Constrained Entity Selection under Partial Knowledge (CES-PK) for LLM-Based Knowledge Graph QA

Forum topic · 小凯 · 2026-08-27

Summary

This post introduces CES-PK (Constrained Entity Selection under Partial Knowledge), a new problem formulation from arXiv paper 2608.24824 by Emanuel Kitzelmann, aimed at improving grounding in LLM-based knowledge graph question answering (KGQA). Existing LLM-based KGQA approaches either rely on full semantic parsing into executable queries such as SPARQL, which is brittle due to complex schemas and incompleteness of real-world knowledge graphs, or on free-form LLM reasoning over the KG, which is more robust but lacks formal guarantees. CES-PK takes a complementary approach: candidate answers are generated by an LLM-based system and then verified against lightweight symbolic constraints derived from the question. Rather than constructing executable logical forms, the formulation focuses on eliminating invalid answers and providing symbolic support for valid ones. To handle incomplete knowledge graphs, it adopts a three-valued constraint semantics (satisfied, violated, unknown), avoiding false rejections under the open-world assumption.

Paper Overview

Field: AI Author: Emanuel Kitzelmann Published: 2026-08-25 arXiv: 2608.24824

Abstract

Large language models are increasingly used for knowledge graph question answering (KGQA), but can fail to correctly ground answers in the underlying graph. Current approaches to LLM-based KGQA either rely on full semantic parsing into executable queries such as SPARQL, which is brittle in practice due to complex schemas or incompleteness of real-world KGs, or on LLM-reasoning and answer generation over KGs, which can be more robust but lacks formal guarantees.

In this work, the author studies a complementary setting in which candidate answers are generated by an LLM-based system and subsequently verified using lightweight symbolic constraints derived from the question.

Key Contributions

CES-PK (Constrained Entity Selection under Partial Knowledge): a problem formulation that focuses on eliminating invalid answers and providing symbolic support for valid answers, without building executable logical forms.
Three-valued constraint semantics (satisfied, violated, unknown) to account for incomplete knowledge graphs, avoiding false rejections under the open-world assumption.

---

*Auto-collected on 2026-08-27.*