Context-based Modeling of Activity in Real-World Projects
Description
Context-based Modeling of Activity in Real-World Projects presents a synthesis of 25 years of research on modeling and using context in real-world applications in a very large spectrum of domains, which allows us to illustrate the keystone aspects of context from an initial operational definition; this opens up a four-level framework under conceptual, operational, implementation and environment aspects of activity modeling.
The result is the Contextual-Graphs (CxG) formalism, thanks to strong connections between context and an actor's focus of attention, leading to a uniform representation of knowledge, reasoning and context for actor and group activity. The results of this research constitute the building blocks for designing future types of AI systems, namely the context-based intelligent assistant systems.
This book presents the proceduralized context as a new definition of context, that is a real-time definition, which is then applied to context modeling for actor or group activity - before finally elaborating the two versions of the CxG formalism including uses in different modeling.
More details
Other editions
Additional editions
Content
Preface ix
Acronyms xiii
Introduction xv
Chapter 1 Theoretical and Pragmatic Basis of Our Context Modeling 1
1.1 Introduction 1
1.2 Theoretical basis 2
1.2.1 Model and representation 2
1.2.2 Concepts used 11
1.2.3 The proposed scientific approach 22
1.3 Projects and applications 28
1.3.1 Introduction 28
1.3.2 The SEPT project (1986-1995) 29
1.3.3 The SART project (1996-2002) 31
1.3.4 The FlexMIm project (2012-2015) 34
1.3.5 The TACTIC project (2013-2015) 35
1.3.6 The ACA project (2005-2009) 38
1.3.7 "Contextualizing scientific workflows" project (2009-2011) 40
1.3.8 The MICO project (2011-2014) 43
1.3.9 The winemaking project (1997-1999) 44
1.3.10 The OSSMOSE project (2009-2011) 46
1.3.11 "Computer-mediated collaborative work" project (1997-1999) 48
1.4 Review of the chapter 50
Chapter 2 Context Modeling for Actor Activity (CxG_1.0) 53
2.1 Introduction 53
2.2 Conceptual level 55
2.2.1 Contextual knowledge and external knowledge 55
2.2.2 Conceptual elements used in the research 57
2.3 Operational level 63
2.3.1 Operational knowledge and context 63
2.3.2 Contextual elements 68
2.3.3 Mental models and mental representations 69
2.4 Implementation level 71
2.4.1 The CxG_1.0 formalism 71
2.4.2 Modeling tools 72
2.4.3 Exploitation tools 79
2.4.4 Review on the CxG_1.0 formalism 88
2.5 Environment level 89
2.6 Variants of the context modeling 91
2.6.1 Introduction 91
2.6.2 The COM 92
2.6.3 Parallel with BPEL for workflows 95
2.6.4 Comparison with two other context-based formalisms 98
2.7 Review of the chapter 103
Chapter 3 Context Modeling for Group Activity (CxG_2.0) 109
3.1 Introduction 109
3.2 Conceptual level 110
3.2.1 Introduction 110
3.2.2 Concept of activity for a group 111
3.2.3 The shared context 113
3.2.4 The turn 117
3.2.5 CxG-based simulation 118
3.3 Operational level 121
3.3.1 Introduction 121
3.3.2 The notions of group activity and interaction 121
3.3.3 The CxG_2.0 formalism 123
3.4 Implementation level 124
3.4.1 Introduction 124
3.4.2 From actor activity to group activity 124
3.4.3 Reserved contextual elements 125
3.4.4 Mechanisms of the CxG-based simulation 128
3.5 Environment level 136
3.6 Two examples 137
3.6.1 "Submission management" example 137
3.6.2 The TACTIC project 146
3.7 Review of the chapter 152
Chapter 4 The Two Versions of the CxG Formalism 157
4.1 Introduction 157
4.2 The key points of the research 158
4.2.1 Discussion on key concepts of the approach 158
4.2.2 Definition of the context of an activity 166
4.2.3 Contextual reasoning 167
4.2.4 Global and local contexts 175
4.3 The "Internship-offer analysis" example 186
4.3.1 The actors 186
4.3.2 Conditions of the experiment 187
4.3.3 Development of the experiment 188
4.3.4 Interpretation of the results 192
4.3.5 Comparison with the "DVD-reader diagnosis" task 199
4.4 CxG formalism for CIAS Design 204
4.5 Review of the chapter 214
Chapter 5 Use of the CxG Formalism in Different Modeling 221
5.1 Introduction 221
5.2 Breast cancer diagnosis 223
5.3 Hierarchical task analysis (HTA) 229
5.4 The ACA project 237
5.4.1 Introduction 237
5.4.2 The ACA method 237
5.4.3 Application in road safety 240
5.5 Workflow modeling in an ACP Department 249
5.5.1 The initial work 249
5.5.2 Contextual modeling of the workflow 252
5.5.3 Discussion on context-based modeling of workflow 257
5.6 Context modeling and semiotics 258
5.7 Review of the chapter 266
Conclusion 271
References 285
Index 297
Introduction
This book is a summary of 25 years of research on the modeling of context and its use in about 20 real-world projects and applications1. The topic "Modeling and using context" is found in a number of disciplines either directly (i.e. as context) or indirectly (i.e. such as situations, constraints, etc.), but we do not present a thorough study of the literature in all the disciplines. Most of the references cited in this book come from Artificial Intelligence (AI) and Computer Science, and from some related disciplines. We made (and continue) a collection of definitions of context (271 in 2024 found in about 30 different domains). The success of our research comes from the challenges arising from the "real-world" projects and applications upon which we have worked and that obliged us to avoid arbitrary simplification. In this book, we present 10 of these projects and applications that have been selected to show that our modeling and use of context cover conceptual aspects, operational aspects and implementation aspects, thanks to the role of environment inactivity modeling being made explicit.
The research was led using AI, from the modeling of task realization, decision-making and problem-solving that we consider now as "activity" to the operational concepts "mental model" and "mental representation" coming from Cognitive Sciences. There are two reasons developed hereafter. First, the activity performed by an actor is closer to the actor's experience than the "focus of attention" that only concerns the current point of the activity development. Second, the activity is larger than the realization of a task by including other elements such as the actor and the environment, far from a task model itself.
Our interest in developing an AI system in decision-making led us first to establish an operational definition of context and then design and implement a model of context in concrete terms for actors with whom we collaborated in very different domains. The research presented in this book has been the object of numerous papers published in different journals, reviews and conferences; thus, we will not enter into all the details of the results obtained. We limit here the research presentation to what is necessary for a coherent view of the research progression on modeling and use of context and the main lessons learned in our research. Our objective is to propose a formalization for approaching the fuzzy concept of "context" from an AI viewpoint - as was done with the concepts of mass, length, speed, etc. in Physics and Biology - for modeling how an actor and a group of actors have an activity in real-world projects or applications. Modeling context for a given activity is supposed to clearly establish its close relationships with human knowledge and reasoning such as in real-world applications. The midterm goal of our context modeling is to establish a pragmatic approach to design, develop and implement context-based AI systems that rely on operational knowledge used by humans in relationships with their activities. Making context explicit makes it possible to use human experience that encompasses the nature of the task, the situation in which the activity must be accomplished, and the available means within the environment that are necessary for that. Human, task, situation and environment are strongly intertwined, and context emerges as the expression of the whole.
A preliminary step for modeling the concept of "context" is to clarify the starting point of our research and provide the readers, before starting the first chapter, with a shared context for the benefit of the lessons learned and provided in this book. We thus want to begin by giving our understanding of different common questions about context such as What is context? Which definition of context? And why do we need context?
To become informed about a topic we generally use a search engine. The first observation is that the information provided depends on the search engine. In 1996, we found about 1,000 links by Yahoo! and 650,000 by Google (Brézillon 1999). Our annual count of the number of webpages containing the word "context" showed an exponential increase over 10 years from 650,000 in 1996 to 325,000,000 webpages in 2006 (after 2006, Google filtered unused webpages). About 90% of the pages were not relevant for our purpose of modeling context (e.g. containing ". . . in this context, we think that . . ."). Nevertheless, the number of relevant pages increased by 500 during this interval of 10 years. In 1999, we made an interdisciplinary survey of the literature (e.g. for our collection of 271 definitions), but it rapidly became clear that such a survey was only possible at the level of each discipline. The situation was aggravated by the move of bibliographic references from printed paper to digital ones. The consequence was that references had to "be on the Web or they do not exist" in most of the disciplines, and the corollary is a lack of accessibility to old papers judged as "THE" references in some disciplines (especially books). With respect to our research, the bibliography we made is mainly on what was accessible on the Web (and updated frequently) and not on what is only printed paper, especially in other disciplines.
The second important point was finally to ask for the definition given by each author because one discovers that each definition was (almost systematically) criticized for its lack of generality. A look at our collection of definitions shows the accuracy of this criticism. One surprise was to discover that the notion of context plays a role in a number of very different disciplines, although this role may be an excuse to simplify discipline problems. For example, three definitions are: in AI, "a context is a collection of relevant conditions and surrounding influences that make a situation unique and comprehensible"; in philosophy, context is "a body of information available to participants in the speech situation"; in economy, context is compared to "the fibers of a rope for describing the conditions of a system." The italics words in the definitions correspond to the actor's focus of attention that is concerned by the context in the definition. Some immediate observations can be made: (1) there seem to exist as many definitions as authors (and sometimes several definitions per author), (2) few definitions make it possible to formalize context to make it operational, (3) definitions are given in terms of the discipline (i.e. highly contextualized) and (4) even if definitions are "author-dependent", some generic features emerge, linking context to activity.
The third challenge is to address the need for context that appears as soon as humans and their activities intervene (Why do we need context?). We consider activities with an operational purpose (especially modeling and simulation). When humans realize a task, make a decision, solve a problem in the real world, they adapt methods, tools, etc., during the process for integrating the constraints of the task, the situation, the environment and also their preferences, emotions, etc., that is context. Our question then is how an AI system can effectively use context for exploiting the human's knowledge and reasoning put into the machine without enumerating all the contextual variants. This supposes, on the one hand, a coherent view on knowledge, reasoning and context in a uniform representation, and, on the other hand, an efficient software for modeling human focus of attention in context. In all our projects and applications for large (and small) enterprises, the human that we considered is an actor that faces an activity.
The book addresses the previous questions in five chapters covering all the aspects of modeling and use context in real-world projects and applications. Note that the important points of our research from Chapter 2 to Chapter 3 are called "key points" (KPs) in the text and are summed up and discussed in Chapter 4.
Chapter 1 presents the scientific foundation of our research based on the distinction between model and representation in mathematics and the concepts of interest such as activity, mental model and mental representation. Section 1.2 is dedicated to our scientific approach, and the framework with its four modeling levels, which is used for activity modeling. Section 1.3 introduces the main real-world projects and applications that rely on these 25 years of research in a large spectrum of domains.
A model is an abstraction of the reality that is used in a large number of disciplines. A model can be formalized in numerous formalisms of representation. If you do not have the right representation formalism for what you are looking for, you may miss some relevant observations in reality and/or obtain a model that is not relevant for testing your assumptions about reality. We illustrate the situation with two examples: the autocatalytic model in Biochemistry and a formal model of the calcium metabolism in Physiology. In these disciplines, a mathematician and a biologist interacted on a given model in two different representation formalisms, ordinary differential equations for the mathematician and a compartmental formalism for the biologist. This is the origin of our four modeling levels.
The role played by context has been identified by leading an approach as scientific as possible, such as in Biology for modeling...
