KNOWNET: Guided Health Information Seeking from LLMs via Knowledge Graph Integration

TL;DR

This work tackles the risk of misinformation in health information retrieved from large language models by introducing KNOWNET, a visualization system that tightly integrates LLM outputs with external knowledge graphs. KNOWNET extracts structured triples from LLM responses, maps them into a validated KG, and provides literature-backed verification to improve accuracy. It also offers structured, progressive exploration by leveraging KG neighborhoods to generate next-step recommendations and guide users through a focused, stepwise graph visualization that bridges prior inquiries with current questions. The approach demonstrates effectiveness through use cases in dietary supplements and expert interviews, highlighting improved interpretability, evidence grounding, and user-guided exploration, with potential applicability to broader health domains and knowledge-intensive tasks.

Abstract

The increasing reliance on Large Language Models (LLMs) for health information seeking can pose severe risks due to the potential for misinformation and the complexity of these topics. This paper introduces KNOWNET a visualization system that integrates LLMs with Knowledge Graphs (KG) to provide enhanced accuracy and structured exploration. Specifically, for enhanced accuracy, KNOWNET extracts triples (e.g., entities and their relations) from LLM outputs and maps them into the validated information and supported evidence in external KGs. For structured exploration, KNOWNET provides next-step recommendations based on the neighborhood of the currently explored entities in KGs, aiming to guide a comprehensive understanding without overlooking critical aspects. To enable reasoning with both the structured data in KGs and the unstructured outputs from LLMs, KNOWNET conceptualizes the understanding of a subject as the gradual construction of graph visualization. A progressive graph visualization is introduced to monitor past inquiries, and bridge the current query with the exploration history and next-step recommendations. We demonstrate the effectiveness of our system via use cases and expert interviews.

Figures (8)

• Figure 1: KnowNet is designed to support the communication among three distinct forms of knowledge: the knowledge users apply in their reasoning process, the knowledge contained within LLMs, and the knowledge stored in KGs.
• Figure 2: System Overview.
• Figure 3: An overview of the back-end. KnowNet accepts text inputs (a), maps entities in the LLM response to corresponding nodes in the KG based on their embeddings (b), and identifies related entities in the KG neighborhood to generate recommendations based on user exploration histories (c). Finally, KnowNet outputs text responses and visualizations that organize the main relationships, provide evidence from the KG, and suggest next-step recommendations (d).
• Figure 4: Interface of KnowNet. Users can log into the tool via GitHub or Google account and securely store their chat history (A1-A2). The main interface consists of a Text Dialogue (B), a Graphical Explorer (C), and a Navigator (D). Information validation (E) and next-step recommendations (F) are provided to facilitate the exploration.
• Figure 5: Edge labels. KnowNet suggests three different edge labels, support, relevant, and unsure, based on the matching between KG and LLM. $sim(n_1, n_1')$ indicates the cosine similarity between the the entity identified in LLM ($n_1$) and the nodes in KG ($n_1'$). $\theta_n$ and $\theta_r$ are the thresholds for entity and relation matching.