dependent constraints.
👉 Pročitaj original: arXiv AI Papers
