142,99 €
Mehr erfahren.
- Herausgeber: John Wiley & Sons
- Kategorie: Wissenschaft und neue Technologien
- Serie: ISTE Invoiced
- Sprache: Englisch
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.
Sie lesen das E-Book in den Legimi-Apps auf:
Seitenzahl: 577
Veröffentlichungsjahr: 2026
Ähnliche
Table of Contents
Cover
Table of Contents
Title Page
Copyright Page
Preface
Acronyms
Introduction
1 Theoretical and Pragmatic Basis of Our Context Modeling
1.1. Introduction
1.2. Theoretical basis
1.3. Projects and applications
1.4. Review of the chapter
2 Context Modeling for Actor Activity (CxG_1.0)
2.1. Introduction
2.2. Conceptual level
2.3. Operational level
2.4. Implementation level
2.5. Environment level
2.6. Variants of the context modeling
2.7. Review of the chapter
3 Context Modeling for Group Activity (CxG_2.0)
3.1. Introduction
3.2. Conceptual level
3.3. Operational level
3.4. Implementation level
3.5. Environment level
3.6. Two examples
3.7. Review of the chapter
4 The Two Versions of the CxG Formalism
4.1. Introduction
4.2. The key points of the research
4.3. The “Internship-offer analysis” example
4.4. CxG formalism for CIAS Design
4.5. Review of the chapter
5 Use of the CxG Formalism in Different Modeling
5.1. Introduction
5.2. Breast cancer diagnosis
5.3. Hierarchical task analysis (HTA)
5.4. The ACA project
5.5. Workflow modeling in an ACP Department
5.6. Context modeling and semiotics
5.7. Review of the chapter
Conclusion
References
Index
Other titles from iSTE in Information Systems, Web and Pervasive Computing
End User License Agreement
List of Tables
Chapter 2
Table 2.1.
Expression of workflow patterns in BPEL and CxG.
...
Chapter 3
Table 3.1.
Comparison of the model- and CxG-based simulations (Brézi
...
Chapter 4
Table 4.1.
List of the key points for modeling and use context
Table 4.2.
List the contextual elements retained for modeling of “In
...
Table 4.3.
Student-centered categorization of contextual elements
Table 4.4.
Grouping of the contextual elements
Chapter 5
Table 5.1.
The GDE matrix for modeling drivers’ behaviors
Table 5.2.
A critical situation, a behavior space and a situation sp
...
Table 5.3.
Workflow table at reception and registration areas of ACP
...
List of Illustrations
Chapter 1
Figure 1.1.
Representations of the autocatalytic model by (a) differ
...
Figure 1.2.
Limit cycle in the phase plane (a) and oscillations in a
...
Figure 1.3.
Decision tree of the procedure for “lack of train power”
...
Figure 1.4.
From a sequence of actions to a macro-action.
...
Figure 1.5.
From tree to graph.
...
Figure 1.6.
The contextual graph for “lack of train power” incident
...
Figure 1.7.
CxG version of the diagnosis of “lack of train power” in
...
Figure 1.8.
Relationship between mental model and situation model
Figure 1.9.
Example of mental representation and mental model in the
...
Figure 1.10.
Two screenshots from Myfitnesspal ().
...
Figure 1.11.
First steps of the scientific approach for Homeostasis
...
Figure 1.12.
Modeling levels of our scientific approach in living sc
...
Figure 1.13.
Compartmental model of the calcium metabolism with circ
...
Figure 1.14.
The three sources of information in a C2 system
.
Figure 1.15.
Tree representation of the “mitosis diagnosis” for a wa
...
Figure 1.16.
Management of the generic framework in OSSMOSE.
...
Chapter 2
Figure 2.1.
Actor activity at four modeling levels
Figure 2.2.
Contextual organization in the incident “Ill traveler in
...
Figure 2.3.
Knowledge organization in a tactical viewpoint (Brézillo
...
Figure 2.4.
Knowledge organization in an operational viewpoint (Bréz
...
Figure 2.5.
Conceptual and operational views on contextual elements.
...
Figure 2.6.
Components of CxG formalism in the CxG software.
...
Figure 2.7.
Example of an ESIA inspired by the MICI application. For
...
Figure 2.8.
Evolution of a contextual graph in example “coffee prepa
...
Figure 2.9.
Tree representation of the ESIA in .
...
Figure 2.10.
Tree-view building (Garcia and Brézillon ).
...
Figure 2.11.
Contextual graph (a) and tree (b) representation with t
...
Figure 2.12.
“Wall of mitosis slides” (MICO project).
...
Figure 2.13.
The three layers of the Context-Oriented Model.
...
Figure 2.14.
“Academic mission” example in a COM modeling.
...
Figure 2.15.
CxBR hierarchical organization for a particular mission
...
Chapter 3
Figure 3.1.
Group activity at the four modeling levels
Figure 3.2.
Model of the collaborative answer building processes. Fo
...
Figure 3.3.
Collaborative building of a proceduralized context. For
...
Figure 3.4.
Conceptual model of a turn.
...
Figure 3.5.
An operational representation in terms of turns of the p
...
Figure 3.6.
“Submission management” example as a contextual graph. F
...
Figure 3.7.
Example of a conceptual graph.
...
Figure 3.8.
Start of the CxG-based simulation of “Manage a unit” wit
...
Figure 3.9.
Instantiations leading to the realization of the indepen
...
Figure 3.10.
CxG-based simulation of the turn S5 that follows turn O
...
Figure 3.11.
Paper-submission example (Torres da Silva et al. ). For
...
Figure 3.12.
Petri-net representation of the submission-process exam
...
Figure 3.13.
UML representation of the submission-process example
Figure 3.14.
CxG representation of the “submission management” examp
...
Figure 3.15.
An extended model of the editor’s activity in the “subm
...
Figure 3.16.
Shared context between an editor (top) and a reviewer (
...
Figure 3.17.
Contextual graph (a), tree representations (b) and lege
...
Figure 3.18.
Modeling of the submission-management example as a grou
...
Figure 3.19.
Task realization “Giving an order of recognition” in Cx
...
Figure 3.20.
Contextual graph of activity “Manage a unit”.
...
Figure 3.21.
Movement between turns in the model.
...
Figure 3.22.
Model of operator–simulator interaction.
...
Chapter 4
Figure 4.1.
Key concepts and their evolution in the four-level frame
...
Figure 4.2.
Conceptual model of a CxG-based simulation of a mental-m
...
Figure 4.3.
The glocal approach of the subway-line responsible. For
...
Figure 4.4.
Flexibility inside workflow lifecycle (Fan et al. )
Figure 4.5.
Virtual slide of invasive ductal carcinoma with six posi
...
Figure 4.6.
Located mitosis on a C40 field of the virtual slide in .
...
Figure 4.7.
The glocal approach with a Command & Control system
...
Figure 4.8.
The global and local approaches in operator’s mental rep
...
Figure 4.9.
Different consequences of the contextual element “confid
...
Figure 4.10.
Part “the offer” of the contextual graph representing “
...
Figure 4.11.
Working context at the start of the simulation (left pa
...
Figure 4.12.
Intermediary results of the CxG-based simulation. For a
...
Figure 4.13.
Representation of the diagnosis page in the user manual
...
Figure 4.14.
The video problem-solving activity (Activity A2 in ). F
...
Figure 4.15.
Collaborative modeling of DVD-Reader diagnosis.
...
Figure 4.16.
Proposal of an architecture of CIASs.
...
Chapter 5
Figure 5.1.
One of the flow charts in Logan-Young and Hoffman (). Fo
...
Figure 5.2.
Introduction of activities in with two lower parts as a
...
Figure 5.3.
The complete representation as a contextual graph (with
...
Figure 5.4.
The contextual graph in with activities.
...
Figure 5.5.
Supermarket checkout task (Shepherd )
Figure 5.6.
CxG representation of the supermarket checkout task with
...
Figure 5.7.
CxG version of the supermarket checkout task in
.
Figure 5.8.
Different possible situations defined by the expert. For
...
Figure 5.9.
Example of the path followed among intermediary categori
...
Figure 5.10.
Evolution of a some driving errors.
...
Figure 5.11.
Examination processing in the ACP Department.
...
Figure 5.12.
Modeling of the reception workflow.
...
Figure 5.13.
Conformity checking and actions (activity 110 in ). For
...
Figure 5.14.
Management of a blocking nonconformity (activity 125 in
...
Figure 5.15.
Modeling of the registration workflow.
...
Figure 5.16.
Saussure’s signification system
Figure 5.17.
Recontextualization model for transfer from country A t
...
Figure 5.18.
The revisited Saussure’s signification system in
...
Figure 5.19.
Double role of contextual elements as signified and sig
...
Figure 5.20.
Contextual graph of the model given in .
...
Figure 5.21.
A representation of the CDR process and its variants
Guide
Cover Page
Table of Contents
Title Page
Copyright Page
Preface
Acronyms
Introduction
Begin Reading
Conclusion
References
Index
Other titles from iSTE in Information Systems, Web and Pervasive Computing
Wiley End User License Agreement
Pages
iii
iv
ix
x
xi
xii
xiii
xiv
xv
xvi
xvii
xviii
xix
xx
xxi
xxii
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
Context-based Modeling of Activity in Real-World Projects
Patrick Brézillon
First published 2026 in Great Britain and the United States by ISTE Ltd and John Wiley & Sons, Inc.
Apart from any fair dealing for the purposes of research or private study, or criticism or review, as permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced, stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers, or in the case of reprographic reproduction in accordance with the terms and licenses issued by the CLA. Enquiries concerning reproduction outside these terms should be sent to the publishers at the undermentioned address:
ISTE Ltd27-37 St George’s RoadLondon SW19 4EUUKwww.iste.co.uk
John Wiley & Sons, Inc.111 River StreetHoboken, NJ 07030USAwww.wiley.com
© ISTE Ltd 2026The rights of Patrick Brézillon to be identified as the author of this work have been asserted by him in accordance with the Copyright, Designs and Patents Act 1988.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s), contributor(s) or editor(s) and do not necessarily reflect the views of ISTE Group.
British Library Cataloguing-in-Publication DataA CIP record for this book is available from the British LibraryISBN 978-1-83669-095-5
The manufacturer’s authorized representative according to the EU General Product Safety Regulation is Wiley-VCH GmbH, Boschstr. 12, 69469 Weinheim, Germany, e-mail: [email protected].
Preface
Everyone knows the word “context”, hears it several times a day, and uses it for making decisions from the choice of a drink to rent for vacations or to move from one place to another. Context is an intrinsic part of our life, objectives, reasoning and actions, of the situation we are in and the local environment where we require resources. We are so steeped in this notion that context seems obvious to us, and we leave it implicit. However, everyone uses a personal definition of what context is, and even different definitions from one day to the next. We all have different (eventually contradictory) views on context. This situation mostly comes from the lack of operational use of context, except in a few domains like Natural Language Processing.
During our research on modeling context, we collected 271 definitions. It is a weakness for communication, especially interdisciplinary communication. The collection of definitions comes from approximately 30 domains. We organized a series of biannual international and interdisciplinary conferences on modeling and using context from CONTEXT-97 (Rio de Janeiro, Brazil) to CONTEXT-19 (Trento, Italy) for an interdisciplinary community of researchers wishing to share their views on context in order to develop a unified definition of context.
The research, which was led using AI on a wide range of applications, was beneficial for attacking modeling and the use of context because each application only puts an emphasis on some specific aspects of context, but results of the context modeling in a particular application may be directly applicable in other applications, often leading to highlighting in these applications some aspects of the domain initially not considered. At a more conceptual level, this research on context modeling intersects with most theoretical approaches to contextual reasoning, decision-making, diagnosis and problem-solving and also cognitive approaches such as activity, mental models and mental representations that do not always arrive at an operational expression. For example, in decision-making, the result is not the most important aspect, rather it is the process leading to this result, a process that is context-dependent.
This book presents the sum of 25 years of research on modeling, and using context in real-world applications, these applications coming from a very large spectrum of domains, such as Power Systems, Control and Command Systems, Psychology and Medicine. Ten of these applications are presented in this book for illustrating the aspects of context that are put center stage. Our approach covers conceptual, operational and implementation aspects as far as a software is implemented and used in all the applications presented (except the first application for Electricité de France, which served as a trigger initiating this research). Thus, the research was not limited to a sketch of what a model of context could be but to an operational and implemented model.
The research presents several original findings. Clearly, one cannot speak of context in an abstract way. This point explains the excessive number of definitions of context: most of these definitions are given without concern about real-world applications. The givers of definitions consider that it is sufficient to have a definition, even though the definition has nothing to do with the subject of their paper (like a justification for their claims). However, context cannot be modeled disconnected from the actors’ focus of attention, the activity, the situation in which the actor has to accomplish the activity, and the environment where the required means for that are. This does not mean that a model of context is specific to each application.
Indeed, context is understood through its relationships with knowledge and reasoning in real-world applications; that is, context, knowledge and reasoning must be modeled in a uniform representation formalism based on a bottom-up approach. It is fascinating to see that the resulting pieces of software, the Contextual-Graphs (CxG) formalism, rely on a handful of items, which makes the software very easy to use (training of 15 min) by students in domains far from computer science and AI (e.g. psychology, medicine, lawyer). The 25 key points of the research identified throughout the presentation are summarized and discussed for a coherent view of the research as a whole. These key milestones rely on the framework with four levels for modeling an activity with the CxG formalism.
The second successful challenge is the extension of the CxG formalism from one actor realizing a task to a group of actors realizing an activity. The transition from one actor to N actors necessitates considering the mental representation of one actor (based on mental models) in a group mental representation (based on independent subtasks of actors). The solution proposed is the modeling of actors’ activities in the group activity in terms of independent subtasks. Thus, group-activity development could be represented as the assembling of a sequence of subtask execution, that is, a dynamic building of a group mental model from the mental representation constituted of independent subtasks of the actors. The evolution from a task realization (for one actor) to an activity (for N actors) is possible with the building of a shared context during group activity development to represent the exchanges of contextual information between actors developing the activity. In a kind of reverse engineering, the notion of activity has been used as a unifying concept in all the books.
Such an evolution of the conceptual approach is not a bifurcation point because – as a return on investment – the actor’s activity has benefited from a deep change in its modeling from an internal viewpoint (the task model) to an external viewpoint (a set of independent subtasks) in the group activity. This change at the conceptual level has left the CxG formalism unchanged on the implementation aspect. However, this transition enriches the modeling of the task realization/activity by making context management explicit in group-activity development, leading to deep simplifications in the modeling of artificial or living systems because the resulting framework covers the strong connections between context and group activity. At the conceptual level, context is a general, non-formalized concept, that is organized and structured at the operational level and is fully implemented for confrontation with the real world at the environment level where the application is. The concept of contextual element shows how a fuzzy concept like “contextual knowledge” becomes formally “pairs of contextual and recombination nodes”. The CxG formalism has a version CxG_1.0 for actor activity and a version CxG_2.0 for group activity of the CxG software.
The CxG formalism allows context-based modeling of activity in AI systems that rely on human-experience bases represented by contextual graphs. The bottom-up approach does not lead to discovery as deep learning, but a possibility would be to use the sum of the bases of experiences developed for different activities in a domain as a support for a glocal (global and local) learning in the domain that, indeed, would be a context-based learning. Some elements are identified in the book for designing and implementing context-based intelligent assistant systems. Beyond applications used during the modeling and use of context, the CxG software (and thus, all our research) has been validated by other researchers on their applications, either using context at least implicitly or not while clearly it was possible. Our validation concerned applications in breast cancer diagnosis, Hierarchical Task Analysis, self-evaluation of drivers, and a context-based workflow in an ACP Department and in semiotics.
Last but not least is my expression of gratitude to the numerous people (about 150) that participated in this successful adventure. In particular I would like to thank two special people, Juliette and Céline, who have been enthusiastic and active supporters with their participation in the organization of two CONTEXT conferences, on the use of the CxG formalism, one in their Master’s dissertation (for modeling the revocation of the Edict of Nantes as a contextual graph) and the other in their PhD thesis (see the ACA project for a context-based intelligent tutorial system for the self-evaluation of drivers’ behavior). Finally, I consider this book, paradoxically, as a new starting point from which this exciting adventure may be continued. I hope to collaborate with researchers in order to transmit this research to the community interested in the modeling and use of context and particularly the CxG software.
Patrick BRÉZILLON
Paris, France
November 2025
Acronyms
The list does not contain acronyms corresponding to applications used in this document, those acronyms only having local interest and definition.
ACA
Adequate Category Learning
ACP
Anatomical and Cytology Pathology
AI
Artificial Intelligence
BPEL
Business Process Execution Language
CBR
Case-Based Reasoning
CDR
Contextualization-Decontextualization-Recontextualization
CIAS
Context-based Intelligent Assistant System
CMB
Context-Mediated Behavior
COM
Context-Oriented Model
CxBR
Context-Based Reasoning
CxG
Contextual Graphs
CxG_1.0 version
Version for an activity realized by a unique actor
CxG_2.0 version
Version for an activity realized by a group of actors
DES
Discrete-Event Simulation
EHV
Extra High Voltage
ESIA
Executive Structure of Independent Activities
FlexMIm
French acronym for “Platform for processing and sharing medical images in a cloud environment”
GADGET
Guarding Automobile Drivers through Guidance Education and Technology
GNU
“
G
NU’s
N
ot
U
NIX
”
HTA
Hierarchical Task Analysis
IBD
Inflammatory Bowel Disease
KBS
Knowledge-Based System
ODE
Ordinary Differential Equations
OSSMOSE
Open Source Software for Medium Organization Enterprises
SART
French acronym for support systems of traffic regulation
SEPT
French acronym for Monitoring of equipment in extra high voltage substation
SWF
Scientific Workflow
TACTIC
French acronym for “Contextualized support on a tablet for the transmission of command information”
XML
eXtensible Markup Language
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 calcium metabolism (Brézillon 1983). In brief, the idea is to identify the concepts of interest for activity modeling, that are refined and organized in a mental model that the actor has in mind when accomplishing the activity. The next step is to find an efficient implementation of the mental model confronting it with the observable phenomena that happens in the environment of the activity.
The spectrum of domains concerned by our research is large, from formal to less formal. The retained projects aim to illustrate the aspects of context put center stage in the projects: SEPT (1986–1995), SART (1996–2002), FlexMIm (2012–2015), TACTIC (2013–2015), ACA (2005–2009), “Contextualizing scientific workflows” (2009–2011), MICO (2011–2014), “Prediction of the stuck of wine fermentations” (1997–1999), the OSSMOSE project (2009–2011) and the project “Computer-mediated collaborative work” (1997–1999). The 10 projects and applications are introduced in a unique frame of presentation in three parts: domain overview, contribution to the project (or application) and lessons learned for our research on context modeling and use. Additional information is given in the following chapters.
The SEPT project highlighted that context modeling cannot be replaced by ad hoc tools or artifacts. The SEPT expert system was designed from technical reports corresponding to a “theoretical model”, and it appeared rapidly that a knowledge configurator was mandatory to play the role of a “contextualizator”. The SART project (for the subway company in Paris) was the beginning of modeling context bridging a conceptual framework, an operational framework and an implementation framework, the latter containing the CxG_1.0 version of the Contextual-Graphs formalism. Procedures and practices show how context intervenes in reasoning. The FlexMim project in Medicine was the opportunity to develop new concepts in group activity like a voting system for representing reasoning in a domain where science is more of an art than a science: experts generally reach the same conclusion by different interpretations of criteria or different combinations of criteria for diagnosing and glocalization for identifying cancer by first looking for global zones of interest on a slide and zooming-in on “spots” at the cell level for a local search. The TACTIC project was the opportunity to study the interaction of an operator, a simulator (cognitive interaction) and an interface (physical interaction). The consideration of the simulator with a similar status to the operator leads to the modeling of group activity in the CxG_2.0 version of the CxG formalism. In the ACA project, an experience base corresponds to all drivers’ practices developed for situation solving in different contexts. An experience base includes good and bad practices executed by novices, bad drivers or drivers under drug influence. Driver’s context awareness is sensible to any change of instantiation of contextual elements. Thus, actors’ behavior can be studied and compared by simulation of their activities to determine, say, those that develop safe or risky practices (and also for statistics). The objective of Scientific-Workflows (SWFs) is to use context for obtaining flexible SWFs. The use of the Contextualization–Decontextualization–Recontextualization (CDR) approach leads to the identification of the SWF model (decontextualization) and specifies how the SWF model is instantiated (recontextualization). It allowed the scientist’s activity to be extended in a contextual graph that guides the scientist in search of a variant of the initial SWF to solve their specific problem. The MICO project highlighted the interest of the tree representation that offers a “cognitive measure” of the diagnosis. More precisely, a mental tree representation offers fast decision-making by identifying mental models as context and situations are presented. Tree representation is an important component of the CxG formalism. The “Wine making” project is based on the onion metaphor for a temporal ordering of contextual elements around a focus. The alcoholic fermentation, being too complex for modeling, is considered a black box by enologists, but a relevant contextual element that can be found far from the focus of attention. The OSSMOSE project concerns the design and configuration of a contextualized platform for a given enterprise, taking into account the advice of the actors, the consultant and the customer, who have different expertise on tools to assemble on the platform. The customer’s expertise concerns the selection and configuration of tools to assemble on the contextualized platform to be used. Consultant’s expertise concerns experience base development and the way to assemble tools. The “Collaborative work” project shows the role of context in the cooperative building of an answer to a given question. Participants first gather contextual information for establishing the shared context and second for building the proceduralized context and preparing the answer.
Chapter 2 shows the initial concepts retained and those used for context modeling in actor activity to reach an operational implementation of the context-based representation formalism in CxG_1.0 version. Context modeling is presented in the framework with four levels, namely conceptual, operational, implementation and environment levels, as presented in Chapter 1.
The conceptual aspects concern the relationships of context with knowledge and reasoning. The notion of “contextual element” gives practical consistency to the concept of “context” as being constituted of objective elements (but possible subjective instantiations of them). Upstream, activity allows for the discrimination of context into contextual knowledge and external knowledge. Downstream, the activity concerns a specific actor that realizes a task in a given situation with the available resources of the local environment.
The selected concepts at the conceptual level are transposed in terms of the mental model at the operational level, showing the coherence between operational knowledge (knowledge effectively used in reasoning) and contextual knowledge in the framework of the CxG formalism (related to the activity) that leads to the importance of contextual reasoning. The implementation level helps to design and develop a contextual graph as a piece of software expressing mental models in projects and applications. The CxG-1.0 version of the CxG formalism is based on four functions but allows the representation of complex models by the combination and recursivity of the functions. In section 2.6, variants of the context modeling proposed here are presented, such as the context-oriented model attempting a programming approach, a parallel with BPEL used for workflow development, and a comparison with the two other context-based formalisms, CBR and CXBR.
Chapter 3 presents the first extension of the results presented in Chapter 2 for adapting the context-based modeling of an actor to an implementation for group activity. The presentation of four-modeling levels (conceptual, operational, implementation and environment) is used in the spirit of the previous chapter. The extensions include the notion of group activity and new notions such as independent subtasks, shared context, turn, CxG-based simulation and the CxG_2.0 version of the CxG formalism. Otherwise, the changes required for modeling, implementing and using context are an accommodation of the context-based modeling levels leading to the CxG representation formalism. Shared context is associated with group-activity development by elaboration from contextual elements of individual working contexts. Group activity is an assembling of independent subtasks realized by different actors. Reserved contextual elements make it possible to represent role and subtask management in a group activity. The notion of turn is introduced naturally as the realization of a subtask by a given actor by crossing the contextual graph and also includes the management of the turn transition and of the next actor (and its subtask). The CxG-based simulation of a sequence of turns requires a cyclic use of the contextual graph as long as the shared context is modified during the turn. The tree view of mental models is an alternative to the contextual graph to get directly to the decision to be made. Three examples chosen in our projects and applications illustrate the mechanisms implemented.
Chapter 4 presents the CxG_1.0 and CxG_2.0 versions of the CxG formalism. Section 4.2 discusses the KPs identified in Chapters 2 and 3, and a new definition (a dynamic one) of context. The role of contextual reasoning is naturally linked to activity accomplishment. An activity is considered on two levels, the operational level with the mental representation and mental-model development and the implementation level with the instantiation of the contextual elements and the deduction of the proceduralized context. Section 4.3, considers the findings of the “Internship-offer analysis” application. The mission was to model the activity as a contextual graph in the CxG_1.0 version; students had the same objective but personal goals. The same activity was later modeled in the CxG_2.0 version. The results are contrasted with those obtained on the modeling of “DVD-reader diagnosis” activity (also modeled by students). The explicit consideration of context in the modeling of an activity as a contextual graph leads to context-based intelligent assistant systems (CIASs) for reuse and an extension of the human experience in the spirit of Generative AI. CxG formalism provides powerful experience bases that can be served as a basis for deep learning in order to complete the known practices.
Chapter 5 presents a discussion on the use of the CxG formalism in five relatively experimental real-world applications where context was implicitly considered, namely breast cancer diagnosis, hierarchical task analysis (HTA), context-based training system for car drivers, workflow modeling in ACP Department and semiotics. The first example concerns a high-level summary about all that was known in the 1990s about breast cancer diagnosis. A context-based model is proposed in the CxG_1.0 formalism. The second example concerns HTA for conducting a task analysis. The idea was to breakdown the overall goal into the next level of subgoals triggered under plans introduced for a flexible approach. Experience was a complex combination of (contextual) elements for flexibility in plans. “Plans”, initially judged as comments, become integrated into activities and their nesting. The third example presents the ACA (Adequate Category Learning) project, which concerns the long-term objective to improve bad drivers’ abilities to detect a pre-critical situation and assess the potential risk of having an accident. The fourth example presents a contextualization of the Anatomical and Cytology Pathology (ACP) workflow for analyzing non-conformity during the reception and registration steps of the department. Non-conformity may be prejudicial to the patient. The semiotic approach concerned the transfer of the Walt Disney Company made at internationalization with the same international strategy. We revisited this study in terms of context-based approaches of the sociocultural environments of the source (USA) and the targets (Japan and France) and finally replaced this study in the CDR paradigm. A parallel is drawn between the concepts of sign, signifier and signified (in semiotics) and contextual elements and instantiations in the CxG formalism.
We want to point out right now for Chapters 1 and 2 that context is defined as a set of contextual elements. The real definition is introduced more formally in Chapter 3; the operational definition and the consequences are detailed in the last Chapters 4 and 5. Finally, we propose an extended evolutive definition of context that is valid for any activity and also at any step of the development of an activity.
Note
1.
The list of the publications (and PDFs for most of them) for most of the projects and applications (including the software) can be obtained from the author (Facebook, LinkedIn, email).
1Theoretical and Pragmatic Basis of Our Context Modeling
1.1. Introduction
Section 1.2 pinpoints the theoretical basis of our approach. First, we return to the conceptual difference between a model and representation. On the one hand, a model is an abstraction of the reality with the power of prediction; and, on the other hand, a model has as many different expressions as representation formalisms. Moreover, the concept “model” has different meanings in different disciplines. We also describe the concepts of activity, mental model and mental representation in Cognitive Sciences in light of the Contextual-Graphs (CxG) formalism for contextual reasoning and conversely, the contribution of the mental-model concept to the CxG formalism. We end this section with the proposed scientific approach and an overview of the origin of the four-modeling level framework.
Section 1.3 briefly presents several real-world applications over a period of 25 years in a very large spectrum of domains: Power Systems, Subway Exploitation, Enology with winemaking, Computer Security, collaborative understanding, car driver support, website analysis, self-training of car drivers, medical image access, contextualization of platforms of interoperable open-source tools for enterprises, database administration, a strategic infrastructure project in transportation, a cognition-driven explorer for histopathology in breast cancer grading, process and sharing of medical images, and contextualized support on a tablet for a Control & Command system for the French military. The ten retained projects used in this book for showing the aspects of context put center stage are: SEPT (1986–1995), SART (1996–2002), FlexMIm (2012–2015), TACTIC (2013–2015), ACA (2005–2009), “Contextualizing scientific workflows” (2009–2011), MICO (2011–2014), “Prediction of stuck wine fermentations” (1997–1999), the OSSMOSE project (2009–2011) and the project “Computer-mediated collaborative work” (1997–1999). All the projects have a common presentation: domain, our contribution to the project, and the lessons learned for our research on context modeling and use.
Section 1.4 sums up the main points covered in this chapter.
1.2. Theoretical basis
The concept of “model” often has different meanings from one discipline to another. After a presentation of our view on models, we point out that a model can have different representations, and their difference must be made explicit. We illustrate our points using the theoretical example of the autocatalytic model and a formal modeling of calcium metabolism with important findings in the scientific approach we followed to illustrate the richness of such an approach in the elaboration of our representation formalism.
1.2.1. Model and representation
Consider a general idea and several examples that illustrate the idea (MUSTIL Team 1996). There are two opposed viewpoints:
In the first viewpoint, the general idea is a model of the examples. It is the viewpoint in applied mathematics (Physics, Economy, Biology, etc.) where the mathematical model is an abstract representation of a tangible reality to provide a better understanding of some aspects of the reality or to better enable action on it. The main use of the mathematical model is the simulation of a real process or system.
In the second viewpoint, examples are models of the general idea. This is the viewpoint in Logics. Models are concrete objects that illustrate or interpret abstracted ideas generally expressed as logical formulas or axioms. Abstracted ideas constitute a theory, and a model (an example) makes a transition from the abstract to the real.
To sum up, a model in applied mathematics is a tool to act on the reality, whereas a theory in logic is a tool to study abstract properties independently of the real objects that can satisfy this theory.
A model is either a model of decision-making or a model of simulation. In decision-making, a model represents a part of reality within the framework of a given theory, and its sensitivity is studied on datasets (Lévine and Pomerol 1995). In biology, the model is a tool for confronting our ideas and assumptions with reality. It has qualitative aspects expressed as a statement and quantitative aspects represented by the specifications associated with the statement (Brézillon 1983).
Our research is ascribed in the first viewpoint. A model is a simplified version of the “real world” containing only the characteristics judged important by an observer. The model is expressed in a representation formalism that plays a crucial role of “concept revealer” between the model and reality. If we do not have the right representation formalism, we may miss some observations in the reality and/or have a model that is not relevant for testing our assumptions about the reality. For instance, in biology, a decay curve of a radioactive element given by discrete measures (observations) is supposed to be representable by the sum of exponentials as a representation formalism (another representation formalism is a set of ordinary differential equations (ODEs)), supposing that differences between the curve and the experimental data could be explainable by measurement accuracy and biological variability. For example, the concept of homeostasis expresses the paradigm of the “constance du milieu intérieur” of Claude Bernard (1865). As a result, a metabolism, like calcium metabolism, was assimilated to a linear system. However, Perault-Staub et al. (1975) highlighted the existence of circadian rhythms that could not be explained by a linear model and required a representation of the calcium metabolism as a nonlinear self-oscillating model. A new set of assumptions was necessary at the level of calcium-hydroxyapatite crystal formation in bone (observations) that could be explained by the role of dissipative structures pointed out in Chemistry (Goldbetter and Lefever 1972). Then, the choice of a nonlinear representation formalism was natural.
A model allows interpolation and extrapolation of a possible evolution of the real-world system beyond the observed behavior, thanks to the predictive power of the model in the limits of the chosen representation formalism (oscillations will never appear in a linear formalism like the sum of exponentials). A model can be expressed in different representation formalisms, each formalism allowing the model to exhibit (or not) targeted behaviors. Thus, the observer must define clearly which characteristics of the model are relevant for the study. However, it is possible to jointly use two different formalisms, like ODEs for a mathematician and a compartmental formalism for a physiologist (Jacquez 1972). The “Paper submission” example (see section 3.6.1) shows that the example has been used for several formalisms of representation: production rules, UML, Bayesian networks, Petri nets, fluence graphs, etc. A model has some properties that are important for a nonlinear model: the steady state – dQ/dt = 0 – and its stability or instability. A nonlinear model may present several steady states, and the choice of the initial conditions is thus important because there are several distinct stability zones. Usually the stability of a steady state is computed by the discriminant of the Jacobian matrix, which is composed of parameters in the model (transfer constants Kji from compartment i to compartment j below) and of the values of the variables at the steady state. Note that when the steady state is unstable, the stability zone generally corresponds to a limit cycle (in natural systems with positive masses there is always a stability zone). There are other subtasks related to parameters and initial conditions such as identifiability, identification and sensibility. However, nonlinear models often have no theoretical solution, only numerical ones, and thus, the solution is obtained by numerical simulation.
Figure 1.1 presents the example of the autocatalytic model (Goldbetter and Lefever 1972) that has the following expression in (a) a nonlinear differential equation representation formalism and (b) the compartmental representation formalism (Jacquez 1972).
Figure 1.1.Representations of the autocatalytic model by (a) differential equations and (b) compartmental formalism
The steady state is:
The conditions of stability of the steady state in the example are given by:
When the steady state is unstable, the stable trajectory is a limit cycle in the state space (Figure 1.2a) and variables present oscillations as a function of time (Figure 1.2b).
Figure 1.2.Limit cycle in the phase plane (a) and oscillations in a temporal representation (b)
We used this approach in the modeling of the calcium metabolism (Brézillon 1983). The concept of homeostasis of the calcium metabolism was considered as a statement relative to the state of a variable y (the calcemia). The concept of homeostasis conventionally corresponds to maintaining the variable y at a constant value y0 (dy/dt = 0). It is built on concepts such as mass, time, error, stationarity, etc., and principles such as regulation and small variations.
In the SART project, we have changed the representation, initially for pragmatic reasons of scarcity. We started from the notion of a decision tree (Pomerol et al. 2002) and its development in the SART project as summed up in (Pasquier 2002). In a decision tree, each event at an event node carries on a part of the uncertainty of the situation. There are no probabilities for the events, but there are several contexts in which a decision has to be made. For example, in Figure 1.3, which represents the incident “lack of train power” in the SART project, the branch with {C12, C21, C32} describes a context in which there is no immediate repair possible but still enough power and a steep incline in view. For this reason, we replaced the event nodes used in decision-making with context nodes, and states of nature are replaced by contexts. There is no interest in the probability of, say, C12 versus C11, because, first, these instantiations are exclusive, and, second, our representation intends to be used for any incident, whatever the probabilities are. In this sense, the Ci are not random variables unlikely in influence diagrams. When an incident occurs, the probability of occurrence no longer matters; what matters is the chaining of troubleshooting actions.
Figure 1.3.Decision tree of the procedure for “lack of train power” incident.
The postponement of actions to the leaves observed in Figures 1.3 amounts to relating each decision to a state of nature, here, the context described in the branch. Note that a branch corresponds to a contextual node that is instantiated to a given value. For example, the branches C11 and C12 are more important than the contextual node C1 that will play a central role in our CxG formalism. Thus, our representation tends to stick to operators’ behavior. This means that when operators undertake an action in the tree, they consider that, due to the contextual information they got, the state of nature between the root and the action undertaken is the true state of nature.
The tree representation was made of two types of elements: actions and contextual nodes, which select a branch depending on the knowledge about the current context. Figure 1.3 shows the decision-tree representation of the procedure for “train lack of power” solving (the meaning of the boxes is not important here but can be found in Pasquier (2002)). In summary, our representation is inspired by decision trees but mainly differs on two points: (1) our trees have no chance nodes but have contextual nodes and (2) there are no probabilities.
This representation shows several important characteristics that have some consequences on the size and structure of a tree:
Operators use many contextual elements to perform their choice. This leads to a large number of pragmatic strategies, even for the same incident, and multiplies the number of branches and the tree size grows rapidly.
Operators prefer to gather a maximum of information before making their decision. This behavior postpones most of actions to the end of tree branches. It is an observation similar to Watson and Perera’s observation (1998) that a hierarchical case representation holds more general contextual features at its top and specific building elements in its leaves.
Operators privilege actions allowing us to reach common intermediary situations. Thus, they can reuse common strategies to clear the incident. Graphically, the tip sequence of actions is often repeated from one branch to another.
Several action sequences are executed in the same order in different situations (paths).
Two actions could be done in any order but must precede a given third one. For example, before linking two trains, both have to be emptied, but the order in which they are emptied does not matter (partial order on the actions as in the onion metaphor).
The large expansion of the tree structure does not easily make it possible to represent highly contextual decision-making in complex applications. We now explain the modification we have done, based on the characteristics discussed above, to obtain a manageable structure for representing operational knowledge on incident solving for subway lines.
The first operation was to reduce the number of objects in the structure by replacing repeated sub-sequences of actions with a single object called macro-action. The choice made for defining the different macro-actions is based on common sub-procedures known by the operators, such as “linking trains” and “return to end-station without travelers”. The principle of replacing is explained in Figure 1.4.
Figure 1.4.From a sequence of actions to a macro-action.
This replacement simplifies the lecture of the tree but does not reduce structural expansion of the tree. The second operation was to merge tree branches as soon as the sequence leading to the end of the incident are similar, as shown in Figure 1.5.
Figure 1.5.From tree to graph.
Cognitively speaking, all of this to use a scarcity principle leads operators to reuse well-known procedures as soon as possible. This operation has several consequences on the representation structure and the model meaning:
We no longer face a tree but an acyclic and directed graph with exactly one root and one goal, and branches express only different strategies, depending on context, to achieve this goal.
Structure size is now under control, and consideration of a new contextual element will add some elements in the graph but will not drastically increase its size.
Structure change introduces a life duration of the instantiation of contextual nodes. When two branches are merged at a recombination node, the contextual node is initially proceduralized, and their instantiations do not matter anymore.
Even in part of a subgraph, it may happen that actions are only partially ordered. In Figure 1.6, all the branches join together in one node, according to an implementation simplification rule (the blue arrows now correspond to an Executive Structure of Independent Activities (ESIA)). As a result, the C1.1 branch of Figure 1.6 is connected directly to the branch C3.1, not at the final node because it connects to the macro-action MA1 for avoiding a duplicate. We observe a similar situation at the level of the contextual element 6. As a result, arriving at the final node, the trace of the development is lost. The same problem appears for mental models through the macro-action MA4. The representation respects syntax but not semantics: the macro-action MA1 exists in two different contexts (C1.1 and C3.1).
Figure 1.6.The contextual graph for “lack of train power” incident in Figure 1.3.
Thus, a simplification at the implementation level for avoiding double occurrences implies an inconsistency at the conceptual level. Coherence between conceptual and operational levels is lost. Moreover, updating the knowledge representation (conceptual level) is difficult because knowledge organization is at the implementation level. At the time of expert systems, screening clauses for representing procedural knowledge in a declarative formalism had the same effect.
Changes introduced in the CxG formalism take into consideration the identified weaknesses:
A distinction between contextual node and instantiation and an association of similar concepts at the operational and the implementation levels of modeling (mental representation and contextual graph, mental-model development and path in the contextual graph).
A contextual element is a pair of contextual and recombination nodes with as many exclusive branches between nodes and values. A contextual node must be associated with a unique recombination node.
Branches between a contextual node and a recombination node represent variants of the reasoning step that is the contextual element, and there are as many variants as branches that differ only on a defined part.
An action or a macro-action can occur in different contexts (like MA1).
If a contextual node C2 is on a branch of a contextual element C1, the recombination node of C2 must be before the recombination node of C1. As a result, the contextual element C2 is embedded on a branch of the contextual element C1. Thus, a contextual element includes another contextual element, as with Russian dolls. There is no partial overlapping.
Figure 1.7 is the rewriting of Figure 1.6 in the CxG_1.0 version (explained later). The blue circles represent the pairs of contextual and recombination nodes, green square boxes represent the actions, the violet boxes represent the macro-actions and the two red bars the ESIA.
Figure 1.7.CxG version of the diagnosis of “lack of train power” incident.
