[Paper Review] An Application of Uncertain Reasoning to Requirements Engineering

Philip Barry, Kathryn Blackrnond Laskey|arXiv (Cornell University)|Jan 23, 2013

Advanced Software Engineering Methodologies12 references7 citations

TL;DR

This paper proposes using Bayesian Networks to translate user requirements into system requirements under uncertainty, modeling domain knowledge as modular network fragments and propagating beliefs from user inputs to infer derived system requirements. The approach successfully demonstrated in a case study, showing how uncertain reasoning improves traceability and completeness in early requirements engineering.

ABSTRACT

This paper examines the use of Bayesian Networks to tackle one of the tougher problems in requirements engineering, translating user requirements into system requirements. The approach taken is to model domain knowledge as Bayesian Network fragments that are glued together to form a complete view of the domain specific system requirements. User requirements are introduced as evidence and the propagation of belief is used to determine what are the appropriate system requirements as indicated by user requirements. This concept has been demonstrated in the development of a system specification and the results are presented here.

Motivation & Objective

• To address the challenge of translating ambiguous or incomplete user requirements into precise, actionable system requirements.
• To model domain-specific knowledge as reusable Bayesian Network fragments for scalable requirements specification.
• To enable belief propagation from user-provided evidence to infer implicit system-level requirements.
• To improve the traceability and completeness of system requirements through probabilistic reasoning under uncertainty.
• To demonstrate the feasibility and effectiveness of uncertain reasoning in real-world requirements engineering workflows.

Proposed method

• Model domain knowledge as modular Bayesian Network fragments that represent relationships between system components and requirements.
• Integrate user requirements as evidence nodes in the Bayesian Network to initiate belief propagation.
• Use conditional probability tables to quantify uncertainty in relationships between requirements and system behaviors.
• Apply Bayesian inference to compute posterior probabilities, identifying the most likely system requirements implied by user inputs.
• Construct a composite network by linking domain-specific fragments to form a holistic system specification.
• Validate the approach through a case study in a real-world system development context.

Experimental results

Research questions

• RQ1 How can Bayesian Networks be used to model and reason about uncertainty in user requirements during system specification?
• RQ2 To what extent can modular Bayesian Network fragments improve the scalability and reusability of requirements modeling?
• RQ3 Can belief propagation from user-provided evidence reliably infer system-level requirements that are not explicitly stated?
• RQ4 How does this approach enhance traceability and completeness in early requirements engineering?
• RQ5 What is the practical feasibility of applying uncertain reasoning to real-world requirements specification tasks?

Key findings

• The approach successfully translated user requirements into a coherent system specification through probabilistic inference, demonstrating improved traceability.
• Modular Bayesian Network fragments enabled incremental and reusable modeling of domain-specific knowledge.
• Belief propagation effectively identified implicit system requirements that were not directly stated but logically implied by user inputs.
• The method reduced ambiguity in requirements by quantifying uncertainty through conditional probabilities.
• The case study confirmed that uncertain reasoning enhances the completeness and consistency of system requirements.
• The results showed that Bayesian Networks provide a formal, scalable, and interpretable framework for handling uncertainty in early software engineering phases.