CAPTURING USER BEHAVIOR IN E-LEARNING ENVIRONMENTS

Enric Mor, Juli`a Minguill´on

Department of Computer Science, Multimedia and Telecommunication, Universitat Oberta de Catalunya, Barcelona, Spain

Francesc Santanach

Learning Technologies, Universitat Oberta de Catalunya, Barcelona, Spain

Keywords:

User tracking, user behavior, web usage mining, e-learning environments.

Abstract:

In this paper we describe a proposal for embedding hidden marks in the web pages of a complex web site

with the aim of simplifying the process of collecting and analyzing users’ behavior. Usually, server log ﬁles

are used for tracking user actions, but these ﬁles present typical drawbacks due to their huge size and the

lack of discrimination between user actions and automated responses. We propose two different strategies

for embedding usage marks into web server log ﬁles, action-oriented and content-oriented. Action marks are

button-oriented and keep track of user behavior oriented to perform tasks thorough system buttons and menus.

Content marks keep track of page loads and navigation thorough links pointing to contents that act as traps.

Both strategies are complementary and can be used at once. A simple experiment designed to capture user

behavior once the initial web site page has been loaded is described as a real example of such proposal usage,

showing the advantages of reducing the number of relevant log entries and focusing on speciﬁc user actions.

1 INTRODUCTION

One of the most interesting possibilities in any e-

learning environment is tracking user navigational be-

havior for analysis purposes, as it may help to dis-

cover unusual facts about the system itself, users’ be-

havior and also can provide useful information for

adaptation and personalization purposes. For exam-

ple, having a user navigational model (Kay and Lum,

2004) may be used to perform an automated usabil-

ity evaluation (Hilbert and Redmiles, 2000), detect-

ing whether the system web interface was properly

designed or not, and where users ﬁnd obstacles and

difﬁculties to reach out their objectives (Marsico and

Levialdi, 2004). This navigation model can be also

used to improve the student learning experience; nav-

igation behavior and learning indicators and can serve

to tailor learning itineraries to speciﬁc student needs.

In the case of a virtual learning environment, sev-

eral data and information levels can be determined

and thus, several analysis levels can be established,

and different research questions can be answered with

the aid of web mining tools. Instructional design-

ers, teachers but also system designers need powerful

tools for understanding all the information collected

in a virtual learning environment in order to improve

the learning process and, thus, user experience and

satisfaction, by means of an adaptive (Paramythis and

Loidl-Reisinger, 2004) environment: monitoring and

interpreting the activities of its users (Thomas et al.,

2003), inferring user requirements and preferences,

and acting upon the available knowledge on its users.

Therefore, the extraction of the real learning and

training patterns can be of great utility, among other

purposes, to determine the quality of the e-learning

environment design and to evaluate the concordance

degree among the usability requirements and the navi-

gation behavior of the users, but especially for person-

alization purposes, tailoring the learning plans to the

student background and needs (Mor and Minguill´on,

2004). The information extracted from behavior and

learning patterns must not be used in any way but

for learning and teaching purposes. The main issue

of user behavior analysis in large e-learning environ-

ments is the great amount of data available from dif-

ferent sources and the lack of speciﬁc data for analysis

purposes. Usually, as users interact with the learning

management system by means of a web browser, it is

464

Mor E., Minguillón J. and Santanach F. (2007).

CAPTURING USER BEHAVIOR IN E-LEARNING ENVIRONMENTS.

In Proceedings of the Third International Conference on Web Information Systems and Technologies - Society, e-Business and e-Government /

e-Learning, pages 464-469

DOI: 10.5220/0001288404640469

 SciTePress

possible to track all interactions by examining the log

ﬁles stored in the servers which support the learning

management system. This is a well known technique

known as web mining (Chi et al., 2000), but it suffers

from several drawbacks when complex (in both size

and internal structure) sites are evaluated. Instead of

analyzing raw log ﬁles directly from the servers, we

propose to establish a minimum set of marks in strate-

gic areas which is considered to be relevant for analy-

sis purposes, with the double aim of capturing users’

real activity and intentions while reducing the volume

of data needed to analyze them.

This paper is organized as follows: Section 2 de-

scribes the basic services and the interactions that

are interesting to capture from a pedagogical point of

view. In Section 3 we describe the proposed mark em-

bedding strategy for capturing these interactions com-

paring two different kinds of mark techniques. Sec-

tion 4 describes the proposed framework for data col-

lecting and analysis, and some preliminary results fol-

lowing a three level decomposition. Finally, the most

important conclusions of this paper and current and

future research lines are summarized in Section 5.

2 E-LEARNING ENVIRONMENTS

In the last years, learning management systems

(LMS) and learning content management systems

(LCMS) have become a common tool for teachers and

institutions willing to set up virtual learning environ-

ments. In fact, both LMS and LCMS pursue the same

goal, but focusing on the learning process or the learn-

ing contents, respectively, and they can be combined

altogether, as in the UOC proprietary e-learning envi-

ronment, the virtual campus. We are interested in cap-

turing users activity when interacting with the LMS,

that is, when accessing to the available services.

Figure 1: Hierarchical structure of navigation levels.

In the context of an e-learning environment, user

activity and behavior can be analyzed from a three-

level (Mor et al., 2006) point of view, as shown in

Figure 1. These three levels, namely short term, mid

term and long term, are based on the study of student

behavior during different time periods and, therefore,

different data is collected at each level. The short

term level, also known as session level, captures the

way users navigate with particular goals in mind. It

considers each student log into the LMS and the ac-

tions performed and the information accessed during

the log session. At this level, as important what users

do is how they do it, which pages they visit and how

they reach them. The mid term level is the academic

semester level and takes into account the single user

sessions in a continuous ﬂow during the course du-

ration. At this level the main goal is to determine the

navigational patterns followed by users but at a higher

scale than in the previous level, according to some ex-

isting teaching plan, in the case of an e-learning en-

vironment. The third and long term level is called

lifelong learning level and the main goal is to analyze

how students evolve from the beginning of a degree

until they successfully ﬁnish it.

Studying the user activity using this three level

approach allows teachers, instructional designers and

usability experts to obtain an accurate description and

provides a complete view of system usage and user

behavior, because at each level the user’s goals can be

learned and correlated with the data analysis results

(Spiliopoulou, 2000). In this work we are focused

on the collection, management and processing of web

server logs and how can be used within the three-level

framework in order to discover knowledge about user

activity and behavior, specially for the ﬁrst two levels.

2.1 The UOC Virtual Campus

The Universitat Oberta de Catalunya (UOC

, in En-

glish known as Open University of Catalonia) is a

completely online university which offers 19 ofﬁcial

degrees, several graduate programs and post-graduate

studies, and a doctoral degree, with almost 40000 stu-

dents and more than 1800 people including instruc-

tional designers, teachers, tutors, academic and tech-

nical staff, and so. The UOC virtual campus is an in-

tegrated web based e-learning environment which al-

lows usersto asynchronouslycommunicate with other

users using a mail system, and includes an agenda, a

news service, virtual classrooms, a digital library and

other e-learning related tools (Sangr`a, 2002). Each

subject has a virtual classroom with all the elements

and tools for the learning/teaching process.

2.2 System Architecture

The UOC virtual campus is built upon a complex

database server system which uses a hierarchical

structure of servers which deal with different kinds

http://www.uoc.edu

CAPTURING USER BEHAVIOR IN E-LEARNING ENVIRONMENTS

465

of user requests. There are up to 24 front-end servers

(depending on the server load at each moment), and

an automated load-balance system moves each user

ment, switching on and off the total number of front-

ends depending on system load. At peak hours, it is

not uncommon to achieve 3000 concurrent connec-

tions, while 1500 is the average number of concur-

rent users in ofﬁce hours. Brieﬂy, the virtual cam-

pus uses client-server web technology and common

interface to integrate a series of services and function-

alities. These functionalities include: access to on-

line educational materials, library resources and gen-

eral academic and cultural information; student man-

agement enquiries service; and interaction with pro-

fessors and other students through pre-deﬁned com-

munication channels (e.g., forums, activity spaces).

Services include: an email account; a collection of

virtual classrooms, where each one has several com-

munication spaces where students and teachers can

interact and share learning resources; a digital li-

brary which integrates all the digital and non-digital

contents into the virtual campus. When users navi-

gate through these services, they leave a track which

can be posteriorly analyzed for user modelling pur-

poses. Most of this information is collected by the

web servers in form of server log ﬁles, according to

the Apache Common Log Format (CLF). Following

the classical architecture of web information systems,

there are three different layers for handling all the re-

quests from the users of the web site. A ﬁrst layer,

the presentation layer, is handled by the web servers,

which generate access logs with all the interactions

(i.e. clicks) performed by users. Most of these in-

teractions are not only navigational but also involve

the use of some services and components, handled by

the business logics layer, which does also generate its

own program logs in different formats, according to

each application. The data model layer is handled by

a back-end database which stores all the data gener-

ated by the applications in the previous layer, which

can also be traced for analysis purposes (i.e. who did

the last change on a speciﬁc record). Figure 2 shows

this three-level architecture.

Web server

(presentation)

Business logics

(services and components)

Data model

(database)

Access logs Program logs

Data

Figure 2: E-learning environment data architecture.

2.3 Server Log Files

For each action a user performs in the virtual cam-

pus, one or more lines representing such action are

logged in several servers. Furthermore, depending on

the type of action, several servers might log the same

action but using different information. In this work

we have used mainly the log ﬁles from the Apache

servers which act as front-ends, once them have been

joined in a single ﬁle which is generated each day.

This ﬁle is ﬁrstly preprocessed in order to remove all

those log lines which are surely not hits produced by

the user, such as the load of icons, style sheets, ban-

ners, and so. This preprocessing reduces the amount

of lines in a 90%. Nevertheless, during a typical

week, the total number of lines that needs to be pro-

cessed is still about 36 million lines, approximately 12

GB, which is indeed a very large ﬁgure. Therefore, a

second preprocessing oriented towards narrowing the

experiment is required, following the recommenda-

tions described in (Cooley et al., 1999).

Users can be uniquely identiﬁed because there is

a unique session number generated each time a user

logs into the virtual campus using his or her user-

name and password. IP addresses are discarded be-

cause there is the possibility of many user accessing

through the same proxy server which might mask the

real IP address. Therefore, it is possible not only to

identify individual users but also each individual ses-

sion, which is useful to establish the different naviga-

tional levels described in the following section. Lines

without session number are removed. Server log ﬁles

are generated on-the-ﬂy by the 24 web servers, and

these ﬁles are copied onto other server which joins

them and applies the ﬁlters aforementioned above, in

a 24 hour basis routine, starting at 01:00am. A simple

analysis of connection patterns reveals that the low-

est server load is between 04:00am and 05:00am, so

consecutive log ﬁles are joined and split again at such

time. These daily (or weekly) chunks of preprocessed

log ﬁles are the starting point for analysis purposes.

3 MARKING STRATEGY

Our goal is to capture the real intentions of each user

navigating through the e-learning environment, but

focusing on those activities related to the learning pro-

cess (accessing the mailbox and virtual classrooms,

using learning resources, and so). From the starting

page, a set of marks is introduced in the most likely

paths followed by the learners, in order to validate

such paths and trying to discover any interesting fact

that might be relevant.

WEBIST 2007 - International Conference on Web Information Systems and Technologies

466

3.1 Action Based Marking Strategy

Each click a learner performs in a critic (i.e. impor-

tant to the learning process) element of a web page,

a call to an application in the business logics layer

is performed, and this generates one or more lines in

the web server log ﬁles. The simplest way to cap-

ture such click is adding a special parameter indicat-

ing the action performed by the learner, according

to the context where such element appears. There-

fore, adding the string "&ACTION=name" (where

name identiﬁes the action being performed by the

user) to each relevant link causes such string to be

logged and, therefore, easily traceable. For analysis

purposes, it is important to choose action names that

are not preﬁxes one of the other, simplifying the use

of “grep” or similar tools for inspecting the generated

log ﬁle. The main drawback of this strategy is that

it might interfere in the correct functioning of the ap-

plication programs handling each call. For example,

if the handling servlet or CGI has a security mecha-

nism for checking the receivedparameters, this mech-

anism should be modiﬁed in order to incorporate the

new parameter. This may be very expensive for com-

plex web sites where hundreds or even thousands of

services available through servelts or CGI are already

running.

3.2 Content Based Marking Strategy

The second strategy tries to hide “transparent” re-

sources in the web pages, which generate a speciﬁc

entry in the log ﬁle when such pages are loaded (par-

tially or completely, depending on the place where

they are embedded). This is typically achieved by

adding an invisible (transparent) tiny image in the

web page, unnoticeable by the user but that will gen-

erate a new entry in the server log ﬁle when loaded,

using a simple HTML sentence like the following:

<img style="display:none"

id="uocTrack_1_0" src="trackingTag.gif"/>

This is used to capture when a learner reaches a

web page considered relevant for the learning pro-

cess. This approach is also useful to generate a log

entry when a link in a web page is accessed. This is

done by forcing the invisible image to be loaded when

the link is clicked, as shown in the following HTML

sentences:

uocTrack_1_0.src=uocTrack_1.0.src + "?ACTION=" + action;

} </script>

<a href="http://www.uoc.edu/wiki"

onClick="tag2log(’WikiAccess’);">Wiki</a>

This strategy is more time consuming than the pre-

vious one, but it has a main advantage: it does not de-

pend on any particular call, it does not interfere with

any existing servlet or CGI, and it can be automated

in order to be incorporated in all the web pages con-

taining a particular resource or service that has been

selected to be logged.

3.3 Example

Both marking strategies can be used at the same time,

recording different kinds of user behavior. In com-

plex web sites where the same goal can be reached

from different paths, it is necessary to know not only

user actions but also other places visited as well, in

order to build a better user model for analysis pur-

poses. The following fragment of a preprocessed log

ﬁle shows marks from both types, action based and

content based. Notice that this ﬁle is simpler than the

equivalent Apache CLF one, as the IP address, times-

tamp and other ﬁelds have been replaced by user and

session identiﬁers and so.

166877 20060920183451 GET /cgi-bin/uocapp?s=166877 -

166877 20060920183508 GET

/trackingTag.gif?PAGE=/UOC/a/cgi-bin/hola

http://cv.uoc.edu/UOC/a/cgi-bin/hola?s=166877

166877 20060920183518 GET

/UOC/a/cgi-bin/ma_mainMailFS?s=166877&ACCIO=L_MNOUS

http://cv.uoc.edu/UOC/a/cgi-bin/hola?s=166877

...

166877 20060920183541 GET

/UOC/a/extcgi_0.html?s=166877&ACCIO=B_INICI&...

http://cv.uoc.edu/cgi-bin/uocapp?s=166877

166877 20060920183543 GET

/trackingTag.gif?PAGE=/UOC/a/cgi-bin/hola

http://cv.uoc.edu/UOC/a/cgi-bin/hola?s=166877

166877 20060920183555 GET

/webapps/classroom/032_common/plan.jsp?s=166877

&ACCIO=plan.jsp&...

http://cv.uoc.edu/webapps/classroom/032_common/

frameset.jsp?s=166877&...

In this case, a student with identiﬁer 166877 logs

in into the virtual campus, it loads the initial web

page is loaded (trackingTag.gif?PAGE), and

he or she accesses the mailbox (ACCIO=L_MNOUS).

Then, he or she goes back to the initial web page

(ACCIO=B_INICI), which is loaded again, and ﬁ-

nally goes to one of the virtual classrooms he or she

is enrolled to (ACCIO=plan.jsp).

3.4 Establishing a Methodology

One of the most important results of the embedding

mark proposal is that provides the web administra-

tors a simple but realistic map of the available ser-

vices in the e-learning environment and how they are

used by learners (or site users in general). Without

this map, any effort to understand user behavior be-

comes too expensive due to the necessity of identify-

ing the speciﬁc entries in log server ﬁles related to the

activity being monitored. Even trying to reproduce

previous experiments (Mor et al., 2006) for validat-

ing them with new available data becomes very difﬁ-

cult if the virtual environment has changed even only

CAPTURING USER BEHAVIOR IN E-LEARNING ENVIRONMENTS

467

slightly, which is not uncommon in complex web en-

vironments. Therefore, any embedding mark process

should be done according to the following require-

ments: (a) Architecture independent: all the changes

in the web site should be done in the presentation

layer only, and not in the business logics layer; (b)

Scalability: the proposed embedding mark strategy

should be the same independently of the size of the

web site being monitored; (c) Consistency: all the rel-

evant information needed for usage analysis is present

in the log ﬁles; (d) Hierarchical: in order to broad or

narrow each experiment carried out, trees and sub-

trees of marks should be activated and deactivated

easily. For example, “analyze only actions related to

the virtual classroom”; (e) Technology independent:

any change in the technological aspects of the busi-

ness logics layer should not interfere with the embed-

ded marking system; and (f) Automatization: a com-

plex web site with thousands of web pages should be

able to be modiﬁed by means of an automated script.

Studying user behavior in a web site needs to es-

tablish which elements and actions will be monitored

and further analyzed. In a web site with a wide range

of services and goals available to users, it will be nec-

essary to identify the main user tasks and the possible

user actions related to such tasks as well. Using the

cognitive walkthrough method (Wharton et al., 1994)

it is possible to obtain a ﬁrst version of the sequence

of steps that users perform in order to achieve a spe-

ciﬁc task. Then, starting from this sequence and the

knowledge of the services to be analyzed, a initial set

of marks can be established for generating a ﬁrst map

of the web site. This map is analyzed to discover any

gap due to an incomplete task description. This pro-

cess is repeated until a reasonable map is obtained.

4 EXPERIMENTAL RESULTS

We provide a simple experiment showing user behav-

ior once they just log in into the virtual campus. Our

goal is to determine the paths followed by the students

once they log in into the virtual campus, which is in-

teresting because the same goal can be reached from

different places, thus evaluating web design and the

relative importance of the available elements.

4.1 Data Set

The Fall 2006 academic semester started September

the 20th, 2006. This is a special day because most

users log in motivated by the fact that it is the ﬁrst day,

and they want to obtain information about the sub-

jects they have enrolled into, the available resources,

and so. The preprocessed log ﬁle for such day is 7,8

million lines long (around 1,5 GB), containing infor-

mation about105000 sessions, with almost 42000stu-

dent sessions from 20000 different students. We have

selected 30000 sessions of a subset of 14600 students

for this experiment, concretely those studying any of-

ﬁcial degree in Catalan. On the other hand, a simple

inspection reveals that some sessions show a bizarre

behavior, probably caused by a malfunctioning of the

sions with no interaction, which are removed, using a

threshold of 3 seconds, the minimum amount of time

for loading the initial page using a fast connection.

From the initial web page, students have access to:

(a) The mailbox, from three different buttons or links:

a message notiﬁcation with the amount of new mails,

an icon representing a ﬂag which means that there are

new messages waiting, and a main mailbox access;

(b) The list of subjects where the learner is enrolled

into; (c) The virtual classroom for each one of the sub-

jects the learner is enrolled into; (d) The notice boards

and forums for each one of the subjects the learner is

enrolled into; and (e) Other services available (digital

library, community, proﬁle, and so).

LIST 51.2 %

CLASSROOM

2.7 %

BOARDS

44.2 %

OTHER

14.9 %

39.3 %

EXIT

2.7 %

12.6 %

START

15.2 %

13.7 %

MAIL

1.7 %

40 %

12.9 %

40.3 %

1.6 %

12 %

11.8 %

20.9 %

6.1 %

4.1 %

2.9 %

31.4 %

11.1 %

36.3 %

8 %

7.9 %

8.7 %

1.8 %

23.6 %

9 %

47.3 %

1.7 %

25.7 %

7.7 %

2.8 %

34.7 %

25.6 %

Figure 3: Most frequent navigational paths.

Results show that 49,5% of the students directly

go to the notice boards and forums as their ﬁrst ac-

tion. This is normal in the ﬁrst day of an academic

semester, as students read the welcome messages and

some of them write short messages introducing them-

selves as well. In second position, a 26,4% of the stu-

dents go to one of the virtual classrooms. A 12,0% of

the students go to the mailbox by means of the mes-

sage notiﬁcation link, while only a 1,7% and 0,6% ac-

cess the mailbox through the main button and the ﬂag

icon respectively. A 8,1% of the students go to the

list of subjects, and the remaining 1,7% (which is a

remarkably low ﬁgure) perform other actions. Figure

3 shows the most frequent paths followed by learn-

ers between the different spaces in the virtual campus.

WEBIST 2007 - International Conference on Web Information Systems and Technologies

468

Only relevant links between nodes have been drawn.

A link is relevant when the percentage of students that

follow such link is at least a 5% for both nodes. No-

tice that this graph includes all the possible paths, that

is, not only the ﬁrst action taken from the initial page.

5 CONCLUSION

In this paper a proposal for embedding hidden marks

in web pages with the aim of user action identiﬁca-

tion has been proposed. Two complementary mark-

ing strategies have been described: the ﬁrst one, ac-

tion based, consists in adding special parameters in

the calls to applications in the business logics layer of

a web site, in order to capture the speciﬁc clicks in

the relevant elements for analysis purposes. The sec-

ond one, content based, consists in hiding invisible

pieces of content that, combined with simple scripts,

generate log entries for both user actions and navi-

gation paths. Preliminary results show that this ap-

proach is useful to face both well known problems

when analyzing large and complex web sites. First,

users’s intentions are better captured as only the rele-

vant actions (for analysis purposes) are logged. Sec-

ond, generated data are only a small fraction of the

original data stored in the web server log ﬁles, thus re-

ducing both time and space requirements for off-line

processing. A simple experiment showing user be-

havior the ﬁrst day of the academic semester has been

described. This experiment reveals interesting infor-

mation, for example, students start sessions reading

all the welcome messages from the teacher and other

students, and replying such messages. This fact could

be used to design a speciﬁc initial page for the ﬁrst

session during ﬁrst day, when all the other available

services are not requested. Similar ideas for improv-

ing the e-learning environment can be extracted from

the results of each activity analysis, depending on the

educational context.

Current and future research in this topic should in-

clude the use of large scale web usage mining tech-

niques for clustering purposes. Analyzing the evolu-

tion of users’ behavior along a period of time is also

an interesting issue. Finally, as web services are be-

coming more “semantic”, it will be important to es-

tablish the appropriate relationships between the new

services available in the virtual campus and real user

intentions when searching or browsing such services.

ACKNOWLEDGEMENTS

This work has been partially supported by a Span-

ish government grant under the project PERSONAL

(TIN2006-15107-C02-01).

REFERENCES

Chi, E. H., Pirolli, P., and Pitkow, J. (2000). The scent of

a site: a system for analyzing and predicting informa-

tion scent, usage, and usability of a web site. In Pro-

ceedings of the SIGCHI conference on Human factors

in computing systems, pages 161–168, New York, NY,

USA. ACM Press.

Cooley, R., Mobasher, B., and Srivastava, J. (1999). Data

preparation for mining world wide web browsing pat-

terns. Knowledge and Information Systems, 1(1):5–

32.

Hilbert, D. M. and Redmiles, D. F. (2000). Extracting us-

ability information from user interface events. ACM

Computing Surveys, 32(4):384–421.

Kay, J. and Lum, A. (2004). Creating user models from

web logs. In Proceedings of the Intelligent User Inter-

faces Workshop: Behavior-Based User Interface Cus-

tomization.

Marsico, M. D. and Levialdi, S. (2004). Evaluating web

sites: exploiting user’s expectations. International

Journal of Human-Computer Studies, 60(3):381–416.

Mor, E. and Minguill´on, J. (2004). E-learning personaliza-

tion based on itineraries and long-term navigational

behavior. In Proceedings of the Thirteenth Interna-

tional World Wide Web Conference, volume 2, pages

264–265, New York, NY, USA. ACM Press.

Mor, E., Minguill´on, J., and Carb´o, J. M. (2006). Data Min-

ing in E-learning, chapter Analysis of User Naviga-

tional Behavior for E-Learning Personalization, pages

227–246. WIT Press.

Paramythis, A. and Loidl-Reisinger, S. (2004). Adaptive

learning environments and e-learning standards. Elec-

tronic Journal of e-Learning, 2(2).

Sangr`a, A. (2002). A new learning model for the informa-

tion and knowledge society: The case of the UOC. In-

ternational Review of Research in Open and Distance

Learning, 2(2).

Spiliopoulou, M. (2000). Web usage mining for web site

evaluation. Communications of the ACM, 43(8):127–

134.

Thomas, R., Kennedy, G., Draper, S., Mancy, R., Crease,

M., Evans, H., and Gray, P. (2003). Generic usage

monitoring of programming students. In Proceedings

of the ASCILITE 2003 Conference, pages 715–719,

Adelaide, Australia.

Wharton, C., Rieman, J., Lewis, C., and Polson, P. (1994).

Usability inspection methods, chapter The cognitive

walkthrough method: a practitioner’s guide, pages

105–140. John Wiley & Sons, Inc.

CAPTURING USER BEHAVIOR IN E-LEARNING ENVIRONMENTS

469