Schemex: Discovering Design Patterns from Examples through Iterative Abstraction and Refinement
Project Overview
The document explores the integration of generative AI in education through Schemex, an AI-powered workflow that facilitates schema induction by employing clustering, abstraction, and refinement techniques. This innovative tool is designed to alleviate cognitive load, particularly in the creation of HCI paper abstracts and the analysis of news TikToks, thereby enhancing the writing process. Schemex engages users in a collaborative cycle where the AI suggests schemas that users can assess and refine, resulting in actionable guidelines applicable to both academic and creative writing. The findings indicate that this approach not only streamlines the writing process but also fosters a collaborative environment where human creativity and AI-generated insights can coexist, ultimately improving educational outcomes by making complex tasks more manageable and interactive.
Key Applications
Schemex
Context: Applied in academic research for writing HCI paper abstracts and analyzing news TikToks, which involves the integration of text, visual, and audio elements. The contexts include generating abstract schemas for academic papers and clustering TikTok examples for script generation.
Implementation: The Schemex workflow clusters examples (including HCI paper abstracts and TikToks) into categories, generates schemas for each cluster, and refines them using AI-generated contrasting examples and evaluations. This iterative process enhances the clarity and quality of the generated outputs.
Outcomes: Achieved high alignment with existing schemas, improved generation of HCI abstracts, and identified new schema insights for TikTok scripts, leading to a better understanding of content creation across different media.
Challenges: Initial schemas may lack precision and require multiple iterations to achieve clarity. Clustering accuracy can be affected by the complexity of the medium, leading to misclassification in some cases.
Implementation Barriers
Technical Barrier
Complexity of clustering diverse examples can lead to misclassification.
Proposed Solutions: Integrate better AI reasoning models to enhance clustering accuracy and provide users with clear explanations of cluster distinctions.
Cognitive Barrier
Schema induction is a high cognitive load task that can overwhelm users.
Proposed Solutions: Utilizing AI to perform the initial heavy lifting and allowing users to focus on high-level evaluations and refinements.
Project Team
Sitong Wang
Researcher
Lydia B. Chilton
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Sitong Wang, Lydia B. Chilton
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai