Visualization Over Large Language Model Using Knowledge Graph

The visualization model using LLM primarily operates based on pre-trained materials. However, these models may experience hallucination problem when dealing with knowledge not found in the training set. In this paper, we propose VisKL, a novel model that synergizes knowledge graphs with large language models in a bidirectional manner, which addresses the above problem through two parts: (i) a LLM-augmented KG construction module that uses LLM to assist in KG construction, and (ii) a KG-augmented LLM visualization module that extracts triples from the knowledge graph, and integrates triples with tasks to visualize through LLM. We evaluate VisKL with other LLM-driven visualization models across multiple tasks and explain the enhancement of VisKL through case study, as well as the impact of different large language models on visualization results.