Project TEAL: Add Adaptive e-Learning to your

Workflows

Oleg Rostanin, Carsten Ullrich, Harald Holz

(German Research Center for Artificial Intelligence, Germany
[email protected], [email protected], [email protected])

Shenwei Song
(Imbus AG, Germany
[email protected])

Abstract: Workflow- or task- embedded e-learning is an actual trend in enterprise

and office environments. Having been integrated into enterprise workflow or task man-
agement systems, e-learning turns into a powerful tool for enterprise knowledge man-
agement: the seamless integration into the working environment allows getting actual
information about potential learning goals of the user; using up-to-date e-learning tech-
nologies enables just-in-time delivery of goal-oriented, user-tailored learning curricula,
helping employees to solve problems autonomously and competently. This paper re-
ports about the results of the project TEAL (Task-embedded adaptive e-Learning)
taking place at German Research Center for Artificial Intelligence (DFKI). The aim of
the project was to prove the concepts and feasibility of task-embedded e-learning by de-
signing a reference architecture and realizing prototypical implementation based on the
existing components built at DFKI: WFMS Taskman, LCMS DaMiT and ActiveMath.
Key Words: task embedded learning, workflow learning, ontology, business process
oriented knowledge management, business process management
Category: H.3.7, H.5.4

1 Introduction

Modern enterprises face hard restrictions and requirements than it was a decade
ago due to the progress of globalization and world economy developments
[UNDP 03]. Ubiquitous computing and emerging internet technologies lead to
more and more complex computational environments that increase the danger of
the information overload in the enterprise that present a challenge for enterprise
employees. In order to achieve and keep the competitive edge, todays enterprises
have on one hand to carefully adjust their business processes to the permanently
changing enterprise business goals [Sheer 05]. On the other hand the enterprise
has to care about the effectiveness of each employee that depends heavily on the
employees‘ skills, competency level and employees‘ ability to effectively learn in
the situation of information overload.
While the problem of business process optimization in the enterprise can
be solved by introducing workflow management systems (WFMS) and business
process management (BPM) software [Ultimus 05], the problem of information
overload and employee skill/competency support can be solved by introducing
knowledge management and e-learning solutions [Prusak 01]. The integration of
the above approaches (i.e., workflow management, knowledge management and
e-learning), named “Workflow learning“TM, is a unique combination that allows
to increase the effect of both. Having been integrated into enterprise workflow
or task management systems, e-learning turns into a powerful tool for enter-
prise knowledge management: firstly, the seamless integration of e-learning into
the working environment (working context) allows to get actual information
about potential learning goals of the user and adjust the learning process to
the current user needs; secondly, the integration of learning into the workplace
increases the acceptance of learning because the learning process in not seen
as extra burden and is a part of daily work. In [Atkins 03], Sam Atkins gives
the definition of “Workflow Learning“TM that assumes integration of enterprise
software and e-learning into one workflow using service oriented architecture
(SOA), generating learning tasks as part of the enterprise workflow and per-
manent control on the learning task execution. Although we agree that it is
a possible approach for solving the problem, we advocate the approach of the
lightweight proactive information delivery and business process-oriented knowl-
edge management described in [Holz et al 05]. In the TEAL project, we extend
the idea of context-specific, proactive information delivery by using up-to-date e-
learning technologies enabling just-in-time delivery of goal-oriented, user-tailored
learning curricula, helping employee to solve problems autonomously and com-
petently [Rostanin and Holz 05]. This paper reports about the results of the
project TEAL taking place at German Research Center for Artificial Intelli-
gence (DFKI). The aim of the project was to prove the concept and feasibility of
task-oriented proactive delivery of e-learning courses by designing a reference ar-
chitecture and realizing a prototypical implementation based on the components
built at DFKI: WFMS Taskman [Taskman], LCMS DaMiT [DaMiT], LCMS
ActiveMath [LeActiveMath].

2 Task-embedded e-learning

As it was said above, the enterprise business processes set hard requirements
to the task-embedded e-learning. The information delivered to workflow partic-
ipants has to satisfy the current users’ information needs (be just-in-time) but
not to overwhelm them (be just-enough). Considering this facts, we argue that
task-embedded learning has to be highly adaptive and goal-oriented (see 2.1).

2.1 Goal-Oriented Learning

Fundamental research on goal-oriented learning in general, and, in particular,
with respect to AI and machine learning has been conducted
by [Ram and Leake 85]. Goal-oriented or targeted learning is something that
one meets in everyday life. Using AI terminology, let us call the person who
deals with any kind of tasks an agent. Such agents often have to cope with pre-
viously unknown kinds of task. Facing a new task, an agent first makes mental
inferences and checks whether he has enough knowledge to perform the task. If
the inference fails (i.e., necessary knowledge is not present), a knowledge gap
is identified. After the knowledge gap is identified, the learning goal “cover the
knowledge gap in the context of the given task“ arises. Learning goals that arose
earlier and had not been satisfied or have been satisfied only partially will in-
fluence the learning process even if the agent currently has a new learning goal.
Moreover, learning goals tend to evolve during the learning process: they can be
generalized, specialized, and - if they become obsolete - eliminated.

2.2 The TEAL Architecture

One of the main tasks in the TEAL project was to develop a flexible architecture
that allows for just-in-time learning in workflow environment (see fig. 1). The
components comprising the architecture are presented below:

WFMS A WFMS is “a system that defines, creates, and manages the execu-
tion of workflows through the use of software, running one or more workflow
engines, which are able to interpret the process definition, interact with work-
flow participants and, where required, invoke the use of IT tools and applica-
tions“ [Allen 05]. WFMS contain two basic components: the first component
is the process modeling (PM) component, which enables administrations and
analysts to define processes and activities, to analyze and simulate them, and
to assign them to people. The second component is the process execution
(PE) component, sometimes called the run-time system. Normally, work-
flow participants deal only with the PE component using a web-based or
desktop-based WFMS client. In the TEAL project we used the Taskman
system [Taskman] as the WFMS for the task-embedded learning.

LCMS LCMS is a multi-user software system designed to enable learning de-

velopers to create, store, reuse, manage, and deliver digital learning con-
tent from a central learning object repository (LO Repository). The LCMS
DaMiT used in TEAL is extended by incorporating an on-the-fly goal-
oriented course generator developed in the LeActiveMath project (see 2.4).

LO Repository Is a part of the LCMS that manages storing and retrieval of

learning objects (LO) as well as maintenance of the learning concept ontol-
ogy. The LO repository provides access to LOs through both a web service
interface (TEAL DB Service) and direct SQL queries to the database.
 Figure 1: Architecture of task-embedded e-learning

DyLeGo - Dynamic Learning Goal A SOA-based middleware system en-

abling task-embedded e-learning by implementing the following functionali-
ties accessible through a web service interface:
1. Building a text index of the learning concept ontology from the LO
repository for a high-performance search of learning concepts.
2. Retrieving learning concepts from the ontology based on the full text
search in the index;

3. Delivering filtered list of potential learning goals to the user based on

the current task context.
The working scenario of the presented system is depicted on fig. 2:
1. DyLeGo call: The software developer Bob is a participant of software de-
velopment workflow in the company C. At the moment when Bob creates a
task in the WFMS, the WFMS automatically launches a learning goal query
to DyLeGo. DyLeGo receives the context information from the WFMS and
starts with learning concept retrieval.
2. Learning concept retrieval and filtering: DyLeGo starts the learning
concept retrieval in the LO repository of the LCMS. If learning concepts
that are relevant to the task are found, they should be filtered using the
information from the workflow context. For example, if Bob is an expert in
SQL, learning content about basic concepts of SQL is out of interest for him.
3. Learning strategy identification: After retrieving and filtering of the
learning concept, learning strategy should be determined. Which learning
strategy in learning process should be chosen depends on the learning goal
itself, the role of the user and the learning history of the user. In the TEAL
project we defined 4 following types of learning strategies:
overview Very short description giving the general impression about the
subject to be learned. One can compare this with glossary description.
On the basis of the overview Bob should be able to judge whether he
needs to learn this subject deeper or not.

cursorily If the learner decided to learn the subject but he/she does not
need to get expert-level knowledge on it, the cursorily strategy should
be chosen. For instance, it would be the case if Bob’s manager would like
to get acquainted with possibilities of the SQL language.

detailed Provides expert-level knowledge on the subject. If Bob would like

to optimize a complex Oracle query and has no idea about optimization,
a detailed course on Oracle SQL tuning should be delivered for him.

repeat Serves as reference material on the subject. If Bob finished the

course on Oracle SQL tuning he might still need a succinct reminder
course on Oracle optimizer hints.
At the end, DyLeGo returns learning concepts and recommended learning
strategy as potential learning goals to the WFMS.

4. Potential learning goal visualization: Bob receives potential learning

goals delivered by DyLeGo and selects one or more of them. The LCMS
generates learning course dependent on the selected potential learning goals
(see 2.4) and delivers it to Bob, so that Bob can start learning.

2.3 Identification of Learning Goals in the Task Context

Task context includes a variety of information about the task environment that
allows identifying potential learning goals:

– Task information
• Task name, description, task-relevant concepts and documents
provide the key information about what the user is currently doing. Using
task name we can identify potential learning goals of the user.

• Reference to the instantiated task model if the current task is

instance of the certain activity model it can give more precise information
on what the user is currently doing than just using a task name.
 Figure 2: Task-embedded e-learning with DyLeGo: sequence diagram

• Project information, connection to other tasks The information

about the project and other tasks (predecessor, successor etc.) of the
user can help to interpret current user actions.

– User information
• User role is the base for determining learning strategy of the user.

• User skills, interests, working and learning profile allow us to

exclude concepts that user already knows or include concepts that are
unknown for the user.

The basis for the retrieval of potential learning goals is the ontology of
learning concepts contained in the LO repository of the integrated LCMS. The
learning concept ontology depicts the outline of the learning content in the LO
repository. The goal of the learning concept ontology is to model the domain
of knowledge related to the given workflow. LO contained in the LO repository
are bound to the corresponding concepts using special metadata. The more con-
cepts are preserved in the ontology, the finer knowledge gap identification can be
achieved. In order to enable potential learning goals search, the DyLeGo system
creates full text search index of the learning concept ontology (concept name,
description). After the request from the WFMS system arrives, the search of
learning concepts in the text index using task context information is performed.
The result of the search are filtered using learning history and competence profile
(delivered from the WFMS). The concepts left after filtering are returned to the
WFMS together with their short descriptions and proposed learning strategy.

2.4 Generating an Adaptive Goal-Oriented Course

In the LeActiveMath project, an advanced course generator [Ullrich 05a] was
developed that implements competency-oriented learning strategies based on
modern pedagogical research. The pedagogical knowledge is generic, that is,
independent of the content. While the developed learning strategies were origi-
nally target at learning mathematics, they can be applied on different content,
for instance the one of the TEAL project. This general applicability is achieved
in the following way: the pedagogical knowledge uses an vocabulary targeted at
describing learning resources from a pedagogical perspective that captures the in-
formation necessary to reason about resources from a pedagogical point of view.
This vocabulary is specified in an ontology of instructional objects [Ullrich 05b].
In order to connect a LO repository to the course generator, one has to specify
a mapping from the ontology to the metadata representation used in the repos-
itory and a wrapper that answers queries about the learning resources coming
from the course generator. In information processing, such an approach is called
a mediator architecture (for a recent overview, see [Doan et al 04]). The result
of the course generation is a structured sequence of learning resource identifiers
similar to an IMS Content Package [IMS 03], which can be further processed by
the system that called the course generator.
The integration into TEAL followed this approach: a mapping between the
ontology of instructional objects was specified and a wrapper connected the
mediator to the LO repository of the LCMS DaMiT. Each time the WFMS
Taskman needs to provide the employee with a sequence of learning objects, it
starts the course generator with the identifiers of the target learning concepts
and chosen strategy. As a result, the WFMS Taskman receives a sequence of LO
identifiers that are presented to the learner by the LCMS DaMiT.

3 Conclusions

In the current paper we reported about the results of the TEAL project:
1. We introduced a reference architecture for integrating e-learning into enter-
prise workflows.

2. We introduced the DyLeGo system serving for dynamic identification of

employee’s learning goals.
 3. We reported about integrating a dynamic course generator into the LCMS
DaMiT that enables just-in-time delivery of the adaptive learning courses in
the workflow context.

In the future we would like to check the DyLeGo concept on WFMS others than
Taskman. The second step would be to evaluate the system in industrial and
academic scenarios.

Acknowledgements
The TEAL project was funded by the Innovation Funding Rheinland-Palatinate.
The LeActiveMath project is funded under the 6th Framework Program of the
European Community.

References

[Allen 05] Allen, R., Open Image Systems Inc., United Kingdom Chair,
WfMC External Relations Committee Workflow: An Introduction. Url:
http://www.wfmc.org/information/Workflow-An Introduction.pdf
[Atkins 03] Atkins, S., Introduction to workflow learning. White paper. 2003. Url:
http://www.internettime.com/workflow/intro wfl.htm.
[DaMiT] Data Mining Tutor. Home page. Url: http://damit.dfki.de.
[Doan et al 04] Doan, A., Noy, N. F., Halevy, A. Y., Introduction to the special issue
on semantic integration. In SIGMOD Rec. Nr.4, V. 33, 2004.
[Holz et al 05] Holz, H., Maus, H., Bernardi, A., Rostanin, O., From Lightweight,
Proactive Information Delivery to Business Process-Oriented Knowledge Manage-
ment. Journal of Universal Knowledge Management. 2, 2005, Pages 101–127.
[IMS 03] IMS Global Learning Consortium, IMS Content Packaging Information
Model. June 2003.
[LeActiveMath] LeActiveMath. Home page. Url: http://www.leactivemath.org.
[Prusak 01] Prusak, L., Where did knowledge management come from? IBM Systems
journal, Vol 40, No. 4, 2001.
[Ram and Leake 85] Ram, A., Leake, D. B. Goal-Driven Learning. MIT
Press/Bradford Books, Cambridge, 1985.
[Rostanin and Holz 05] Rostanin, O., Holz, H., Task-Embedded e-Learning.
Url: http://137.251.109.40/dl2000/publicationdetail.php?&setchannelid=1
&publicationid=912
[Sheer 05] By IDS Sheer. The ARIS platform. Url: http://www.ids-
scheer.com/sixcms/media.php/2188/PR0905-E-BR.pdf
[Taskman] FRODO Taskman. Home page. Url: http://www.dfki.de/frodo/taskman.
[Ullrich 05a] Ullrich, C., Course Generation Based on HTN Planning. In Proceedings
of 13th Annual Workshop of the SIG Adaptivity and User Modeling in Interactive
Systems, 2005, Pages 74-79.
[Ullrich 05b] Ullrich, C., The learning-resource-type is dead, long live the learning-
resource-type! Journal of Learning Objects and Learning Designs, 1, 2005, Pages
7-15.
[Ultimus 05] Ultimus Inc. A Closer Look at BPM. Whitepaper.
Url:http://www.bitpipe.com/detail/RES/1103736098 704.html.
[UNDP 03] United Nations Development Program. MAK-
ING GLOBAL TRADE WORK FOR PEOPLE. 2003 Url:
http://www.undp.org/dpa/publications/globaltrade.pdf.