Enhancing Online Discussion Forums with a Topic-driven Navigational

Paradigm

A Plugin for the Moodle Learning Management System

Damiano Distante

, Luigi Cerulo

, Aaron Visaggio

and Marco Leone

Unitelma Sapienza University, Rome, Italy

University of Sannio, Benevento, Italy

Keywords:

Discussion Forums, Navigability, Searchability, Information Search, Information Extraction, Text Mining,

Topic Modeling, e-Learning, Learning Management Systems.

Abstract:

One of the most popular means of asynchronous communication and most rich repository of user generated

information over the Internet is represented by online discussion forums. The capability of a forum to satisfy

users’ needs as an information source is mainly determined by its richness in information, but also by the way

its content (messages and message threads) is organized and made navigable and searchable. To ease con-

tent navigation and information search in online discussion forums we propose an approach that introduces in

them a complementary navigation structure which enables searching and navigating forum contents by topic

of discussion, thus enabling a topic-driven navigational paradigm. Discussion topics and hierarchical relations

between them are extracted from the forum textual content with a semi-automatic process, by applying Infor-

mation Retrieval techniques, speciﬁcally Topic Models and Formal Concept Analysis. Then, forum messages

and discussion threads are associated to discussion topics on a similarity score basis. In this paper we present

an implementation of our approach for the Moodle learning management system, opening to the application

of the approach to several real e-learning contexts. We also show with a case study that the new topic-driven

navigation structure improves information search tasks with respect to using Moodle standard full-text search.

1 INTRODUCTION

Online discussion forums represent one of the main

sources of user generated information (i.e., social me-

dia) over the Internet and enable asynchronous com-

munication between Internet users in the form of mes-

sage posts. Most visited websites, including blogs

and social networks, use forums to support user in-

teraction and knowledge sharing. In several domains

ranging from e-commerce (Otterbacher, 2008)(Gruen

et al., 2006), to news (Li et al., 2010), and healthcare

(Sudau et al., 2014) discussion forums constitute rich

and widely accessed repositories of information for

Internet users.

As an example, software developers forums are

an effective source of information where program-

mers search for and describe solutions to speciﬁc

problems

. In e-learning contexts, discussion fo-

An example of such forum is the Microsoft MSDN De-

veloper Network forum. http://social.msdn.microsoft.com/

Forums/en/categories/

rums enable asynchronous communication student-

to-student, and teacher-to-student, e.g., to sup-

port collaborative learning and group work (Stefan,

2009)(Hrastinski, 2008). Whatever the forum do-

main, discussions held in a certain period of time be-

come a source of information for any user accessing

the forum afterwards.

In general online forums organize messages into

a chronological order. A user starts a new discussion

by posting an initial message, other users post their

replies or comments to it, and the list of messages

form a discussion thread. If users are allowed to re-

ply to other users’ replies in additional to the original

message, discussions take the form of trees, with dis-

cussion branches.

The effectiveness of a discussion forum as infor-

mation source depends on its richness in information,

but especially on the searching paradigm users can

adopt to ﬁnd contents of their interest.

Search features usually provided with online dis-

cussion forums are limited to full-text search which

returns a list of forum messages that include (and/or

Distante D., Cerulo L., Visaggio A. and Leone M..

Enhancing Online Discussion Forums with a Topic-driven Navigational Paradigm - A Plugin for the Moodle Learning Management System.

DOI: 10.5220/0005078600970106

In Proceedings of the International Conference on Knowledge Discovery and Information Retrieval (KDIR-2014), pages 97-106

ISBN: 978-989-758-048-2

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

Topics-Terms

Matrix

Discussion

Topics

Lattice

discussion threads

Terms extraction

- Outlier ﬁltering

- Stopwords ﬁltering

- Terms stemming

Topic Modeling

Latent

Dirichlet Allocation

Documents

to Topics

Assignment

Formal Concept

Analysis

Forum

(ﬂat structure view)

Forum

(hierarchical topic-

centered view)

Documents-Terms

Matrix

Topics-Documents

Matrix

Forum

threads/messages

Forum

messages

Forum

messages

Figure 1: The topic-driven forum navigation enhancement process (Cerulo and Distante, 2013).

do not include) one or more of the query keywords

in their body and/or their title. Such a search feature

may return too many or too few results (depending on

the forum size and the query keywords) and may miss

messages which are semantically related to the query

keywords but do not actually include them (Baeza-

Yates and Ribeiro-Neto, 1999).

Hierarchical graphs constitute an effective

paradigm to represent users’ knowledge (Zhang

and Peck, 2003). In a previous work (Cerulo and

Distante, 2013) we have introduced an approach to

improve information retrieval and content navigation

in online discussion forums by introducing in them a

complementary hierarchical topic-driven navigation

structure. Information Retrieval (IR) techniques,

speciﬁcally Topic Models (Blei, 2011) and formal

concept analysis (FCA) (Ganter and Wille, 1999),

are used to discover discussion topics and hierar-

chical relations between them in the forum content.

Then, forum messages and discussion threads are

associated to discussion topics based on a similarity

score, thus to enable searching and navigating them

on a topic-driven basis, additional to conventional

chronological order and full-text search approaches.

In this paper we present an implementation of this

approach as a plugin for the Moodle learning man-

agement system which makes the topic-driven navi-

gation approach accessible and evaluable in several e-

learning contexts. We also present a case study which

provides a ﬁrst qualitative assessment of the beneﬁts

of topic-driven navigation and access to forum con-

tent, with respect to traditional full-text search.

The rest of the paper is organized as follows. Sec-

tion 2 describes our forum navigation enhancement

approach introduced earlier in (Cerulo and Distante,

2013). Section 3 presents the implementation of the

approach for the Moodle

learning management sys-

tem. Section 4 reports on a case study we have con-

www.moodle.org

ducted to qualitatively assess the beneﬁts of the topic-

driven forum enhancement approach in searching fo-

rums for information of interest for the user. Section

5 overviews related work and Section 6 draws conclu-

sions and introduces future works.

2 THE TOPIC-DRIVEN FORUM

NAVIGATION ENHANCEMENT

PROCESS

The topic-driven forum navigation enhancement pro-

cess, introduced recently by some of the authors of

this paper (Cerulo and Distante, 2013), is shown in

Figure 1. It consists of four main steps represented

in the ﬁgure as rectangles and described brieﬂy in the

following.

2.1 Terms Extraction

We represent a forum message as a vector of index-

ing terms, {t

, . . . , t

}, extracted, from the corpus of n

messages, {d

, . . . , d

}, through a standard text analy-

sis pipeline usually adopted in Information Retrieval

that comprises: outlier ﬁltering, stopwords ﬁltering,

and stemming (Baeza-Yates and Ribeiro-Neto, 1999).

The outcome of this step is a document-term ma-

trix DT, where each element {DT}

is the tf-idf of

the term t

in the forum message d

(Baeza-Yates and

Ribeiro-Neto, 1999).

2.2 Topic Modeling

Topic modeling, in particular Latent Dirichlet Alloca-

tion (LDA), is a statistical technique that is able to ex-

tract frequently co-occurring terms, known as topics,

from a corpus of documents (Blei, 2011). The input

is the document-term matrix, DT, obtained from the

KDIR2014-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval

Table 1: Examples of topic-term and topic-document matri-

ces.

topic topic-term topic-document

(top terms) d

problem, email, setup 0.6 0.7 0.1

problem, email, con-

nection, setup

0.3 0.1 0.1 0.5

problem 0.1

problem, video, de-

coder, setup

0.1 0.2 0.8 0.5

problem, video 0.1 0.2

previous task, while the output is a topic-document

matrix, TD, and a topic-term matrix, TT. The number

of topic k is a parameter that controls the granularity

of the topics and must be ﬁxed a priori.

Intuitively, the top terms of a topic are semanti-

cally related and represent some real-world concepts.

For example the concept related to problems e-mails

setup is represented by the terms “mail”, “problem”,

“setup”. The topic membership of a document de-

scribes which concepts are present in that document.

Table 1 shows and example of topic-document and

topic-term matrices.

2.3 Formal Concept Analysis

Using the topic membership of a term, we prune

a topic lattice by means of Formal Concept Analy-

sis (FCA). FCA is a computational way to derive a

concept hierarchy or formal ontology from a collec-

tion of objects and their properties (Ganter and Wille,

1999)(Birkhoff, 1967).

We model the topics as the objects of a formal

context and the terms as their attributes. The relation

R of the formal context is computed from the topic-

term matrix TT by means of a decision threshold h

i.e., a term (attribute) t

belongs to a topic (object) z

, t

) ∈ R iif {TT}

≥ h

As a clariﬁcation example consider the formal

context shown in Table 3 and the topic lattice ob-

tained from such a formal context shown in Figure

2. Topics are mapped on circles and hierarchical re-

lationships are represented by arcs. Large circles are

mapped on topics extracted with the topic modeling

approach, while small circles are intermediate topics

extracted with the formal concept analysis. The lat-

tice shows the hierarchical relationships between top-

ics. In the lattice the top most topic (z

) is the most

general topic. A path starting from the top most topic

is a more speciﬁc topic. For example z

is reachable

by the path from z

(problem), setup, z

(email), and

(connection), and represent the more speciﬁc topic

of problems related to the email connection setup.

Table 2: The formal context obtained from the topic-term

matrix shown in Table 1.

problem

connection

video

decoder

setup

× × ×

× × × ×

× × ×

Figure 2: The topic-lattice pruned from the formal context

shown in Table 3.

2.4 Documents to Topics Assignment

During this step each document (i.e. forum mes-

sage/thread) is mapped onto the topics to which it is

more likely to belong by estimating the probabilities

of each topic for that message (topic-document ma-

trix). For this purpose we adopted the topic-document

matrix TD and a decision threshold h

, i.e., a docu-

ment d

belongs to a topic z

iif {TD}

i j

≥ h

2.5 Parameter Setting and Accuracy

Evaluation

Selecting the number of topic k is one of the most

problematic modeling choice in topic models (Wal-

lach et al., 2009). We adopt a metric, introduced by

Meil

a (Meila, 2003) for clustering comparison, that

measures the Variation of Information as the entropies

the mutual information associated with cluster assign-

ments. Intuitively, the entropy measures the uncer-

tainty of allotting an item to a cluster, while the mu-

tual information measures the reduction of such un-

certainty when the allocation in the other cluster is

known. Following the approach adopted by Wallach

et al. (Wallach et al., 2009) the assignment of doc-

uments (forum messages or threads in our context)

EnhancingOnlineDiscussionForumswithaTopic-drivenNavigationalParadigm-APluginfortheMoodleLearning

ManagementSystem

to topics can be assimilated to a sort of cluster as-

signment. In our previous work (Cerulo and Distante,

2013) we showed that above a certain value of k no

signiﬁcant increment of Variation of Information can

be observed in a speciﬁc context. We consider such a

value the optimal number of topic in that context.

We evaluated the document assignment task to

check whether the documents assigned to topics by

the Latent Dirichlet Allocation were congruent with

the semantics of their content (Bakalov et al., 2012).

In our previous work (Cerulo and Distante, 2013) we

addressed this question with a controlled experiment

obtaining in average a precision ranging between 52%

and 74%.

3 TDForum: A PLUGIN FOR THE

MOODLE LEARNING

MANAGEMENT SYSTEM

Topic-Driven Forum (TDForum) is a Moodle plugin

(particularly, an activity module) that implements the

topic-driven forum navigation enhancement approach

described in Section 2 for the Moodle open-source

learning management system.

In Moodle, activity is a general name for a group

of features in a course. Usually an activity is some-

thing that a student will do that interacts with other

students and/or the teacher. Assignments, quizzes,

surveys, workshops, chats, and forums are examples

of activities that can be created in a course and that

are provided in Moodle by default. Each activity is

implemented by a software module (plugin) located

in the mod sub-folder of the Moodle instance. Addi-

tional activities can be included by installing the cor-

responding Moodle plugin

From a source code point of view, each Moodle

activity module consists of a series of mandatory ﬁles

(e.g., install.xml, lib.php, and view.php) used to in-

stall the module and integrate it within the Moodle

system, and other ﬁles speciﬁc to the plugin.

Figure 3 shows the architecture of the TDForum

Moodle plugin that we developed. In the ﬁgure, we

can distinguish the components representing the plu-

gin front-end (the graphical interfaces that Moodle

users interact with), and those that are part of the plu-

gin back-end.

The plugin front-end comprises the components

Main View and Discussion Topics View corresponding

to the two possible views on the forum content: (i)

A rich and up-to-date list of Moodle plugins can

be found in the Moodle Plugins Directory at http://

www.moodle.org/plugins

standard chronological list of discussions augmented

with discussion topics and scores, and (ii) navigable

hierarchical discussion topics graph. The last view

is built using the JavaScript InfoVis Toolkit

. It also

includes the Admin User Interface component which

lets administrators manage the forum data processing

and customize the visualization plugin parameters.

The plugin back-end contains the components im-

plementing the forum analysis and indexing process

described in Section 2 to build the additional topic-

driven navigation structure. In particular, the Process

Controller controls the process by executing the com-

mands provided through the plugin admin user inter-

face. It also exports forum content from the Moodle

database into a local temporary csv text ﬁle and im-

ports the data on the new navigation structure from

the local ﬁlesystem into the Moodle database.

The Data Processing component includes the fol-

lowing sub-components:

• Data Preprocessing: a Perl script which extracts

threads and messages from the csv ﬁle into sepa-

rated text ﬁles and performs terms extraction and

text ﬁltering such as stopwords and stemming (cf.

Section 2.1).

• Topics Identiﬁcation and Documents to Topics As-

signment: a R

script which uses the Topic Model

library

to perform discussion topics identiﬁca-

tion and documents to topics assignment. The ma-

trices Topics-Terms and Topics-Documents of the

detected forum discussion topics and scores asso-

ciated to them are generated in this step (cf. Sec-

tions 2.2 and 2.4).

• Formal Concept Analysis and Topics Graph Ex-

port: this component uses the FcaStone

Formal

Concept Analysis command-line utility to gener-

ate the lattice representative of the hierarchy of

topics and to export the topics graph used in the

graph view of the plugin (cf. Section 2.3).

The TDForum activity implemented by our plugin

offers the same features provided by a standard Moo-

dle forum (particularly, a main view which lists forum

discussions and messages organized in a chronologi-

cal order, the functionality of posting new messages

or replying to existing ones, full-text search of mes-

sages, etc.) and adds to them a discussion topics view

which acts as a topic-driven navigation index to the

forum content.

The main view (Fig. 4) presents the list of discus-

sion threads of the forum in a chronological order and

http://philogb.github.io/jit/

http://cran.r-project.org/

http://cran.r-project.org/web/packages/topicmodels/

http://fcastone.sourceforge.net/

KDIR2014-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval

100

Figure 3: Architecture of the TDForum Moodle plugin (with a gray background color, standard Moodle components).

adds to each of them the list of discussion topics in it

identiﬁed, and the calculated similarity score (column

’Discussion topics’ in the ﬁgure). Score values range

between 0 and 1 (with 1 representing the maximum

similarity value) and the list of topics associated to a

discussion is ordered by score. By right-clicking one

of the topics of the list, the user can search for discus-

sions or messages which are related to the selected

topic. The results of this search is presented sorted by

decreasing values of score.

The discussion topics view (Fig.5) shows the list

of discussion topics found by the analysis process for

the considered forum (scrollable list on the left side

of the ﬁgure) and a graph that the user can pan and

zoom which highlights the hierarchical relations be-

tween the identiﬁed topics. The user can navigate the

discussion topics graph or the topics list and once she

ﬁnds a topic of her interest she can retrieve the list of

discussions/messages associated to it with a click.

The plugin has been designed to extend a standard

Moodle forum and, at the same time, to be indepen-

dent from it. As such, if it is installed, applied on a

forum, and then deactivated, none of the content of

the original forum are lost, nor the additional mes-

sages/discussions that will have been added in it after

the plugin instantiation.

4 CASE STUDY

We evaluated qualitatively that, with the topic-driven

approach, searching and browsing tasks of forum con-

tents can be signiﬁcantly improved with respect to tra-

ditional full-text search. The case study has been con-

ducted on forums inside an instance of the Moodle

learning management system. The context is com-

posed by a reduced version of 2 forums extracted

from the Moodle user and development communities

(Table 3). The Installation Help Forum includes all

discussions about user difﬁculties with ﬁrst Moodle

installations or errors happening during the installa-

tion process, or with migration to different OSs, or to

newer Moodle versions. The General Help Forum in-

cludes discussions about problems not included into

other Moodle community forums, such as problems

with database access, ﬁle upload, block modules and

student enrollment.

We evaluated effectiveness in 11 searching tasks

in terms of (i) the number of items (forum posts) the

user had to inspect in order to satisfy the information

need, and (ii) the time spent to accomplish the task

(Table 4). The nature of the 11 searching tasks has

been deﬁned by the ﬁrst two authors of this paper.

For each task the search goal, i.e., the expected posts

EnhancingOnlineDiscussionForumswithaTopic-drivenNavigationalParadigm-APluginfortheMoodleLearning

ManagementSystem

101

Figure 4: The main view of TDForum showing the list of forum discussion threads enhanced with discussion topics and scores

associated to them.

Table 3: Case study context.

Forum # threads # posts # users time period

Installation Help 253 777 78 May 1, 2013 – May 24, 2013

General Help 115 714 107 Jul 15, 2013 – Aug 8, 2013

Table 4: Search tasks deﬁnition.

ID Search goal Adopted Keywords

1 Retrieve the 10 posts related to css problems in the General help forum css, problem

2 Retrieve the 5 posts related to login issues in the General help forum login, problem

3 Retrieve the 3 posts related to uploading ﬁles problems in the General help forum ﬁle, not, upload

4 Retrieve the 4 posts related to changing Moodle fonts in the General help forum change, font

5 Retrieve the 5 posts related to not sent enrollment email in the General help forum enrollment

6 Retrieve the 3 posts related to editing Moodle theme in the General help forum change, theme

7 Retrieve the 5 posts related to web hosting in the General help forum moodle, web, hosting

8 Retrieve the 10 posts related to Moodle upgrading problems in Installation help

forum

problem, moodle, upgrade

9 Retrieve the 5 posts related to editing admin password in the Installation help forum admin, password

10 Retrieve the unique post related to missing ﬁles after Moodle migration in the In-

stallation help forum

missing, ﬁles, after, migration

11 Retrieve the 2 post related to slower system after upgrade in the Installation help

forum

css, problem

that should be retrieved, is known beforehand. The

other two authors performed the searching tasks with

two complementary approaches: (i) full-text search,

and (ii) topic driven navigation. The ﬁrst approach is

accomplished with the default full-text search engine

implemented in the Moodle platform which performs

a full-text search from a set of user deﬁned keywords.

By default, the keywords are linked by an AND op-

erator and the system retrieves the list of all posts

containing all searched terms. The second approach

is executed with the TDForum Moodle plugin intro-

duced in Section 3. While performing the tasks, we

KDIR2014-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval

102

Figure 5: The Discussion Topics view of TDForum showing the topics list and the hierarchical topics graph.

collected the number of items inspected and the time

needed to achieve the search goal. With the Moo-

dle full-text search the number of inspected items is

computed by counting the number of posts examined

before ﬁnding the correct expected posts. With the

topic-driven navigation approach the number of in-

spected items is the sum of two quantities: the number

of links followed to reach the closest topic in the Dis-

cussion Topics View of the TDForum plugin and the

number of posts examined before ﬁnding the correct

expected posts.

Table 5 reports the results obtained by executing

the evaluation protocol on the 11 tasks. The table re-

ports for each search goal the number of inspected

items and the time spent to ﬁnd the correct posts. For

Moodle full-text search we also reported the number

of search attempts (queries) performed with differ-

ent search keywords necessary to reach the goal. In

general the number of items inspected with full-text

search is in average higher than the number of items

inspected with TDForum (14 vs 9). The time neces-

sary to obtain the correct answer is in average less in

TDForum (137 sec. vs 170 sec.) because with full-

text search more time is spent to choose the correct

search keywords. The difference is not statistically

signiﬁcant due to the limited number of samples, thus

further experiment are necessary to draw more gen-

eral conclusions.

Table 5: Case study results (time in seconds).

Task Moodle full-text search TDForum search

ID # queries # items time # items time

1 2 15 201 20 254

2 1 5 109 5 92

3 1 17 131 8 168

4 3 12 230 7 135

5 3 13 225 11 187

6 5 40 275 9 113

7 1 5 134 4 62

8 2 42 287 10 131

9 1 4 122 6 90

10 1 1 86 16 192

11 1 0 70 6 85

average 2 14 170 9 137

We evaluated qualitatively that, with the topic-

driven approach, searching and browsing tasks of fo-

rum contents can be signiﬁcantly improved with re-

spect to traditional full-text search. The case study

has been conducted on forums inside an instance of

the Moodle learning management system. The con-

text is composed by a reduced version of 2 forums ex-

tracted from the Moodle user and development com-

munities (Table 3). The Installation Help Forum in-

cludes all discussions about user difﬁculties with ﬁrst

Moodle installations or errors happening during the

installation process, or with migration to different

EnhancingOnlineDiscussionForumswithaTopic-drivenNavigationalParadigm-APluginfortheMoodleLearning

ManagementSystem

103

OSs, or to newer Moodle versions. The General Help

Forum includes discussions about problems not in-

cluded into other Moodle community forums, such

as problems with database access, ﬁle upload, block

modules and student enrollment.

We evaluated effectiveness in 11 searching tasks

in terms of (i) the number of items (forum posts) the

user had to inspect in order to satisfy the information

need, and (ii) the time spent to accomplish the task

(Table 4). The nature of the 11 searching tasks has

been deﬁned by the ﬁrst two authors of this paper.

For each task the search goal, i.e., the expected posts

that should be retrieved, is known beforehand. The

other two authors performed the searching tasks with

two complementary approaches: (i) full-text search,

and (ii) topic driven navigation. The ﬁrst approach is

accomplished with the default full-text search engine

implemented in the Moodle platform which performs

a full-text search from a set of user deﬁned keywords.

By default, the keywords are linked by an AND op-

erator and the system retrieves the list of all posts

containing all searched terms. The second approach

is executed with the TDForum Moodle plugin intro-

duced in Section 3. While performing the tasks, we

collected the number of items inspected and the time

needed to achieve the search goal. With the Moo-

dle full-text search the number of inspected items is

computed by counting the number of posts examined

before ﬁnding the correct expected posts. With the

topic-driven navigation approach the number of in-

spected items is the sum of two quantities: the number

of links followed to reach the closest topic in the Dis-

cussion Topics View of the TDForum plugin and the

number of posts examined before ﬁnding the correct

expected posts.

Table 5 reports the results obtained by executing

the evaluation protocol on the 11 tasks. The table re-

ports for each search goal the number of inspected

items and the time spent to ﬁnd the correct posts. For

Moodle full-text search we also reported the number

of search attempts (queries) performed with differ-

ent search keywords necessary to reach the goal. In

general the number of items inspected with full-text

search is in average higher than the number of items

inspected with TDForum (14 vs 9). The time neces-

sary to obtain the correct answer is in average less in

TDForum (137 sec. vs 170 sec.) because with full-

text search more time is spent to choose the correct

search keywords. The difference is not statistically

signiﬁcant due to the limited number of samples, thus

further experiment are necessary to draw more gen-

eral conclusions.

5 RELATED WORK

Recently, on-line education systems are becoming

widespread tools adopted by both historical and

newly founded educational institutions. E-learning

and e-teaching are new contexts for education where

large amounts of information are generated and ubiq-

uitously available. Most of generated information has

the form of free text without a structure crucial for

automating knowledge retrieval.

Data Mining has been historically used to extract

knowledge from free text (Baeza-Yates and Ribeiro-

Neto, 1999). Knowledge extraction from e-learning

systems, in particular from user generate data, has

been introduced in (Castro et al., 2007b; Hanna,

2004). Patterns of system usage by teachers and

learning behavior by students has been investigated in

(Tang and McCalla, 2005). Data clustering was sug-

gested to promote group-based collaborative learning

and to diagnose students incrementally (Castro et al.,

2007a).

Web Mining techniques to meet some of the cur-

rent challenges in distance education was presented

in (Sung Ho Ha, 2000) where a clustering of forum

messages are in fact grouped into similar discussion

topic classes. Association Rules mining has been

widely adopted in e-learning, in particular recommen-

dation systems (Za

ıane, 2002; Yang et al., 2010),

learning material organization (Tsai et al., 2001),

student learning assessments (Romero et al., 2005),

course adaptation to the students behavior (Hogo,

2010), and evaluation of educational websites (dos

Santos Machado and Becker, 2003). In educational

research the development of cooperative learning and

knowledge sharing inside student groups constitute

recent research trends(Jakobsone et al., 2012). To

this aim, Web technologies should grasp the oppor-

tunities raised by mixing the Social and the Semantic

Web (Ghenname et al., 2012) and on adopting Seman-

tic and Artiﬁcial Intelligence techniques for discov-

ering information objects and restructure large dig-

ital collections (Martin and Leon, 2012). Concept

maps and their use for navigation in educational con-

texts has been investigated in the recent past by dif-

ferent authors. As a representative of this research ef-

fort we cite the work of Dicheva and Aroyo (Dicheva

and Dichev, 2006). In this work the authors propose

a framework and a set of tools for the development

of ontology-aware repositories of learning materials.

While the idea and use of concept maps is similar to

our topic-driven navigation structure, in our approach

topics are extracted from free text in a semi-automatic

way, by leveraging information retrieval techniques

and then validated by the user, while concepts have

KDIR2014-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval

104

to be manually deﬁned by the authors of the learning

materials in the work of Dicheva and Aroyo.

6 CONCLUSIONS AND FUTURE

WORK

Online discussion forums are one of the main asyn-

chronous communication means and repositories of

user generated content over the Internet. Learning

management systems (LMSs), such as Moodle, use

forums to support interaction and collaboration be-

tween students and students-to-teachers. Discussions

taken place in a forum at some time represent a source

of information for users accessing the forum after-

wards. However, the effectiveness of a forum as a

source of information for its users, additionally to be

closely related to its richness in content, is also in-

ﬂuenced by the way its contents are organized made

searchable.

In this paper we presented an approach and a plu-

gin for the Moodle LMS that enhances content navi-

gation and information search in online discussion fo-

rums with a topic-driven navigational paradigm. The

approach enables the automatic recovery of a lattice

of discussion topics from the forum content, and the

introduction of an additional navigation structure and

graphical user interface which enable navigating and

searching forum contents by topics of discussion.

While the approach has proven correctness for

both the identiﬁed topics and the document-to-topics

assignment (Cerulo and Distante, 2013), in this paper

we have also shown with a case study that the addi-

tional navigation structure signiﬁcantly improves the

search of information stored in forum discussions.

In the future we aim to apply our approach in the

context of social networks, in order to explore how

it could improve social organization and user interac-

tion. As a matter of fact, social networks are increas-

ingly used in e-learning as side means for connecting

students and teachers.

REFERENCES

Baeza-Yates, R. A. and Ribeiro-Neto, B. (1999). Mod-

ern Information Retrieval. Addison-Wesley Longman

Publishing Co., Inc., Boston, MA, USA.

Bakalov, A., McCallum, A., Wallach, H. M., and Mimno,

D. M. (2012). Topic models for taxonomies. In

Proceedings of the 12th ACM/IEEE-CS Joint Con-

ference on Digital Libraries, JCDL ’12, Washington,

DC, USA, June 10-14, 2012, pages 237–240.

Birkhoff, G. (1967). Lattice theory. In Colloquium Publi-

cations, volume 25. Amer. Math. Soc., 3. edition.

Blei, D. M. (2011). Introduction to probabilistic topic mod-

els. Communications of the ACM.

Castro, F., Nebot, A., and Mugica, F. (2007a). Extraction of

logical rules to describe students’ learning behavior.

In Proceedings of the sixth conference on IASTED In-

ternational Conference Web-Based Education - Vol-

ume 2, WBED’07, pages 164–169, Anaheim, CA,

USA. ACTA Press.

Castro, F., Vellido, A., Nebot, A., and Mugica, F. (2007b).

Applying data mining techniques to e-learning prob-

lems. In Jain, L., Tedman, R., and Tedman, D., edi-

tors, Evolution of Teaching and Learning Paradigms

in Intelligent Environment, volume 62 of Studies in

Computational Intelligence, pages 183–221. Springer

Berlin Heidelberg.

Cerulo, L. and Distante, D. (2013). Topic-driven semi-

automatic reorganization of online discussion forums:

A case study in an e-learning context. In Global

Engineering Education Conference (EDUCON), 2013

IEEE, pages 303–310.

Dicheva, D. and Dichev, C. (2006). Tm4l: Creating and

browsing educational topic maps. British Journal of

Educational Technology, 37(3):391–404.

dos Santos Machado, L. and Becker, K. (2003). Distance

education: A web usage mining case study for the

evaluation of learning sites. In 2003 IEEE Interna-

tional Conference on Advanced Learning Technolo-

gies (ICALT 2003), 9-11 July 2003, Athens, Greece,

pages 360–361. IEEE Computer Society.

Ganter, B. and Wille, R. (1999). Formal concept analysis:

mathematical foundations. Springer.

Ghenname, M., Ajhoun, R., Gravier, C., and Subercaze, J.

(2012). Combining the semantic and the social web

for intelligent learning systems. In Global Engineer-

ing Education Conference (EDUCON), 2012 IEEE,

pages 1 –6.

Gruen, T. W., Osmonbekov, T., and Czaplewski, A. J.

(2006). eWOM: The impact of customer-to-customer

online know-how exchange on customer value and

loyalty. Journal of Business Research, 59:449456.

Hanna, M. (2004). Data Mining in the e-Learning Domain.

Campus-Wide Information Systems, 21(1):29–34.

Hogo, M. A. (2010). Evaluation of e-learning systems based

on fuzzy clustering models and statistical tools. Ex-

pert Syst. Appl., 37(10):6891–6903.

Hrastinski, S. (2008). What is online learner participa-

tion? a literature review. Computers & Education,

51(4):1755 – 1765.

Jakobsone, A., Kulmane, V., and Cakula, S. (2012). Struc-

turization of information for group work in an online

environment. In Global Engineering Education Con-

ference (EDUCON), 2012 IEEE, pages 1 –7.

Li, Q., Wang, J., Chen, Y. P., and Lin, Z. (2010). User

comments for news recommendation in forum-based

social media. Information Sciences, 180:49294939.

Martin, A. and Leon, C. (2012). An intelligent e-learning

scenario for knowledge retrieval. In Global Engineer-

ing Education Conference (EDUCON), 2012 IEEE,

pages 1 –6.

Meila, M. (2003). Comparing clusterings by the variation of

information. In Computational Learning Theory and

EnhancingOnlineDiscussionForumswithaTopic-drivenNavigationalParadigm-APluginfortheMoodleLearning

ManagementSystem

105

Kernel Machines, 16th Annual Conference on Compu-

tational Learning Theory and 7th Kernel Workshop,

COLT/Kernel 2003, Washington, DC, USA, August

24-27, 2003, pages 173–187.

Otterbacher, J. (2008). Searching for product experience at-

tributes in online information sources. In Proceedings

of the International Conference on Information Sys-

tems (ICIS 2008). Association for Information Sys-

tems.

Romero, C., Ventura, S., and Bra, P. D. (2005). Knowl-

edge discovery with genetic programming for provid-

ing feedback to courseware authors. User Modeling

and User-Adapted Interaction, 14(5):425–464.

Stefan, H. (2009). A theory of online learning as online

participation. Computers & Education, 52(1):78–82.

Sudau, F., Friede, T., Grabowski, J., Koschack, J., Make-

donski, P., and Himmel, W. (2014). Sources of in-

formation and behavioral patterns in online health fo-

rums: qualitative study. Journal of medical Internet

research, 16:e10.

Sung Ho Ha, Sung Min Bae, S. C. P. (2000). Web mining

for distance education.

Tang, T. and McCalla, G. (2005). Smart Recommenda-

tion for an Evolving e-Learning System: Architecture

and Experiment. International Journal on e-Learning,

4(1):105–129.

Tsai, C.-J., Tseng, S.-S., and Lin, C.-Y. (2001). A two-

phase fuzzy mining and learning algorithm for adap-

tive learning environment. In Proceedings of the

International Conference on Computational Science-

Part II, ICCS ’01, pages 429–438, London, UK, UK.

Springer-Verlag.

Wallach, H. M., Mimno, D. M., and McCallum, A. (2009).

Rethinking lda: Why priors matter. In Advances in

Neural Information Processing Systems 22: 23rd An-

nual Conference on Neural Information Processing

Systems 2009, pages 1973–1981.

Yang, Q., Sun, J., Wang, J., and Jin, Z. (2010). Seman-

tic web-based personalized recommendation system

of courses knowledge research. In Proceedings of the

2010 International Conference on Intelligent Com-

puting and Cognitive Informatics, ICICCI ’10, pages

214–217, Washington, DC, USA. IEEE Computer So-

ciety.

ıane, O. R. (2002). Building a recommender agent for e-

learning systems. In Proceedings of the International

Conference on Computers in Education, ICCE ’02,

pages 55–, Washington, DC, USA. IEEE Computer

Society.

Zhang, K. and Peck, K. (2003). The effects of peer-

controlled or moderated online collaboration on group

problem solving and related attitudes. Canadian Jour-

nal of Learning and Technology / La revue canadienne

de lapprentissage et de la technologie, 29(3).

KDIR2014-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval

106