GraphScape: A Model for Automated Reasoning about Visualization Similarity and Sequencing
summary
GraphScape models visualization space as a directed graph whose nodes are grammar-based chart specs (implemented with Backlight) and whose edges are atomic edit operations (marks, transforms, encodings) weighted by interpretive cost. Transition costs derive from an expert-ranked partial order solved by a linear program, plus a global discount that rewards repeated subsequence patterns. A controlled study with 51 graduate students shows GraphScape’s top suggestion matches user preferences in all but one case. Code is on GitHub repository.
AI-written summary · about
chapters (14)
- 0:00Introduction to GraphScape
- 2:35What is GraphScape?
- 3:19GraphScape model overview
- 4:21Finding design alternatives
- 5:04Elaborating transitions
- 5:50Recommending chart sequences
- 6:12Building transition cost model
- 8:22User study validation
- 10:09Global structure preferences
- 11:36Refining sequence cost
- 13:23Sequence rating discounting
- 14:05Evaluating sequence cost
- 15:32Applications and conclusion
- 15:55Q&A: Future research directions
related talks
01h 41mGraph Intelligence: Enhance Reasoning and Retrieval Using Graph Analytics
16mType-Directed Synthesis of Visualizations from Natural Language Queries
30mAgent Reasoning with Graph World Models
56mConnecting the Dots with Context Graphs
21mLinked Visualisations via Galois Dependencies
17mContext Graphs for Explainable, Decision-Aware AI Agents
31mGraph Analytics - From the Whiteboard to Your Toolbox
21mAgentic GraphRAG: Autonomous Knowledge Graph Construction and Adaptive Retrieval