Coreset-Guided Personalized Summarization of Knowledge Graphs
A new method for personalized summarization of large Knowledge Graphs (KGs) is proposed, using coreset theory to sample relevant triples based on user-specific query workloads. The approach, detailed in arXiv:2605.14900, applies sensitivity-based importance sampling to create compact summaries that approximate full dataset characteristics. These summaries reduce storage and query runtime while maintaining information relevant to individual users. The work addresses the challenge of unwieldy KGs in tasks like question answering and visualization, offering a viable alternative through personalized summarization.
Key facts
- Knowledge Graphs are extensively used across domains but often very large.
- Summarization offers a viable alternative for tasks like question answering and visualization.
- Personalized KG summarization captures specific user requirements based on query patterns.
- The method adapts coreset theory to create personalized KG summaries.
- Sensitivity-based importance sampling selects a relevant subset of triples.
- The subset approximates characteristics of the full dataset with bounded error.
- Personalized summaries result in smaller storage requirements and query runtime.
- The paper is published on arXiv with identifier 2605.14900.
Entities
Institutions
- arXiv