Knowledge-Assisted Visual Analytics Utilizing Multimodal Models

This work explores how multimodal models can support Knowledge-Assisted Visual Analytics (KAVA) by automatically generating and evolving a structured knowledge model from unstructured image data. Leveraging recent advances in vision-language models and traditional NLP techniques, the system extracts and clusters semantic concepts to initialize a graph-based knowledge structure. A prototype implementation will enable human-machine collaboration through interactive refinement operations like grouping and splitting, addressing the cold start problem in KAVA systems.