Web analytics of user path tracing and a novel algorithm for generating recommendations in Open Journal Systems
| Date | 23 September 2013 |
| Published date | 23 September 2013 |
| DOI | https://doi.org/10.1108/OIR-09-2012-0152 |
| Pages | 672-691 |
| Author | Behnam Taraghi,Martin Grossegger,Martin Ebner,Andreas Holzinger |
| Subject Matter | Library & information science,Information behaviour & retrieval |
Web analytics of user path
tracing and a novel algorithm for
generating recommendations in
Open Journal Systems
Behnam Taraghi, Martin Grossegger, Martin Ebner and
Andreas Holzinger
Graz University of Technology, Graz, Austria
Abstract
Purpose – The use of articles from scientific journals is an important part of research-based teaching
at universities. The selection of relevant work from among the increasing amount of scientific
literature can be problematic; the challenge is to find relevant recommendations, especially when the
related articles are not obviously linked. This paper seeks to discuss these issues.
Design/methodology/approach – This paper focuses on the analysis of user activity traces in
journals using the open source software “Open Journal Systems” (OJS). The research questions to what
extent end users follow a certain link structure given within OJS or immediately select the articles
according to their interests. In the latter case, the recorded data sets are used for creating further
recommendations. The analysis is based on an article matrix, displaying the usage frequency of
articles and their user selected successive articles within the OJS. Furthermore, the navigation paths
are analysed.
Findings – It was found that the users tend to follow a set navigation structure. Moreover, a hybrid
recommendation system for OJS is described, which uses content based filtering as the basic system
extended by the results of a collaborative filtering approach.
Originality/value – The paper presents two original contributions: the analysis of user path tracing
and a novel algorithm that allows smoo th integration of new articles into t he existing
recommendations, due to the fact that scientific journals are published in a frequent and regular
time sequence.
Keywords Algorithms, OpenJournal System, Path analysis, Recommendersystem,
Reseach-basedteaching, Web analytics
Paper type Research paper
Introduction and motivation for research
Scientific literature is an important knowledge source for research-based teaching
activities in universities (Handelsman et al., 2004; Holzinger, 2010, 2011). Established
engineering education is mostly deductive, i.e. the teacher starts with basics and
fundamentals, then moves to theories and consequently progresses to the applications
of those theories. Alternative teaching approaches are more inductive, e.g. topics are
introduced by presenting specific observations, case studies or problems and theories
are taught using examples or the students are assisted to discover them.
Student-centred and inductive teaching methods can be helpful in reaching such
goals (Motschnig-Pitrik and Holzinger, 2002; Prince and Felder, 2006).
For all these approaches, the recommendation of appropriate literature is a major
issue and Recommender Systems (RS) are an alternative method of presenting and
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/1468-4527.htm
OIR
37,5
672
Received 2 September 2012
First revision accepted
26 February 2013
Online Information Review
Vol. 37 No. 5, 2013
pp. 672-691
qEmerald Group Publishing Limited
1468-4527
DOI 10.1108/OIR-09-2012-0152
recovering knowledge from various resources and therefore a further step towards
personalised e-learning systems (Khribi et al., 2009). In contrast to search engines,
which deliver results from a user query, RS are focused on delivering unexpected
results (Kim et al., 2004), which also support the pooling and sharing of information for
informal learning (Linton and Schaefer, 2000). They even try to create
recommendations for resources, whereas search engines are restricted in that regard
due to limitations in automatic content analysis.
In this paper we introduce an RS for an online journal system. We focus on the
analysis of user paths (the sequence of articles read by users known as user activity) at
first. The influence of an indexing structure such as the table of contents in the
navigation path is examined and visualised. We emphasise that the recorded data can
be misleading when applied in the recommendation algorithms, as they do not reflect
the interests of users appropriately. The analysis shows that users mainly follow the
existing table of contents. Based on this analysis we introduce a hybrid RS consisting
of a content-based technique and a collaborative filtering algorithm. The collaborative
filtering algorithm is based on the weighted user paths. We introduce a method to
reduce the undesirable effect of the navigation path (in our case a table of contents) so
that the proposed collaborative filtering algorithm can provide relevant serendipitous
and novel recommendations.
For our analysis a journal with an installation of Open Journal Systems (OJS) was
used. OJS (Willinsky, 2005) is a journal management and publishing program that
includes the possibility to generate peer reviews. To extend the functionality of OJS,
the possibility of plug-in development is offered. Consequently a RS can easily be
integrated and made available for all installations of OJS. The main indexing structure
is the list of articles of an issue of one particular journal.
Background and related work
There are different types of RS that vary in terms of the addressed domains and user
requirements. Content-based RS try to recommend items that are similar to those the
users have liked before. They rely on matching the users’ interests and preferences to
the attributes of the items (Mladenic, 1999). Similar items are those that have the same
attributes in common. For example in a movie recommendation application attributes
such as actors, genres and subjects can be taken into consideration for each movie.
Each item is described as a set of predefined attributes in a structured way. The
attributes can have a known set of values in this case. Some similarity measures such
as cosine similarity or machine learning algorithms can be used to learn the similar
items and the associated user profiles (Pazzani and Billsus, 2007; Amatriain et al.,
2011). In social tagging RS the user tagging activity is taken into account. Tags are
freely chosen keywords that are used by users to annotate and categorise the resources
within a social content sharing system. Szomszor et al. (2007) introduced a movie RS.
The algorithm used in their approach is based on the similarity between tags of a
movie and the tags of movies the user has already rated. Diederich and Iofciu (2006)
introduced a prototype that uses tag-based profiles to find and recommend people with
similar interests. Instead of using objects directly, they used the tags associated with
the objects to build the user’s profile and enhance a user’s current community of
practice.
To generate content-based recommendations for text-based resources such as
documents or articles, it is necessary to analyse and compare the textual contents.
Open Journal
Systems
673
To continue reading
Request your trial