Introduction
Design tasks, such as architectural planning, often involve "ill-defined" problems where requirements are not fully understood at the outset. In these scenarios, "problem setting" and "problem solving" co-evolve; as a designer works on a solution, they uncover new constraints and goals. Traditional information retrieval (IR) systems struggle here because they require precise, pre-defined queries. When a designer lacks a clear query, navigating vast libraries of examples becomes cognitively taxing. This paper introduces CATALOGEXPLORER, a system designed to bridge this gap by automatically retrieving relevant design knowledge based on the emerging task.
The Multifaceted Architecture
The authors propose a multifaceted architecture for design environments, implemented within the JANUS system. This architecture integrates five core components to support the design process:
- Construction Kit: A palette for building artifacts (e.g., kitchen floor plans).
- Argumentative Hypermedia: A repository of design rationale, including issues, answers, and arguments.
- Catalog: A library of prestored design examples.
- Specification Component: A tool for defining high-level goals and priorities.
- Simulation Component: A tool for testing usage scenarios.
The system's intelligence relies on specification-linking rules. These rules map high-level "hidden" features (like "safety") to concrete physical conditions (like "the stove must be at least 12 inches from the door"). These rules are derived from the argumentative hypermedia, ensuring that system suggestions are grounded in established design rationale.
Key Insights: Situations "Talking Back"
A central theme of the work is the idea that the design situation should "talk back" to the user. Borrowing from Donald Schön, the authors argue that when a designer makes a choice in the construction kit, the system should provide immediate feedback by highlighting relevant arguments or suggesting safer alternatives from the catalog.
Furthermore, the system distinguishes between surface features (e.g., "kitchens with islands") and hidden features (e.g., "suitability for a small family"). By using domain knowledge to infer relevance based on hidden features, CATALOGEXPLORER provides more meaningful assistance than simple keyword-based search.
System Implementation and Results
The CATALOGEXPLORER prototype demonstrates an incremental "retrieval by reformulation" process. As users adjust specifications or weights, the system re-ranks catalog examples in real-time using an appropriateness metric—a weighted sum of satisfied conditions.
The prototype shows that capturing a user's "task at hand" through partial specifications allows the system to shift the user's burden from "searching" to "evaluating suggestions," significantly reducing the cognitive load of information retrieval in complex design tasks.
Limitations and Future Work
While successful as a prototype, the authors note several areas for improvement. Current information spaces are relatively small, and scaling to massive libraries remains a challenge. Additionally, specification-linking rules are currently derived manually from textual arguments; future work aims to automate this process. Other goals include enriching catalog examples with richer backstories and developing mechanisms to detect contradictions between a user's stated specifications and their actual constructions.
Summary updated