From Changes to Dynamics: Dynamics Analysis of
Linked Open Data Sources
Renata Dividino1 , Thomas Gottron1 , Ansgar Scherp2 , and Gerd Gröner3
1
WeST – Institute for Web Science and Technologies
University of Koblenz-Landau, Germany
{dividino,scherp,gottron}@uni-koblenz.de
2
Kiel University and Leibniz Information Center for Economics, Kiel, Germany
{ansgar}@informatik.uni-mannheim.de
3
Paluno – University of Duisburg-Essen, Germany
{gerd.groener}@paluno.uni-due.de
Abstract The Linked Open Data (LOD) cloud changes frequently. Recent ap-
proaches focus mainly on quantifying the changes that occur in the LOD cloud
by comparing two snapshots of a linked dataset captured at two different points in
time. These change metrics are able to measure absolute changes between these
two snapshots. However, they cannot determine the dynamics of a dataset over a
period of time, i.e., the intensity of how the data evolved in this period. In this
paper, we present a general framework to analyze the dynamics of linked datasets
within a given time interval. We propose a function to measure the dynamics
of a LOD dataset, which is defined as the aggregation of absolute, infinitesimal
changes, provided by change metrics. Our method can be parametrized to in-
corporate and make use of existing change metrics. Furthermore, our framework
enables the use of different decay functions within the dynamics computation for
different weights on changes depending on when they occurred in the observed
time interval. We apply our framework to conduct an investigation on the dynam-
ics of selected LOD datasets. We apply our analysis on a large-scale LOD dataset
that is obtained from the LOD cloud by weekly crawls over more than a year. Fi-
nally, we discuss the benefits and potential applications of our dynamics function
in a real world scenario.
1 Introduction
The Linked Open Data (LOD) cloud is a global information space to structurally re-
present and connect data. The LOD principles provide a flexible publishing paradigm
to integrate and interlink any kind of data from arbitrary datasets, published by various
data providers. From the time the Linked Open Data principles have been created until
now, the LOD cloud has grown significantly and is a place of continuous changes.
Knowledge about these changes and especially about the change behavior of a
dataset over time, i.e., the dynamics of a dataset, is important for many purposes and
applications involving Linked Data such as data caching [18], indexing of distributed
data sources [13], searching in large graph databases [9], optimizing the execution of
queries [14] and recommending appropriate vocabularies to Linked Data engineers [16].
2
For example, as the data changes, caches and indexes that rely on this data need to be
updated since they do no longer reflect the current state of the data anymore. Umbrich
et al. [18] proposed a hybrid query execution engine that takes into account the knowl-
edge if a dataset is rather static or dynamic in order to automatically decide whether
data is retrieved from caches or from the LOD cloud. Caches are created only for data
from static datasets. Therefore, knowledge about the level of dynamics for different
data sources is vital to make best use of the resources available for computing caches or
performing index updates (e. g., network bandwidth for crawling, computation time).
In related work, changes of LOD sources are analyzed w.r.t. their sets of triples,
sets of links, sets of entities, or schema signatures. For example, given two snapshots
of a dataset captured at two different points in time, the change analysis at the triple
level includes which triples from the previous snapshot have been preserved in the later
snapshot, which triples have been deleted, or which ones have been added. For instance,
in [6] snapshots of a dataset were analyzed with respect to their set of domain entities,
i.e. it was verified if the set of entities described in the datasets has changed. In [7], the
authors measure changes with respect to usage of the schema information of a dataset.
Käfer et al. [12] quantify changes w.r.t the set of triples, set of links, schema signature.
While these kinds of analyses are capable to quantify changes of a dataset captured at
two different points in time, they do not really grasp the dynamics of a dataset.
The dynamics of a dataset involves a notion of how “fluid” a dataset is, i. e., how
it behaves and evolves over a certain period of time. In the context of this paper, we
understand a period of time to be a continuous time interval beginning at an initial point
in time up to a final one. Therefore, the dynamics of a dataset involves the analysis of its
development over more than two points in time. Due to this time-dependence a measure
for dynamics should capture the frequency, degree and regularity of the changes of the
data. To the best of our knowledge, there is no established method for measuring the
dynamics of LOD datasets.
To fill this gap, we present a formal notion of dynamics for LOD datasets. First,
we define dynamics as an aggregation of changes, built on top of contemporary change
metrics. Then we extend this notion to incorporate the use of different decay functions
for stressing or weakening periods within a time interval. Finally, we analyze the dy-
namics of different LOD datasets obtained via weekly crawls from the period between
May, 2012 and November, 2013. We compute the dynamics of these datasets and com-
pare their change behavior over time. The notion of dynamics has benefits and potential
impact in different real world scenarios, which we discuss before concluding the paper.
2 Motivating Scenario
Let us introduce a toy example to illustrate the differences between change and dynam-
ics analysis on LOD datasets. In this example, we describe three snapshots of a dataset
captured at three distinct points in time t1 , t2 , and t3 . We are using the FOAF vocabulary
for describing persons working at the University of Koblenz-Landau, the University of
Kiel, and the University of Duisburg, all of them located in Germany. In addition, we de-
scribe relations between persons and their associations to different projects. Besides the
FOAF vocabulary, we use domain-specific LOD vocabularies under the domain of uni-
3
Table 1. Scenario: Example dataset at time t1 .
@prefix uni-koblenz: .
@prefix foaf: .
@prefix foaf: .
@prefix foaf: