Link Analysis in Mind Maps: A New Approach to
Determining Document Relatedness
Jöran Beel
Otto-von-Guericke University
FIN / ITI / VLBA-Lab
Germany
beel@sciplore.org
Bela Gipp
Otto-von-Guericke University
FIN / ITI / VLBA-Lab
Germany
gipp@sciplore.org


ABSTRACT
In a previous paper we presented various ideas on how
information retrieval on mind maps could enhance applications
such as expert systems, search engines and recommender
systems. In this paper we present the first research results. In a
brief experiment we researched link analysis respectively citation
analysis, if applied to mind maps, is suitable to calculate
document relatedness. The basic idea is that if two documents A
and B are linked by the same mind map, these documents are
likely to be related. This information could be used by item-
based document recommender systems. In the example,
document B could be recommended to those users interested in
document A. In addition, we propose that those documents linked
in high proximity within a mind map are more closely related
than those documents linked in lower proximity. The results of
our experiment support our ideas. It seems that link analysis
applied to mind maps can be used for determining the
relatedness of documents and therefore for improving document
recommender systems.
Categories and Subject Descriptors
H.3.3 [Information Storage and Retrieval]: Information Search
and Retrieval – information filtering, retrieval models, search
process
H.3.7 [Information Storage and Retrieval]: Digital Libraries –
system issues, user issues
General Terms
Algorithms, Measurement, Documentation.
Keywords
mind maps, recommender systems, research paper recommender,
document recommender, metrics, citation analysis, link analysis
1. INTRODUCTION
Mind mapping is a common method to structure and visualize
ideas, manage electronic literature and to draft documents. Some
users do link in their mind map to external documents such as
PDFs or websites. Some even cite scholarly literature, for
instance by adding BibTeX keys to a mind map’s node (see
Figure 1 for an example). In a recent paper we proposed to
analyze these links and references to determine the relatedness of
those documents that are linked in the mind map [1]1.
The basic idea is that two documents are related if they are both
linked by a mind map. In addition, it was assumed that the closer
the links occur in the mind map, the higher related the linked
documents are. If the assumption proves to be right, Link
Analysis in Mind Maps (LAMM) could be used to enhance
search engines and document recommender systems since these
systems often present related documents to their users.
We conducted a brief experiment to test the proposed idea and
present the results in this paper. The focus of this paper lies on
calculating the relatedness of scholarly literature and on
enhancing research paper recommender systems as we plan to
integrate LAMM into our academic search engine and research
paper recommender system SciPlore2. However, it's highly
probable that the results would be similar for other kind of
documents linked by a mind map such as websites.
In the next section, related work about research paper
recommender systems and citation analysis is presented. It is
then followed by a section showing the methodology which has
been used to evaluate LAMM. Finally, the results, a discussion,
and an outlook towards future work conclude.
2. RELATED WORK
Several attempts have been made to establish research paper
recommender systems [2-7]. Some of them use citation analysis
to determine the degree of relatedness between two papers. An
overview of different citation analysis approaches for
determining the relatedness of research papers is given in [8]. At

1 In this paper we do not distinguish between linking files and
referencing scholarly literature, for instance with a BibTeX
key. Citations, links to files on the user’s hard drive and
hyperlinks to websites are all considered as ‘link’.
2 http://www.sciplore.org

Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
Proceedings of the 4th International Conference on Ubiquitous
Information Management and Communication’10, January 14–15, 2010,
Suwon, Korea.
Copyright 2010 ACM, ISBN 978-1-60558-893-3

Preprint of: Jöran Beel and Bela Gipp. Link Analysis in Mind Maps: A New Approach To Determine Document Relatedness. In Proceedings of the Fourth
International Conference on Ubiquitous Information Management and Communication (ICUIMC’10). ACM, January 2010. Downloaded from
http://www.sciplore.org

this time, our research focuses on co-citation analysis [9] and its
extension citation proximity analysis [10].
According to co-citation analysis, two papers A and B are related
if a third paper C references both. If more than one paper
reference paper A and B together, their relatedness is supposed
to be even higher. Citation proximity analysis additionally
considers the location of citations in the full text: Two papers A
and B are supposed to be more highly related when they are
closely referenced by a third paper C in the text. For instance, if
paper C references paper A and B in the same sentence, A and B
are likely to be highly related. If paper C references paper A in
the beginning of a 100-page document and paper B at the end,
their relatedness is probably not nearly as high.
Co-citation analysis and citation proximity analysis can be used
by research paper recommender systems to make item-based
recommendations: If paper A and B are related, paper B may be
recommended to those users interested in paper A (but not
knowing paper B yet).
However, co-citation analysis and citation proximity analysis
have to cope with some drawbacks.
1. Availability of Data: Co-citation analysis and citation
proximity analysis cannot be applied to all research
papers due to a lack of (correct) data [11, 12]: many
research papers are not cited at all; citation databases
such as ISI Web of Knowledge do not cover all
available publications; and due to technical difficulties,
citations are not always recognized correctly, which in
turn leads to incorrect data in citation databases.
2. Robustness of Data: Citations are often considered as
biased because authors do cite papers they should not
cite and do not cite papers they should cite [12].
Accordingly, citation based recommender systems
might provide irrelevant recommendations.
3. Timeliness of Data: Publishing scientific articles is a
slow process and it takes months or even years before
they are published and citations are received.
Accordingly, documents recommended based on
citation analysis are, at the very least, several months
old.
4. Metrics: There exist metrics for measuring the
relatedness of research papers based on citation
analysis (for instance, coupling strength [13] or the
citation proximity index [10]). However, to our
knowledge, each metric focuses solely on one citation
analysis approach and no combining metric exists yet.
Consequently, relatedness of research papers based on
citations cannot be measured and expressed thoroughly.
Summarized, citation analysis applied to scholarly literature can
do a good job in identifying related articles, but there is room for
improvement.
3. METHODOLOGY
Our intention was to conduct an experiment to obtain first
indications if Link Analysis in Mind Maps (LAMM) might be
suitable for determining research paper relatedness. Two
assumptions were researched:
1. Two research papers A and B are related if at least one
mind map links them both
2. Two research papers A and B are more highly related
the more closely they are linked within a mind map
As part of the experiment, five mind maps were analyzed which
were originally created for drafting research papers, respectively
Masters Theses3. That means each of the mind maps links at
least to a few PDF files representing academic articles. From
each mind map, links (respectively citations) to three articles
were extracted and pairs were built (see Figure 2 for
illustration). The first pair was built from the first and second
link in a mind map. Since the distance between them was low,
we expected this pair to be ‘highly related’. The second pair was
built from the first and last link in the mind maps. Here, the
distance between the links was high. Accordingly, we expected
the corresponding articles to be less closely related.

3 Two mind maps represented drafts of our own papers and three
mind maps were created by some of our students for their
Masters’ theses.

Figure 1: Mind map draft of a paper (arrows indicate a link to a PDF file; the tooltip displays a BibTeX key)