Abstract
This paper introduces a formal framework for handling "soft" (imprecise) requirements in knowledge-based systems using fuzzy logic. By moving away from binary "all-or-nothing" specifications, the authors provide a mathematical way to analyze, quantify, and trade off conflicting goals, such as maximizing profit while minimizing environmental impact.
Introduction
In traditional software engineering, requirements are typically treated as "crisp"—they are either met or they aren't. However, in real-world expert systems, requirements are often vague (e.g., "the response time should be short" or "the cost should be low"). When these imprecise requirements conflict, traditional methods struggle, often labeling the system as inconsistent. This paper addresses the lack of a formal mechanism to capture this impreciseness and systematically evaluate the trade-offs between competing qualitative goals.
Methodology
The authors propose a multi-step approach rooted in fuzzy logic to manage these "soft" requirements:
Soft Requirements via Test-Score Semantics
Requirements are formulated using Zadeh’s canonical form. Instead of a binary pass/fail, a requirement is represented as a fuzzy set where a "degree of satisfaction" (between 0 and 1) is assigned to different state transitions.
Relationship Classification
Mathematical measures are defined for the conflicting degree and cooperative degree between requirements. This allows the system to formally determine if satisfying one goal helps or hinders another.
The Analytic Matrix (AM)
The Analytic Matrix is a key tool for bridging qualitative expert knowledge with quantitative analysis. Experts provide a set of typical "cases" and potential "solutions." The matrix records how well each solution satisfies each requirement, allowing the system to estimate the actual degree of conflict or cooperation between goals.
Iterative Engineering Process
These tools are integrated into a five-step requirement engineering loop: Acquisition, Qualitative Analysis, Quantitative Analysis (via AM), Aggregation, and Evaluation/Prototyping. Conflicts are resolved using "compromise operators" (like the arithmetic mean) from fuzzy multi-criteria optimization to aggregate individual soft requirements into a single, feasible overall objective.
Case Study: ISPBEX
The framework was evaluated using the ISPBEX (Integrated Southern Pine Beetle Expert System), designed for the US Forest Service. The system manages five soft goals: Low Resource, Low Cost, Short Time, Low Environmental Impact, and High Profit. By populating the Analytic Matrix with four typical cases and five major treatment recommendations, the authors were able to quantify specific conflict degrees, such as a 0.8 conflict degree between resource amount and response time.
Key Insights and Results
- Conflict as a Managed Relationship: In this fuzzy framework, conflict is a relationship to be managed rather than an error to be deleted. This allows for more realistic system modeling where "optimal trade-offs" replace non-existent "perfect" solutions.
- Validating Expert Intuition: Quantitative analysis via the Analytic Matrix can refine or correct qualitative expert opinions. In ISPBEX, it revealed cooperative relationships that domain experts had initially overlooked.
- Elastic Constraints: Treating requirements as "elastic" allows the system to find solutions that partially satisfy multiple conflicting goals simultaneously.
Limitations and Future Work
The system's effectiveness relies heavily on the quality of the cases and solutions provided by experts. Future work will explore the selection of appropriate compromise operators and how the relative importance (criticality) of requirements should influence aggregation. Additionally, the scalability of the Analytic Matrix to systems with hundreds of requirements remains to be tested.
Summary updated