Topological analysis of knowledge maps
Jun Liu
a,
⇑
, Jincheng Wang
a
, Qinghua Zheng
a
, Wei Zhang
b
, Lu Jiang
a
a
SPKLSTN Lab, Department of Computer Science, Xi’an Jiaotong University, Xi’an, China
b
Amazon.com, Inc., Seattle, WA 98109, USA
article info
Article history:
Received 2 February 2012
Received in revised form 24 May 2012
Accepted 22 July 2012
Available online 31 July 2012
Keywords:
Knowledge map
Complex network
Learning-dependency
Topological property
Topological analysis
abstract
A knowledge map can be viewed as a directed graph, in which each node is a knowledge unit (KU), and
each edge is a learning-dependency between two KUs. Understanding the topological properties of
knowledge map can help us gain better insights into human cognition structure and its mechanism,
design better knowledge map construction algorithms, and guide learners’ navigational learning through
knowledge map. In this paper, we perform topological analysis on 12 knowledge maps from computer
science, mathematics, and physics. We discover that they exhibit small-world and scale-free properties
like many other networks. Specifically, we show the locality of learning-dependency and hierarchical
modular structure in the 12 knowledge maps. In addition, we study how KUs affect the network effi-
ciency by removing KUs based on different centrality measures. We find that the importance of KUs var-
ies greatly.
Ó 2012 Elsevier B.V. All rights reserved.
1. Introduction
The learning process can be considered as a hierarchical pro-
cess, because understanding a new knowledge unit (KU) often re-
lies on the understanding of other existing KUs [1]. KU is defined
as the basic integral knowledge of a given domain, such as a defi-
nition, a theorem, or an algorithm. For example, in ‘‘Plane Geome-
try,’’ we find the learning-dependency from the KU ‘‘Definition of
right angle’’ to KU ‘‘Definition of right triangle’’, which indicates
that one should learn the ‘‘Definition of right angle’’ before learning
the ‘‘Definition of right triangle.’’ The KUs of a given domain and
the learning dependencies among them form a directed graph,
which is called a knowledge map [2]. Generally a knowledge
map contains numerous KUs and learning dependencies as well
as complex topological properties, such as heterogeneous degree
distribution and high local density. For example, the graphs A
and B in Fig. 1 are the partial and global graphic representation
of the knowledge map of ‘‘Plane Geometry’’, respectively.
Knowledge maps have been employed in the navigational learn-
ing environment as a knowledge representation tool [1,2]. This
helps learners avoid disorientation by providing them a visual nav-
igation path [3]. However there have been few studies on the
inherent properties of a knowledge map. In this paper, we focus
on topological properties of knowledge maps. The main motivation
for this study has three aspects:
(a) Knowledge map is known to be correlated with a range of
human endeavors [4]. Accordingly, understanding the topo-
logical properties of knowledge maps may help gain better
insights into human cognition structure and its mechanism.
(b) Mining the learning dependencies among KUs is a prelimin-
ary and significant task. Taking the topological properties
into consideration provides new solutions to develop better
link-prediction algorithms. For example, we proposed a
learning-dependency mining algorithm for knowledge map
[5], by exploiting the property of locality of learning-depen-
dency, which will be discussed in Section 4.3. The learning-
dependency mining is a key step in the semi-automatic or
automatic knowledge map construction process.
(c) The topological properties of a knowledge map also help
improve the efficiency of a user’s navigational learning activ-
ity. For example, the centralities of knowledge units can be
used to guide the learners’ attention allocation in the naviga-
tional learning.
For this reason, we applied topological analysis to 12 knowledge
maps, including ‘‘C Language,’’ ‘‘Plane Geometry,’’ ‘‘Physics,’’ etc.
Our main contributions are as follows: First, we discovered
knowledge maps from computer science, mathematics and physics
show not only the small-world and scale-free properties exhibited
by many networks, but also the locality of learning dependencies
and hierarchical modular structure. Second, we studied the effect
on network efficiency by removing KUs based on different central-
ity measures, and find that the importance of KUs varies greatly.
The analysis of topological properties of knowledge map is conduc-
tive to the discovery of key factors in the cognition process.
0950-7051/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.knosys.2012.07.011
⇑
Corresponding author.
E-mail address: liukeen@mail.xjtu.edu.cn (J. Liu).
Knowledge-Based Systems 36 (2012) 260–267
Contents lists available at SciVerse ScienceDirect
Knowledge-Based Systems
journal homepage: www.elsevier.com/locate/knosys