[Paper Review] Synthesis of Attributed Feature Models From Product Descriptions: Foundations

Guillaume Bécan, Razieh Behjati+2 more|arXiv (Cornell University)|Feb 16, 2015

Advanced Software Engineering Methodologies67 references3 citations

TL;DR

This paper presents a sound, complete, and parametrizable algorithm for synthesizing attributed feature models (AFMs) from product descriptions using configuration matrices. It enables automatic extraction of feature hierarchies, attribute placements, domain values, and constraints, with theoretical and empirical validation of scalability up to large domain sizes and configuration sets.

ABSTRACT

Feature modeling is a widely used formalism to characterize a set of products (also called configurations). As a manual elaboration is a long and arduous task, numerous techniques have been proposed to reverse engineer feature models from various kinds of artefacts. But none of them synthesize feature attributes (or constraints over attributes) despite the practical relevance of attributes for documenting the different values across a range of products. In this report, we develop an algorithm for synthesizing attributed feature models given a set of product descriptions. We present sound, complete, and parametrizable techniques for computing all possible hierarchies, feature groups, placements of feature attributes, domain values, and constraints. We perform a complexity analysis w.r.t. number of features, attributes, configurations, and domain size. We also evaluate the scalability of our synthesis procedure using randomized configuration matrices. This report is a first step that aims to describe the foundations for synthesizing attributed feature models.

Motivation & Objective

• To address the gap in existing reverse engineering techniques that do not synthesize feature attributes, domain values, or attribute constraints.
• To formalize the relationship between configuration matrices and attributed feature models (AFMs), ensuring semantic fidelity.
• To design a parametrizable synthesis algorithm that computes all possible valid AFMs from input product descriptions.
• To evaluate the theoretical and practical time complexity of the synthesis procedure with respect to features, attributes, configurations, and domain size.
• To lay the foundational basis for tool-supported reverse engineering of AFMs in software product line engineering.

Proposed method

• The approach uses a configuration matrix formalism to represent product descriptions, encoding Boolean features and numerical attributes across multiple configurations.
• The algorithm computes all possible feature hierarchies and feature groups (e.g., xor, or) using a constraint-based inference mechanism.
• Attribute placements are determined by analyzing co-occurrence patterns and consistency across configurations, ensuring semantic correctness.
• Domain values are extracted by clustering attribute values across configurations, with constraints inferred via implication and dependency analysis.
• Binary implications and complex constraints (e.g., RC constraints) are computed using time complexities of O(v²·c·d) and O(v⁴·d + v²·d²), respectively, where v is the number of values, c configurations, and d attributes.
• The synthesis procedure is parametrizable, allowing users to inject prior knowledge (e.g., hierarchy or attribute placement) to guide the output.

Experimental results

Research questions

• RQ1 How can attributed feature models be synthesized soundly and completely from a set of product descriptions?
• RQ2 What is the theoretical and practical time complexity of synthesizing AFMs with respect to the number of features, attributes, configurations, and domain size?
• RQ3 To what extent does the algorithm scale with increasing domain size and number of configurations?
• RQ4 How does the algorithm handle over-approximation in the resulting AFM compared to the input configuration matrix?
• RQ5 What are the dominant computational bottlenecks in the synthesis pipeline, and how can they be optimized?

Key findings

• The algorithm is sound and complete with respect to the configuration semantics of the input matrix, ensuring no valid AFM is missed and no invalid model is generated.
• The practical time complexity grows approximately as the square root of the execution time, which correlates linearly with the maximum domain size (correlation coefficient 0.932).
• The computation of binary implications and complex constraints (RC) accounts for over 90% of the total execution time, making them the primary performance bottlenecks.
• The number of attributes has minimal impact on the maximum domain size that the algorithm can handle, indicating scalability independent of attribute count.
• The algorithm successfully handles configuration matrices with up to 1000 configurations and domain sizes up to 2000, demonstrating strong practical scalability.
• Manual validation and test suites with minimal AFM constructs (e.g., single xor-groups) confirmed the correctness of the implementation with no detected anomalies.