Enhancing Architecture Frameworks by Including Modern Stakeholders and their Views/Viewpoints

Aug 5, 2026

Upload by Liam

Architecture frameworks for software, systems, and enterprises define stakeholders and modeling perspectives, including viewpoint and view concepts for addressing stakeholder concerns. Existing frameworks, however, omit stakeholders with data science and machine learning (ML) responsibilities, such as data scientists and data engineers, which prevents holistic architecture descriptions for ML-enabled systems. ML behavior must be distinguished from traditional software behavior because functionality is inferred from data and ML structural models use different notations than software engineering artifacts. An empirical online survey with 61 experts supports the proposed extension.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages9

Reading Time23 min

AccessFree to download

Abstract

Introduction

Background on architecture frameworks

Gap: missing ML-related stakeholders

Need for ML-enabled system architecture descriptions

Enhancing Architecture Frameworks by Including Modern Stakeholders and their Views/Viewpoints Armin Moin Department of Computer Science University of Colorado, Colorado Springs CO, USA [amoin@uccs.edu](amoin@uccs.edu) * Corresponding author Atta Badii Department of Computer Science University of Reading Reading, United Kingdom [atta.badii@reading.ac.uk](atta.badii@reading.ac.uk) arXiv :2308 .05239v 3 [ cs . SE] 9 Jan 2025 Stephan G¨unnemann School of Computation, Information, and Technology and Munich Data Science Institute (MDSI) Technical University of Munich Munich, Germany[s.guennemann@tum.de](s.guennemann@tum.de) Moharram Challenger Department of Computer Science University of Antwerp and Flanders Make Antwerp, Belgium [moharram.challenger@uantwerpen.be](moharram.challenger@uantwerpen.be) Abstract—Various architecture frameworks for software, systems, and enterprises have been proposed in the literature. They identified several stakeholders and defined modeling perspectives, architecture viewpoints, and views to frame and address stakeholder concerns. However, the stakeholders with data science and Machine Learning (ML) related concerns, such as data scientists and data engineers, are yet to be included in existing architecture frameworks. Only this way can we envision a holistic system architecture description of an ML-enabled system. Note that the ML component behavior and functionalities are special and should be distinguished from traditional software system behavior and functionalities. The main reason is that the actual functionality should be inferred from data instead of being specified at design time. Additionally, the structural models of ML components, such as ML model architectures, are typically specified using different notations and formalisms from what the Software Engineering (SE) community uses for software structural models. Yet, these two aspects, namely MLand non-ML, are becoming so intertwined that it necessitates an extension of software architecture frameworks and modeling practices toward supporting ML-enabled system architectures. In this paper, we address this gap through an empirical study using an online survey instrument. We surveyed 61 subject matter experts from over 25 organizations in 10 countries. Index Terms—architecture frameworks, viewpoints, views, machine learning, modeling, empirical research I. INTRODUCTION Architecture frameworks provide conventions, principles, and practices for architecture descriptions in a particular application domain or stakeholder community [1] . There exist several well-established examples, including The Open Group Architecture Framework (TOGAF) [2], [3], the U.S. Department of Defense Architecture Framework (DoDAF) [4], the Treasury Enterprise Architecture Framework (TEAF) [5], the British Ministry of Defence Architecture Framework (MODAF) [6], the Zachman Framework [7], the “4+1” View Model of Software Architecture [8] and its updated version with the Decision View [9], as well as the Reference Model of Open Distributed Processing (RM-ODP) [10]–[13] . TOGAF, DoDAF, TEAF, and MODAF were primarily concerned with enterprise architectures. MODAF was replaced by the NATO Architecture Framework (NAF), which in its fourth version (NAFv4) [14] provided guidance not only on describing enterprise architectures but also system architectures for military and business use. Furthermore, the Zachman Framework, a generic framework for information systems and enterprise architectures, and the “4+1” View Model of Software Architecture were among the early works that offered the foundations for the more recent architecture frameworks. Last but not least, RM-ODP was not merely a reference model but a set of four international standards, including an architecture framework for distributed information processing in heterogeneous environments. Each of the above-mentioned architecture frameworks identified stakeholders, such as endusers, soft

FAQs

Why do existing architecture frameworks need to be extended for ML-enabled systems?

Existing frameworks do not include viewpoints or views dedicated to stakeholders with ML-related concerns, which makes it hard to represent ML components and their interactions with non-ML components in a holistic architecture description.

What makes ML component behavior different from traditional software component behavior?

ML functionality should be inferred from data rather than specified at design time, and ML structural models often use different notations and formalisms than those used in software engineering structural models.

How did the paper investigate the proposed architectural gap?

The authors conducted an empirical study using an online survey, collecting responses from 61 subject matter experts across more than 25 organizations in 10 countries.

Aug 5, 2026

Upload by Liam