1.9.2020 - 31.8.2023
Range on year:
1.60 FTE | 2021
prof.dr. Aleš Žiberna
The phases of the project and their realization:
The project entails several work packages (WPs), which follow the two main project objectives. The project work will be divided into WPs as shown below:
WP1 – Preparation phase
a. Literature review
b. Assessment of candidate approaches to blockmodeling dynamic networks for evaluation and a search for existing
implementations that are appropriate for simulations
c. Assessment of approaches to generating networks that may be used or adapted to generate dynamic networks for use in
WP2 – Development of algorithms for generating dynamic random networks
a. A detailed study of existing approaches
b. Proposals concerning adaptations needed to generate networks of a required type (known blockmodels and partition at each
time point) or entirely new algorithms
c. Software implementation of new algorithms into the R programming language
d. Evaluation of the algorithms in terms of their resemblance to actual network dynamics
e. Implementing these algorithms in a software package (R)
WP3 – Evaluations of blockmodeling algorithms through simulations:
a. Adaptation of the algorithm implementations for simulations
b. Design of the simulations
c. Evaluation of the results
d. Preparation of recommendations based on the results
WP4 – Application of appropriate blockmodeling approaches(s) to dynamic co-authorship networks of Slovenian researchers
a. Overview of literature on scientific collaboration
b. Operationalization of scientific collaboration
c. Preparation and selection of data
d. Preliminary analysis
e. Analysis with selected approach(es)
f. Preparation of reports
WP5 – General project management and dissemination
a. Project coordination
c. Reports, as required
d. The final report
Below, a more detailed description appears of individual WPs or, where more appropriate, groups of WPs.
The first work package (WP1) is relevant to all three objectives and may be seen as a preparation phase in the project for all things in common to the main project branches. The core focus in this work package is to be: (i) collecting all possibly relevant blockmodeling approaches for dynamic networks (and possible corresponding software implementations); and (ii) evaluating the approaches that can be adapted to generate dynamic networks with the properties described in the previous sections.
Within WP2, the existing algorithms for generating random networks will be studied in terms of possible adaptations for their use in Monte Carlo simulation studies. If needed, completely new approaches to generating such networks will also be proposed in WP2. The modified and/or proposed algorithms will be implemented as software in the R programming language (R Core Team, 2008). Since the purpose of the modified or proposed algorithms is to generate dynamic networks with properties that are commonly seen in the real-world social networks, the algorithms will also be evaluated in this sense.
WP3 will provide the main results of the proposed research project. Its chief goal is to evaluate different approaches to blockmodeling dynamic networks. In cases where the approach may be used with different parameters or different versions of the approach exist, these will be also evaluated. Depending on existing software implementations of the selected blockmodeling approaches (within WP1), the selected implementations will be adapted or linking components will be created if needed: (i) to enable the use of these approaches in Monte Carlo simulations (i.e. interface, type of input data, type of output data); and (ii) to ensure an appropriate level of unbiased comparisons between different blockmodeling approaches (i.e. amount of resources available). Parallel to the latter, the design of the simulations will be established. Here the factors whose effects are to be studied will be determined. Factors like size of the network (number of units), number of time points, stability of the network, partition and ties among groups, and density of the network will be considered. The data obtained with Monte Carlo simulations will be analysed and recommendations for the use of the analysed approaches will then be formulated.
In WP4, empirical data on scientific collaboration will be analysed by using one or more approaches to blockmodeling dynamic networks. This will include basic retrieval and preparation of the data, yet special attention will be paid to different types of operationalisations of scientific collaboration. Regarding this: (i) the impact of considering different types of publications on the blockmodeling solution will be discussed together with; (ii) the impact of different ways of binarizing co-authorship networks or analysing weighted networks. These are very important factors because different scientific disciplines have different publication cultures with respect to both the preferred type of scientific publication(s) and the number of authors (Kronegger et al., 2015). The analysis will be done at a minimum on the level of scientific disciplines for time periods after 1990. The blockmodeling results will be compared against those obtained by Cugmas et al. (2016). This WP is strongly interwoven with WP2 and WP3. On one hand, it will rely on the results of WP3 in terms of proposing which method to use. On the other hand, its preliminary analysis will be used to make the results of both WP2 and WP3 more relevant for co-authorship networks.
The final WP (WP5) represents general management of the project and is therefore active for the whole project duration. It serves as support for all the other WPs, especially in terms of coordination and dissemination of the results. The latter will occur through: (i) publishing the research results in international journals; (ii) active participation at conferences in the fields of social network analysis, statistics and sociology; (iii) publishing open-source packages with the implemented software for the R programming language; (iv) informing the general public about the research results on official social network channels (e.g. Twitter, Facebook, LinkedIn). The annual reports and final report will also form part of this WP.
Citations for bibliographic records:
BACKGROUND: Network analysis has become the main approach to analysing social interactions. A network is defined by the set of nodes (or vertices, units, actors) and by the set of links which represent ties between the nodes.
Recently, much attention has been devoted to the analysis of dynamic networks. We use the term “dynamic networks” to describe a set of networks with the same set of nodes (incomers and outgoers are allowed), observed at/in different consecutive time points/periods.
Blockmodeling is an approach for partitioning the nodes of a network (according to the structure of their links) and determining the ties among the clusters then obtained and thus may be used to describe the global structure of a studied network. Blockmodeling is widely used in the operationalization of social role and for data reduction.
Since most approaches to blockmodeling dynamic networks were developed only in the last decade, researchers usually blockmodel networks from different time points separately. However, this approach is inadequate because it assumes that the networks from the different points in time are independent. The dependency between links from different points in time can be taken into account by the approaches to the blockmodeling of dynamic networks. Such analysis allows for more accurate results, smoother changes in time and hence improves the study of these changes.
PROBLEM DEFINITION: Blockmodeling approaches for dynamic networks have not yet been systematically evaluated and compared using Monte Carlo simulations. Their evaluation is needed to help develop guidelines on the selection and use of blockmodeling approaches.
To perform such simulations, one needs to generate random dynamic networks based on local mechanisms with a specified blockmodel type and partition for at each point in time. Currently, none of the existing approaches is able to do this.
The approaches to blockmodeling dynamic networks should be very useful for studying scientific collaboration as operationalized by co-authorship (due to ‘noise’, lag in publications, etc.), and should make observing changes over time easier, namely, a very important issue in the social sciences.
RESEARCH OBJECTIVES: The project has three main objectives:
(i) To evaluate different blockmodeling approaches to dynamic networks that have different characteristics (e.g. density, nodes’ fluctuation, number of nodes at different time points, change of a blockmodel type in time).
(ii) To develop algorithms for generating dynamic random networks based on local network mechanisms that allow the specification of the partition and the global structure for each time point. Using local mechanisms is necessary so that the networks then generated are closer to the real-world network dynamics, while known partition and global structure is needed for evaluation of the blockmodeling results.
(iii) To apply one or several appropriate blockmodeling approaches to the dynamic co-authorship networks of Slovenian researchers. This should add to the validity of the results by way of more stable partitions and more easily observable changes.
Back to list of projects