PROMOTING RESEARCH COLLABORATION BASED ON DATA MINING TECHNIQUES IN LIBRARY INFORMATION SYSTEMS

Abstract

Research collaboration connects distributed knowledge and competencies into new ideas and research institutes and has been the subject of many research projects. We argue that academic libraries, including libraries of universities and research institutes, hold a wealth of information regarding patrons' research interests hidden in their data bases. Mining these databases can provide better understanding of researchers' needs and interests. This paper has two main contributions. Firstly, it proposes a new methodology based on data mining techniques in library information systems to uncover patrons' research interests in order to facilitate research collaboration including interdisciplinary research. The proposed methodology, studies data mining techniques in a library information system as a case study and makes advantage of clustering algorithms to cluster researchers based on their library usage which is interpreted as their research interests. The second contribution of this paper is that, it presented a knowledge map as a visual representation of usage trends of an academic library to portray virtual interest groups based on item use information. The result of this study can support managers and decision makers for strategic decision making regarding future research directions and collaborations. The outcome of the case study confirms our hypotheses by revealing clusters of library users with similar research interests validated by their academic backgrounds.

Full text

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[73]
PROMOTING RESEARCH COLLABORATION BASED ON DATA
MINING TECHNIQUES IN LIBRARY INFORMATION SYSTEMS
Elaheh Homayounvalaa, , Ammar Jalalimanesha
a Information engineering department, Iranian Research Institute for Information Science and
Technology, 1090, Enqelab St., Tehran, Iran, PO Box: 13185-137,1 Tehran, Iran,
Email: vala@irandoc.ac.ir
Abstract
Research collaboration connects distributed knowledge and competencies into new ideas and
research institutes and has been the subject of many research projects. We argue that academic
libraries, including libraries of universities and research institutes, hold a wealth of information
regarding patrons' research interests hidden in their data bases. Mining these databases can provide
better understanding of researchers' needs and interests. This paper has two main contributions.
Firstly, it proposes a new methodology based on data mining techniques in library information
systems to uncover patrons' research interests in order to facilitate research collaboration including
interdisciplinary research. The proposed methodology, studies data mining techniques in a library
information system as a case study and makes advantage of clustering algorithms to cluster
researchers based on their library usage which is interpreted as their research interests. The second
contribution of this paper is that, it presented a knowledge map as a visual representation of usage
trends of an academic library to portray virtual interest groups based on item use information. The
result of this study can support managers and decision makers for strategic decision making
regarding future research directions and collaborations. The outcome of the case study confirms
our hypotheses by revealing clusters of library users with similar research interests validated by
their academic backgrounds.
Keywords: Research collaborations, data mining, Library Information Systems, academic library,
patrons' research interests, researcher profiling, organizational knowledge map.

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[74]
1. Introduction
Research collaboration has received a great deal of
attention from governments and organizations
recently and many research projects has been carried
out in this subject (Bukvova, 2010). Research
collaboration connects distributed knowledge and
competencies into new ideas and research institutes
(Heinze and Kuhlmann, 2008). Amabile et al. Refers
to the definition of collaboration as "the coming
together of diverse interests and people to achieve a
common purpose via interactions, information
sharing, and coordination of activities". They identify
the core concept of this definition as “individuals
who differ in notable ways sharing information and
working toward a particular purpose”(Amabile et al.,
2001). Research collaborations have been modelled
in (Lee et al., 2011) by detecting dependency patterns
in research collaboration environments based on co-
authorship data at the organisation level. Dependency
patterns are extracted in this investigation by a
"cross-association clustering" technique. (Yang et al.,
2010) models research collaborations between
universities by exploring the link relation between
their websites.(Cucchiarelli and D'Antonio, 2010)
define a methodology for discovering potential
collaboration through the investigation of the relation
among actors in research networks. They used their
methodology to analyse a research-oriented network
that their users share potential research interests and
paper co-authorship. Heinze and Kuhlmann analyze
research collaboration in the growing domain of
nanoscience within the German public research
system. They based their research on multiple data
sources, such as co-publications, macro research
statistics, and in-depth interviews. They developed
governance structures that support scientists’ efforts
to start teamwork across institutional boundaries
(Heinze and Kuhlmann, 2008). Modelling research
collaboration based on library usage however, seems
not to be covered in the literature.
Data mining, on the other hand, has been applied in
information science generally and in library
information systems specifically, for many
applications. The applications of data mining to the
field of information science can be categorised in
three main categories of personalised environments,
electronic commerce and search engines (Chen and
Liu, 2004). In library information systems, however,
Papathedorou et. al summarise data mining
applications in three main categories of service
optimisation, decision support and personalization
(Papatheodorou et al., 2003). Scott Nicholson has
coined the term bibliomining for "use of data mining
to examine library data records" (Nicholson and
Stanton, 2003). He proposes utilizing data mining
tools to datasets of libraries in order to aid
organizational decision-making within the library or
improving library services or external reporting and
justification by tailoring services to meet the needs of
user groups (Nicholson, 2003, Nicholson and
Stanton, 2003).
He also suggests researchers to use bibliomining to
generalise their findings about one library to other
libraries by creating data warehouses for multiple
libraries in order to help obtaining deeper
understanding of institutions with the same tools
previously used with only one library (Nicholson,
2006). The bibliominig process, he describes,
consists of the following steps: “determining areas of
focus”, “identifying internal and external data
sources”, “collecting, cleaning, and anonymizing the
data into a data warehouse”, “selecting appropriate
analysis tools”, “discovery of patterns through data
mining and creation of reports with traditional
analytical tools” and finally “analyzing and
implementing the results” (Nicholson, 2003).
Data mining can reveal patterns of behaviour among
library users and staff and patterns of information
resource use throughout the organization. Sources of
data for data mining process in libraries includes, but
are not limited to, user information, circulation
information and searching and navigation
information in case of digital libraries (Nicholson and
Stanton, 2003). Data mining has been also applied in
other literatures for clustering scientific journals in
Digital Library (Lee et al., 2010) and for reader
classification or constructing user communities with
common interests by analyzing queries posed to a
digital library (Chang and Chen, 2006,
Papatheodorou et al., 2003). User communities are
defined as “groups of users who exhibit common
behaviour in their interaction with an information
system” (Orwant, 1994) in (Papatheodorou et al.,

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[75]
2003). (Kim et al., 2006) present a visual user-model
data mining tool based on user tracking information
such as queries and browsing result sets.
The application of data mining techniques in library
information systems for research collaboration seems
to be a new and novel application. We argue that
academic libraries including libraries of universities
and research institutes hold a wealth of information
regarding patrons' research interests hidden in
libraries data bases. Mining these databases can
provide better understanding of researchers' needs
and interests. This paper specifically studies applying
data mining techniques in Library Information
Systems of a research institute. It makes advantage of
an academic library usage to uncover patrons'
research interests in order to facilitate research
collaboration including interdisciplinary research.
The result of this study can help managers for
strategic decision making regarding future research
directions and also initiation of interdisciplinary
research projects in order to promote research
collaborations.
The organisation of this paper is as follows. First, the
proposed methodology in this paper is explained in
section 2. Then, a case study, namely IRANDOC
case study, is described in section 3 and finally
section 4 concludes the paper and makes suggestions
for future work.
2. Proposed Methodology
In order to uncover library users’ research interests,
especially to promote interdisciplinary research, we
made the assumption that researchers' borrowing
history from the research institute or university
library is an important source of information. Based
on this assumption, our proposed methodology
consists of four main steps as explained bellow and
depicted in figure 1:
1. Create a data warehouse based on
library operational data.
This step includes data collection,
refinement and pre-processing of raw data
and creation of a data warehouse.
2. Create knowledge map for library
subjects and their usage.
This step is designed in order to gain better
understanding of most studied subjects and
their relationships to other subjects by
creating a visual representation of library
usage. Creation of this organisational
knowledge map is the result of our former
study reported in detail in (Jalalimanesh and
Homayoun vala, 2011). The created
knowledge map is a network in the form of
an undirected graph. Each node represents a
subject based on library of congress
categorisation and the weight of each node
shows the total number of books borrowed
in that subject. Each edge between two
nodes of S1 and S2 represents total number
of books borrowed by all patrons for both
subjects S1 and S2. Since one person may
have borrowed different numbers of books
for subject S1 and subject S2, the minimum
of these two numbers is used for each person
in total calculation. Some parts of the
knowledge map for IRANDOC library is
represented in figure 2.
3. Analyse knowledge map to choose
specific subjects for further analysis.
Analysis of the knowledge map created in
step two will help decision makers to choose
some of the subjects for creation of research
teams or initiation of interdisciplinary
research projects. Decision makers can
choose number of subjects for further
analysis, considering their organizations
policies in addition to the created knowledge
map. Analysis of selected subjects will
result in better understanding of researchers’
needs and interests who are active in those
subjects.
4. Apply data mining techniques to
find active researchers in selected subjects
from knowledge map analysis.
Data mining can be applied to cluster researchers
based on their similarities in borrowing books in
selected subjects. Cluster analysis will consequently
create better understanding of patrons' research

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[76]
interests which assists decision makers regarding research collaborations.
Data Warehouse
Creation
Knowledge Map
Creation
Knowledge Map
Analysis
Applying Data
Mining Techniques
on Selected
Subjects
Operational
Database
Data Warehose
Knowledge
Map
Data Mining
Techniques
Selected Subjects
cluster0 cluster1 cluster2
Clusters of
Researchers
Decisions Regarding
Research Collaborations
Decision Makers Opinions
Figure 1 – Visual representation of the proposed methodology.
Organisational knowledge map, as it can be seen in
figure 2, can offer a visual representation of most
studied subjects as well as inter-connections of
subjects to identify interdisciplinary research fields in
the research institute under study. Then clustering
techniques such as k-means or k-medoids can be
applied in order to find researchers with similar
research interests.
Figure 2 – Organisational knowledge map based on library information system.
k-means (MacQueen, 1967) is one of the simplest and
most popular unsupervised clustering algorithms and
among top ten data mining algorithms(Wu et al.,
2008). Each point, in this algorithm, is assigned to
the cluster whose centre is the nearest. k in k-means
algorithm is fixed a priori and k centroids (centres)

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[77]
are placed randomly at the beginning of the
algorithm. k-means algorithm then recalculates the
centroid or mean of each cluster and repeats
assigning points to new centroids until centroids do
not change anymore(Witten et al., 2011). k-means
algorithm measure distance between two points based
on squared Euclidean distances, but another
unsupervised clustering algorithm which is called k-
medoids minimises a sum of pair wise dissimilarities.
Both k-means and k-medoids use expectation-
maximisation strategy to converge to a minimum
error condition. Centres in k-medoids algorithm must
be one of the points to be clustered in contrary to k-
means that centres could be any point in the space.
"k-medoids is more robust to noise and outliers as
compared to k-means" (Han and Kamber, 2006) and
works well for small data sets. In our proposed
methodology, a comparison of k-mean and k-medoids
is presented for clustering.
In order to evaluate clustering algorithm, Davis-
Bouldin index (Davies and Bouldin, 1979) is applied.
The smaller the index the better the clustering
algorithm, because Davis-Bouldin index calculates
the ratio of the sum of within-cluster scatter to
between-cluster separation as follows:
Here n is the number of clusters equal to k in k-
means and k-medoids algorithms. 󰇛󰇜is the ratio
of the average distance of all objects in cluster  to
their cluster centre and 󰇛󰇜 is the distance
between cluster  and cluster  centres. Therefore
Davis-Bouldin is a small number if the clusters are
dense and also distant from each other.
3. IRANDOC Case Study
Iranian Research Institute for Information Science
and Technology (IRANDOC) is an institute affiliated
with the Ministry of Science, Research, and
Technology (MSRT) which was established to work
in the field of science and technology of Information
and Librarianship. IRANDOC library has about
14000 Latin books. IRANDOC library books are
organized based on the Congress Classification
System and is run on the basis of the Open-shelf
System. The users of the library are comprised of
university professors, students, researchers, and the
IRANDOC staff. Based on the Library Collection
Policy, the IRANDOC Library, at present, provides
the following subjects: Information Science and
subjects related to the Library Science, Information
Systems Management, Information Technology,
Information Analysis, Knowledge & Information
Management, Linguistics, Computerized
Terminology and Technology.
We used the methodology, we have proposed in our
earlier study, for drawing knowledge map based on
library information system user logs (Jalalimanesh
and Homayoun vala, 2011). IRANDOC knowledge
map drawn based on this methodology is partly
illustrated in figure 2. The size of sphere in each node
shows the amount of books that studied by
IRANDOC researchers totally. The width of bars that
connect subjects together shows the amount of
studies. IRANDOC knowledge map shows most
studied subjects and also interrelation between them
which are invaluable source of knowledge for
IRANDOC decision makers in order to initiate
research projects.
This figure shows the high interests of library users in
bibliography, library science and general information
resources. Assuming that decision makers in
IRANDOC decides to initiate a research project in
interdisciplinary subjects of Z (writing, book
industries and trade, libraries, bibliography), HD
(industries and land use) and QA (Mathematics), then
by applying data mining techniques for these subject
Z, HD and QA, we can cluster library users according
to their interests in mentioned subjects. Figure 3
demonstrates a three dimensional scatter plot of
library users who are interested in these three
subjects based on the number of books studied in
each subjects.
 = 1
max{󰇛󰇜+
,

=1 }

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[78]
Figure 3 – 3D Scatter plot of subject Z, QA and HD.
Figure 4 demonstrates clustering of library users based on their similarities in books they have studied in mentioned
subjects.
Figure 4 – Two dimensional Scatter plot for subjects Z, QA and HD clustered by k-means algorithm.
k-means clustering algorithm in Rapidminer software is used for clustering process. Figure 5 represents same users,
clustered by k-medoids algorithm. Number of clusters in both clustering is four.

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[79]
Figure 5 – Two dimensional Scatter plot for subjects Z, QA and HD clustered by K-medoids algorithm.
Table 1: Summarises performance of two algorithms.
Performance Table
K-means Algorithm
K-Medoids
Average within centroid
distance
400.869
1354.053
Average within centroid
distance cluster_0
0.000
14395.167
Average within centroid
distance cluster_1
116.533
567.944
Average within centroid
distance cluster_2
1530.346
38.478
Average within centroid
distance cluster_3
4475.250
638.000
Davies Bouldin index
0.946
1.527
Table 1 – Performance tables for two algorithms

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[80]
Although Davies-Bouldin index for k-means
algorithm is better than k-medoids algorithm,
analysis of created clusters in two algorithms reveals
that k-medoids clusters are more meaningful than
clusters in k-means algorithm. This supports the fact
that k-medoids is more suitable for small data sets
and is more robust to noise and outliers (Han and
Kamber, 2006). Clustering with k-means algorithm
resulted in clusters with just one and two members
whose average within centroid distance are zero or
very small and hence have smaller Davies-Bouldin
index.
Analysis of clusters created by k-medoids algorithm,
on the other hand, proves to be more meaningful in
our special case study as discussed here. Cluster_2
members (red points) consist of library users who
studied very limited number of books in all three
subjects under investigation. This presumably
indicates that they are not interested in mentioned
subjects and we can call them "uninterested cluster".
Cluster_0 members (light green points) are far from
all three x, y and z axes. These members borrowed
substantial number of books in all three subjects,
although they might be much more interested in one
or two subjects than others, but they all are good
candidates for creating research teams on
interdisciplinary subjects of Z, QA and HD. We can
call this cluster "cross disciplinary active
researchers". The average within centroid distance of
this cluster is the highest in k-medoids algorithm
which confirms the fact that members of this cluster
are far from each other in the space, but for our
special case study they form a meaningful cluster.
Cluster_3 members (green dots) are mostly close to
z-axis. They have borrowed many books from library
science (Z) subject and few books from HD and QA
class. This cluster is named "single disciplinary
active researchers". Cluster_1 (blue dots) members
borrowed books from all subjects but not as much as
those borrowed by cluster_0.
In order to evaluate the quality of the clustering
results, samples of cluster members were further
investigated based on their research backgrounds
reflected in their academic resumes and their research
projects. k-medoids clusters are proved to be more
meaningful clusters. As an example, by choosing
three library users from cluster_1 in k-medoids
algorithm randomly, it was figured out that all of
them have been researching in library science and
information resource, but they used system
engineering techniques to handle their research
projects. This confirms our findings regarding their
research interests. Investigating number of
researchers in cluster_0 against their research
backgrounds also validate the fact that they are very
active researchers who do not restrict their studies in
one subject and are interested to bring other subjects
to their major field of study.
4. Conclusions and Future Work
This paper had two contributions. Firstly, it proposed
a new methodology for promoting research
collaboration based on data mining techniques. The
proposed methodology is applicable in the case of
academic institutes such as universities and research
institutes with active library usage. We argue that
information hidden in databases of academic libraries
is a key resource for eliciting patrons' research
interests. The case study, in which this hypothesis is
studied, is a library information system in a research
institute, namely IRANDOC research institute.
Library databases were mined to gain better
understanding of patrons' research interests based on
their book borrowing history in order to assists
managers in making decisions regarding promoting
collaborations including adding new features to
library information system. These features may
include creating reports for managers on potential
research teams, based on library usage. Such reports
assists decision makers in research team formation,
especially in interdisciplinary research subjects. It
can also support sending messages to patrons
introducing researchers with similar research interests
in order to promote collaborative research. Another
important feature can be consulting active researchers
for ordering new books.
The second contribution of this paper is that, it
presented a visual representation of usage trends of
IRANDOC library to portray virtual interest groups
based on item use information. The organizational
knowledge map presented here is a valuable visual
tool for supporting decision makers and offer

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[81]
substantial contribution facilitating analysis process
in our proposed methodology.
In summary, this study demonstrated the application
of data mining for research collaboration and built a
case study to extract research collaboration strategies
based on library usage. Two types of strategies can
be extracted by this methodology: library users'
research collaboration strategies and also managers
or decision makers' strategies for research
collaborations. Therefore, the result of this study has
two folds; it can provide better service provision for
individuals and also provides important information
for library managers as well as research institute
managers and decision makers. The case study
approves validity of discovered patrons' research
interests compared to their research background.
The research reported in this paper may be further
continued by working on sub-categories of each
library subject which gives a more detailed analysis
of patrons' research interests.
Acknowledgment
The authors gratefully acknowledge the support of
the Iranian Research Institute for Information Science
and technology, Specially Dr Babak Seyfe, Dr.
Gholam Ali Montazer, Dr. Hamid Reza Jamali and
Dr. Omid Fatemi.
5. References
1. AMABILE, T. M., PATTERSON, C.,
MUELLER, J., WOJCIK, T., ODOMIROK,
P. W., MARSH, M. & KRAMER, S. J.
2001. Academic-practitioner collaboration
in management research: A case of cross-
profession collaboration. The Academy of
Management Journal, 44, 418-431.
2. BUKVOVA, H. 2010. Studying Research
Collaboration: A Literature Review.
Sprouts: Working Papers on Information
Systems, 10(3).
3. CHANG, C. C. & CHEN, R. S. 2006. Using
data mining technology to solve
classification problems: A case study of
campus digital library. The Electronic
Library, 24, 307-321.
4. CHEN, S. Y. & LIU, X. 2004. The
contribution of data mining to information
science. Journal of Information Science, 30,
550.
5. CUCCHIARELLI, A. & D'ANTONIO, F.
Mining Potential Partnership through
Opportunity Discovery in Research
Networks. Advances in Social Networks
Analysis and Mining conference, 2010.
IEEE, 404-406.
6. DAVIES, D. L. & BOULDIN, D. W. 1979.
A cluster separation measure. Pattern
Analysis and Machine Intelligence, IEEE
Transactions on, 224-227.
7. HAN, J. & KAMBER, M. 2006. Data
mining: concepts and techniques, Morgan
Kaufmann.
8. HEINZE, T. & KUHLMANN, S. 2008.
Across institutional boundaries?:: Research
collaboration in German public sector
nanoscience. Research Policy, 37, 888-899.
9. JALALIMANESH, A. & HOMAYOUN
VALA, E. Organizational knowledge
mapping based on library information
system (IRANDOC case study) IADIS
International Conference, Collaborative
Technologies, 2011 Rome, Italy.
10. KIM, S., LELE, S., RAMALINGAM, S. &
FOX, E. 2006. Visualizing user
communities and usage trends of digital
libraries based on user tracking information.
Digital Libraries: Achievements, Challenges
and Opportunities, 111-120.
11. LEE, B., KWON, O. & KIM, H. 2011.
Identification of dependency patterns in
research collaboration environments through
cluster analysis. Journal of Information
Science, 37, 67.
12. LEE, J. Y., KIM, H. & KIM, P. J. 2010.
Domain analysis with text mining: Analysis
of digital library research trends using

International Journal of Information Technology and Business Management
29th December 2012. Vol.8 No. 1
© 2012 JITBM & ARF. All rights reserved
ISSN 2304-0777 www.jitbm.com
[82]
profiling methods. Journal of Information
Science, 36, 144.
13. MACQUEEN, J. Some methods for
classification and analysis of multivariate
observations. 5-th Berkeley Symposium on
Mathematical Statistics and Probability,
1967 Berkeley. California, USA.
14. NICHOLSON, S. 2003. The bibliomining
process: Data warehousing and data mining
for library decision making. Information
technology and libraries, 22, 146-151.
15. NICHOLSON, S. 2006. The basis for
bibliomining: frameworks for bringing
together usage-based data mining and
bibliometrics through data warehousing in
digital library services. Information
processing & management, 42, 785-804.
16. NICHOLSON, S. & STANTON, J. M.
2003. Gaining strategic advantage through
bibliomining: Data mining for management
decisions in corporate, special, digital, and
traditional libraries. In: NEMATI, S. &
BARKO, C. D. (eds.) Organizational Data
Mining: Leveraging Enterprise Data
Resources for Optimal Performance. Idea
Group Publishing.
17. ORWANT, J. 1994. Heterogeneous learning
in the Doppelgänger user modeling system.
User Modeling and User-Adapted
Interaction, 4, 107-130.
18. PAPATHEODOROU, C., KAPIDAKIS, S.,
SFAKAKIS, M. & VASSILIOU, A. 2003.
Mining user communities in digital libraries.
Information technology and libraries, 22,
152-157.
19. WITTEN, I. H., FRANK, E. & HALL, M.
A. 2011. Data mining: practical machine
learning tools and techniques, Morgan
Kaufmann San Francisco.
20. WU, X., KUMAR, V., ROSS QUINLAN,
J., GHOSH, J., YANG, Q., MOTODA, H.,
MCLACHLAN, G. J., NG, A., LIU, B. &
YU, P. S. 2008. Top 10 algorithms in data
mining. Knowledge and Information
Systems, 14, 1-37.
21. YANG, B., LIU, Z. & MELOCHE, J. A.
2010. Visualization of the Chinese academic
web based on social network analysis.
Journal of Information Science, 36, 131.