Original title: Exploratory Unstructured Data Analysis: A Formative Study and Implications for Human-AI Collaboration
Johannes Eschner, Dominik Eitler, Max Irendorfer, Patrick Kramml +2 · September 2026
Looking through messy, unorganized files like huge sets of pictures is hard because there is no built-in system to organize them. To fix this, researchers introduced a new framework called EluDA to help people explore unstructured dataInformation like images or text that doesn't have a predefined format or labels. by mixing standard tools, like queries and visual charts, with active organizing. In a test with an image collection, people built their own systems from the bottom up by making multi-trait categories, rather than grouping items across physical space. The team also tested an AI model named CLIP. They found that CLIP struggles to accurately label pictures with user-created concepts, but it works well for grouping items by general meaning. These insights point to new ways people and AI can work together.
Imagine you're sorting through a large box of mixed spices without labels. You start by grouping them by smell: sweet, savory, spicy. You don't lay them out in a grid by color or size; instead, you build groups as you go. In this paper, when people explore data, they do something similar—they create categories based on features step by step, not by spatial arrangement. The AI tries to help by grouping similar items, but it can't always match the specific groups you have in mind.
When dealing with large unstructured data like images, understanding how people naturally organize it can lead to better AI tools that assist without overwhelming. This helps researchers and analysts make sense of data faster, which is crucial in fields from healthcare to social media.
The abstract doesn't specify the scale or diversity of the image dataset used in the study. The framework is conceptual and hasn't been fully tested in practice. CLIP's unreliability for assigning user-defined concepts suggests it may not be suitable for all applications without further improvement.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Exploratory Unstructured Data Analysis: A Formative Study and Implications for Human-AI Collaboration — Johannes Eschner, Dominik Eitler, Max Irendorfer et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.