Query Relaxation for LLM-Generated SPARQL Queries over Building Knowledge Graphs

Дата публикации: 18-08-2026 00:00:00

When Knowledge Graph (KG) queries fail to match a pattern in a KG, they return no results. Identifying the statements causing these failures is tedious, especially for LLM-generated queries, which tend to be longer and more complex than queries written by hand. Query relaxation addresses this by systematically loosening query constraints until results are recovered. To evaluate the effectiveness of query relaxation against LLM generated queries, we propose a two-stage relaxation method combining triple deletion and path relaxation and test it against 1,823 failed queries for building KGs.

Схожие новости

#	Наименование новости	Тональность	Информативность	Дата публикации
1	A unified large language model–based framework for heterogeneous PV image diagnosis	0	9.5	18-08-2026
2	UQLM: A Python Package for Uncertainty Quantification in Large Language Models	0	10.29	17-08-2026
3	Как я построил Graph RAG систему с точностью 96.7% за 5 дней: от научных статей до production-ready пайплайна	0	20.89	25-02-2026
4	LLMRouter - Library for LLM Routing	0	10	08-02-2026
5	RAG vs. Agentic RAG: Architecture, Tradeoffs, and How to Choose	0	7.77	29-07-2026
6	Оптимизация запросов под выдачу ИИ	0	5	06-05-2026
7	Appier Research Unveils Agentic AI Breakthrough: A Risk-Aware Decision Framework	5	7	10-03-2026
8	TileRT - Tile-Based Runtime for Ultra-Low-Latency LLM Inference	0	35	28-06-2026
9	mongosense 0.1.0	0	37.27	14-08-2026

Классификация: Пресс-релизы. Схожих патентов: 0. Схожих новостей: 9. Тональность: 0. Информативность: 8.02. Источник: escholarship.org.