# ViStruct: Simulating Expert-Like Reasoning Through Task Decomposition and Visual Attention Cues

arXiv: https://arxiv.org/abs/2506.21762 · HTML: https://mlsanity.com/paper/2506.21762/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Multimodal reasoning, one of 17 directions in LLM reasoning.
- Builds on 4 of its 34 references that are in the map.
- First posted on arXiv 26 Jun 2025.

## Placed in

- LLM reasoning / [Multimodal reasoning](https://mlsanity.com/lr/multimodal-reasoning/)

## Builds on (most cited first, 4 of 4)

- [ReAct: Synergizing Reasoning and Acting in Language Models](https://mlsanity.com/paper/2210.03629/) (2022)
- [Do Vision-Language Models Really Understand Visual Language?](https://mlsanity.com/paper/2410.00193/) (2024)
- [Unraveling the Truth: Do VLMs really Understand Charts? A Deep Dive into Consistency and Robustness](https://mlsanity.com/paper/2407.11229/) (2024)
- [RelationVLM: Making Large Vision-Language Models Understand Visual Relations](https://mlsanity.com/paper/2403.12801/) (2024)

## Built on by (most cited first, 1 of 1)

- [Multimodal Information Fusion for Chart Understanding: A Survey of MLLMs -- Evolution, Limitations, and Cognitive Enhancement](https://mlsanity.com/paper/2602.10138/) (2026)

## Abstract (arXiv)

Data visualization tasks often require multi-step reasoning, and the interpretive strategies experts use, such as decomposing complex goals into smaller subtasks and selectively attending to key chart regions are rarely made explicit. ViStruct is an automated pipeline that simulates these expert behaviours by breaking high-level visual questions into structured analytic steps and highlighting semantically relevant chart areas. Leveraging large language and vision-language models, ViStruct identifies chart components, maps subtasks to spatial regions, and presents visual attention cues to externalize expert-like reasoning flows. While not designed for direct novice instruction, ViStruct provides a replicable model of expert interpretation that can inform the development of future visual literacy tools. We evaluate the system on 45 tasks across 12 chart types and validate its outputs with trained visualization users, confirming its ability to produce interpretable and expert-aligned reasoning sequences.