CoNode: Visualizing Workflows for Knowledge Reuse and Recombination in Team–AI Collaborative Design

Paper Title

CoNode: Visualizing Workflows for Knowledge Reuse and Recombination in Team–AI Collaborative Design

Publication Info

• Topic area: Generative AI and collaborative design processes.
• Keywords: Generative AI, collaborative design, process knowledge, workflow visualization, industrial design, team–AI collaboration, semantic–visual reasoning, knowledge reuse, creative recombination, CoSense.

Background and Problem

• Problem / challenge: Early-stage industrial design generates fragile process knowledge (e.g., sketches, semantic tags) that is rarely captured or reused systematically. Existing AI tools focus on generating outputs but fail to preserve, trace, or recombine design reasoning, leading to fragmented artifacts, context loss, and inefficient collaboration.
• Significance: Addressing these challenges is critical for improving design quality, enabling shared understanding, and enhancing team efficiency in industrial design workflows.
• Motivation and related work: Prior research highlights generative AI's role in expanding design spaces and supporting cross-modal exploration but lacks mechanisms for preserving and reusing process knowledge. Existing tools often focus on linear interactions and outcome generation, leaving gaps in traceability and multi-round collaboration.

Solution

• Proposed approach: CoNode—a two-layer system that embeds generative AI nodes into a shared whiteboard and structures design reasoning through triplet workflows. It includes the CoSense module for proactive knowledge consolidation, reuse, and recombination.
• Novelty:
  1. Embedding generative AI nodes directly into a shared whiteboard for seamless semantic and visual exploration.
  2. Structuring fragmented design reasoning as traceable triplet workflows (input → AI node → output).
  3. Introducing workflow-level features and the CoSense module to consolidate, reuse, and recombine design knowledge.
  4. Supporting non-linear, cross-modal, and iterative team–AI collaboration.
• Procedure and key techniques:
  • Layer 1: Embeds AI nodes into a shared whiteboard, enabling semantic and visual ideation through operations like divergence, decomposition, summarization, and image-to-text.
  • Layer 2: Provides workflow-level features, including a workflow library for saving and reusing paths, and CoSense for consolidation and recommendation.
  • Implementation: Combines rule-based retrieval and LLM-based augmentation for proactive suggestions and semantic expansion.

Results

• Concrete findings:
  • CoNode significantly improved knowledge consolidation (M = 5.82 vs. baseline M = 4.22; p < .001), reuse (M = 5.83 vs. baseline M = 4.40; p < .001), and creative recombination (M = 5.35 vs. baseline M = 4.62; p = .013).
  • Behavioral data showed higher content block collections (M = 8.60 vs. baseline M = 3.00; p < .001) and workflow reuse rates (63.6% vs. baseline).
  • Teams using CoNode produced more final outputs (M = 8.47 vs. baseline M = 5.87; p = .004).
• Advantage over baselines:
  • CoNode outperformed the baseline system in supporting shared understanding, reducing redundant work, and enabling efficient reuse and recombination of design knowledge.
  • Its workflow-level features and CoSense module provided unique advantages in maintaining continuity and fostering creative exploration.
• Experiments / evaluation:
  • Study I (N = 12): Validated CoNode's foundational interaction paradigm, showing high usability (SUS score M = 77.48, SD = 7.81) and effective support for non-linear exploration.
  • Study II (N = 30): Compared CoNode against a baseline system, demonstrating significant improvements in process knowledge management and team collaboration efficiency.
• Limitations and future work:
  • Limited study duration prevents assessment of long-term knowledge evolution.
  • Scalability challenges in larger teams and distributed collaboration settings.
  • Opportunities to enhance generative capabilities (e.g., 3D modeling, counterfactual prompts) and refine recombination mechanisms to mitigate fixation.

Summary

CoNode introduces a novel interaction paradigm for generative AI in collaborative design, embedding AI nodes into a shared whiteboard and structuring reasoning through triplet workflows. Its workflow-level features and CoSense module enable efficient consolidation, reuse, and recombination of design knowledge, addressing fragmentation and context loss in early-stage ideation. Two user studies demonstrate CoNode's effectiveness in supporting non-linear, cross-modal exploration and enhancing team collaboration. By shifting AI from outcome generation to process knowledge evolution, CoNode offers valuable insights for advancing AI-powered creativity tools in industrial design.