EA-Analyzer: automating conflict detection in a large set of textual aspect-oriented requirements

Abstract

One of the aims of Aspect-Oriented Requirements Engineering is to address the composability and subsequent analysis of crosscutting and non-crosscutting concerns during requirements engineering. A composition definition explicitly represents interdependencies and interactions between concerns. Subsequent analysis of such compositions helps to reveal conflicting dependencies that need to be resolved in requirements. However, detecting conflicts in a large set of textual aspect-oriented requirements is a difficult task as a large number of explicitly defined interdependencies need to be analyzed. This paper presents EA-Analyzer, the first automated tool for identifying conflicts in aspect-oriented requirements specified in natural-language text. The tool is based on a novel application of a Bayesian learning method. We present an empirical evaluation of the tool with three industrial-strength requirements documents from different domains and a fourth academic case study used as a de facto benchmark in several areas of the aspect-oriented community. This evaluation shows that the tool achieves up to 93.90 % accuracy regardless of the documents chosen as the training and validation sets.

Appendix: Naive Bayes—the Bayesian approach in EA-Analyzer

The problem of detecting conflicts in EA-Analyzer is formulated as a classification problem. Thus, the input vector of features is extracted from the text of the requirements, and an estimated target function maps these features into a set of classes V={Conflict,Harmony}. The aim of the Bayesian approach (Mitchell 1997) is to assign the most probable target value, v _NB , given the input vector of features 〈a ₁,a ₂,…,a _n 〉:

$$ v_{NB}= \operatornamewithlimits {argmax}_{v_j \in V} P(v_j |a_1, a_2,\ldots, a_n)$$

(5)

Equation (6) presents the Bayes theorem:

$$ P(v_j | a_1, a_2,\ldots,a_n) = \frac{P(a_1, a_2,\ldots, a_n | v_j)P(v_j)}{P(a_1, a_2,\ldots, a_n)}$$

(6)

Hence, from the Bayes theorem in Eq. (6), we can rewrite Eq. (5) as follows:

$$ v_{NB}= \operatornamewithlimits {argmax}_{v_j \in V} \frac{P(a_1, a_2,\ldots, a_n | v_j)P(v_j)}{P(a_1, a_2,\ldots, a_n)}$$

(7)

Since the denominator in Eq. (7) does not depend on v _j and the values of a ₁,a ₂,…,a _n are given, this denominator is a constant independent of v _j . Thus, we can rewrite Eq. (7) as follows:

$$ v_{NB}= \operatornamewithlimits {argmax}_{v_j \in V} P(a_1,a_2,\ldots, a_n | v_j) P(v_j)$$

(8)

The Naive Bayes approach assumes that the features are conditionally independent given v _j . In other words, the probability of observing a ₁,a ₂,…,a _n is the product of the probabilities of each individual feature:

$$ P(a_1, a_2,\ldots, a_n |v_j) = \prod_i P(a_i |v_j)$$

(9)

Hence, we can rewrite Eq. (8) as follows:

$$ v_{NB}= \operatornamewithlimits {argmax}_{v_j \in V}P(v_j) \prod_i P(a_i |v_j)$$

(10)