Collaborative filtering and inference rules
for context-aware learning object
recommendation
1. INTRODUCTION
With the proliferation of the Internet, demand for
on-line learning has grown rapidly. Often used to
inexpensive just-in-time information, students now
expect similar services from learning institutions.
For teachers, these new expectations can be chal-
lenging. The design and production of on-line
courses is expensive and time-consuming. When
providi g digital content, it is no longer adequate
Interactive Technology & Smart Education (2005) 2: 179––190
©2005 Troubador Publishing Ltd.
Daniel Lemire
UQAM, 4750, avenue Henri-Julien, Montréal, QC H2T 3E4 Canada
Email: lemire@ondelette.com
Harold Boley
NRC, 46 Dineen Drive, Fredericton, New Brunswick E3B 9W4, Canada
Email: harold.boley@nrc.gc.ca
Sean McGrath and Marcel Ball
3 UNB, 540 Windsor Street, Fredericton, New Brunswick E3B 5A3, Canada
Email {sean.mcgrath,marcel.ball}@unb.ca
Learning objects strive for reusability in e-Learning to reduce cost and allow personalization of content. We show why
learning objects require adapted Information Retrieval systems. In the spirit of the Semantic Web, we discuss the
semantic description, discovery, and composition of learning objects. As part of our project, we tag learning objects
with both objective (e.g., title, date, and author) and subjective (e.g., quality and relevance) metadata. We present the
RACOFI (Rule-Applying Collaborative Filtering) Composer prototype with its novel combination of two libraries and
their associated engines: a collaborative filtering system and an inference rule system. We developed RACOFI to gen-
erate context-aware recommendation lists. Context is handled by multidimensional predictions produced from a data-
base-driven scalable collaborative filtering algorithm. Rules are then applied to the predictions to customize the rec-
ommendations according to user profiles. The RACOFI Composer architecture has been developed into the context-
aware music portal inDiscover.
Keywords: Learning Objects, Semantic Web, Collaborative Filtering, Recommender Systems, Slope One, Inference
Rules, RuleML
VOL2 NO3 AUGUST2005 179

to merely manually adjust or adapt the course con-
tent for students; students should also be empow-
ered to navigate independently (Lundgren-Cayrol et
al., 2001).
The Web also faces similar challenges. As the
Web becomes ubiquitous, our needs become more
sophisticated. For example, while we may have been
satisfied in the past with weather reports for a given
area, we now want to be able to plan our vacations to
other areas and thus, coordinate data coming from
weather reports, hotels, and air travel companies.
While we generally know how to find Web sites on
a given topic using search engines, we still can’t eas-
ily find all air travel companies offering flights from
Montréal to Rio next week for under 1000 dollars. It
follows that we can’t expect our computers to sug-
gest travel packages automatically from data gath-
ered over the Web. We observe that the Web is not
a database (Mendelzon, 1998) in that there are no
built-in common schemas or sophisticated data
retrieval mechanism. Yet the Web is the most suc-
cessful data management tool ever developed. We
distinguish two different future challenges for the
Web: Information Retrieval and Composition. One
approach to the solution to these problems can be
found in the Semantic Web(Berners-Lee, 1998).
Essentially, the Semantic Web adds to the current
Web enough metadata so that the Web can be con-
sidered machine-parseable (Koivunen & Miller,
2001). In theory, it should render the Information
Retrieval problem easier, and one approach to
Composition can then be achieved through
Artificial Intelligence (AI) planning techniques.
Just like a Web site, digital knowledge and train-
ing material should be reusable. For example, once
basic arithmetic is achieved, individuals reuse this
knowledge all their life; a document about arith-
metic is also reusable in a similar fashion. In this
spirit, the notion of a learni g object(Downes, 2001;
Gibbons et al., 2002) was proposed as a way to
enable content reuse in e-Learning. Essentially, a
learning object is any component that can be used
and reused for learning. A map, a Web site, a piece
of software, and a video stream are all examples of
learning objects. Thanks to projects such as
eduSource, learning objects have become tangible
(Bhavsar et l., 2003, 2004). For example, the Web
site KnowledgeAgora (http://www.knowledgeago-
ra.com/) offers a learning object taxonomy and a
search engine. Users can sell, buy, exchange or offer
learning objects. In KnowledgeAgora, unlike the
Web at large, Intellectual Property, age range, and
education level are explicitly handled, which is vital
for content reuse.
Object composition in education is already pres-
ent in fields s ch as adaptive testing (Raîche, 2000);
wever, learning objects belong to a more hetero-
geneo s and distributed setting than educators are
accustomed to. Few examples of computer support-
ed object composition have been reported (Fiaidhi
et al., 2004; Fiaidhi, 2004). Learning objects have to
be fine-grained enough so that reusability is sensi-
ble. That is, we need to have many small objects and
we can aggregate to form larger objects, such as
courses. The problem in locating learning objects is
much harder than simply finding the proper digital-
ized textbook for a c urse. An example of the prob-
lem we want to solv would be to find all resources
related to “Java inheritance” for first year Computer
Scie c students, such hat the students are likely
t find the content of at least average interest, but
such that it w s rated above average for accuracy by
instructors.
O e might think that Information Retrieval as
acc mplished on the Web can be applied to learn-
ing o jec s. After all, Google is very efficient at
finding content as long as requests can be expressed
as a list of keywords. However, one should note that
Google works well in part because the Web is made
of links in a fundam ntal way, unlike learning
objects. The analogy between the HTML Web and
l arning objects is not perfect. Learning objects are
no necessarily text-based and they are not linked to
each other as explicitly as Web sites. Google
searching is based on the assumption that a Web
page frequently linked to must offer relevant con-
tent and returns such results to the searcher.
Naturally, cours content does include links and
relationships, possibly through Dublin Core, but
learning objects w ’t make up a graph the same
way the Web is a directed graph of Web pages.
Effectively, finding the right resource to express
a given idea remains a challenge. When an instruc-
tor or student wants to find a particular type of
learning object that is of interest to them, they need
a way to specify their interest to the system that
i terfaces with the learning object repository.
Figure 1 shows some of the common methods
of doing so: navigating through a taxonomy,
perf rming keyword searches, specifying their
Lemire et al.: Collaborative Filtering and Inference Rules
INTERACTIVE TECHNOLOGY& SMART EDUCATION180
Figure 1 A few of the options a user has for finding learning
objects