LinkQ: An LLM-Assisted Visual Interface for Knowledge Graph Question-Answering

TL;DR

LinkQ addresses the challenge of querying knowledge graphs by introducing an LLM-assisted, multistep workflow that converts natural language questions into well-formed SPARQL queries while grounding outputs in ground-truth KG data. The system orchestrates natural language interpretation, ID retrieval via KG APIs, and transparent query generation with previews and visualizations (entity-relations and query graphs), achieving low hallucination risk. A qualitative study with five KG practitioners shows strong perceived value, with insights into trust-enabled explanations and the importance of interactive visuals, though it also highlights instances of inaccuracies and the need for user guidance. Overall, LinkQ demonstrates a practical pathway for visible, iterative KG querying and exploratory data analysis driven by LLMs, with clear directions for quantitative evaluation and enterprise-data extension.

Abstract

We present LinkQ, a system that leverages a large language model (LLM) to facilitate knowledge graph (KG) query construction through natural language question-answering. Traditional approaches often require detailed knowledge of a graph querying language, limiting the ability for users -- even experts -- to acquire valuable insights from KGs. LinkQ simplifies this process by implementing a multistep protocol in which the LLM interprets a user's question, then systematically converts it into a well-formed query. LinkQ helps users iteratively refine any open-ended questions into precise ones, supporting both targeted and exploratory analysis. Further, LinkQ guards against the LLM hallucinating outputs by ensuring users' questions are only ever answered from ground truth KG data. We demonstrate the efficacy of LinkQ through a qualitative study with five KG practitioners. Our results indicate that practitioners find LinkQ effective for KG question-answering, and desire future LLM-assisted exploratory data analysis systems.

Figures (2)

• Figure 1: Example of how an LLM can hallucinaterawte2023survey or misrepresent the data in a KG. Here, we ask ChatGPT brown2020language to write a query about gold medal winners from the 2020 Summer Olympics. In total, ChatGPT wrote three consecutive failing queries with incorrect IDs, while stating to us that it had corrected its previous mistakes.
• Figure 2: Illustrative state-machine of LinkQ's system architecture, as described in Section \ref{['sec:architecture']}. In our pipeline, the LLM, System, and User have distinct roles, responsible for communicating with one another to complete different tasks across different states.