Tracking Communities over Time in Dynamic Social Network
This poster paper presents an approach for tracking community structures. In contrast to the vast majority of existing methods, which are based on time-to-time consecutive evaluation, the proposed approach uses a similarity measure that involves the globa
- PDF / 1,294,719 Bytes
- 5 Pages / 439.37 x 666.142 pts Page_size
- 35 Downloads / 163 Views
2
Department of Computer Science, University of Sherbrooke, Sherbrooke, QC, Canada {etienne.gael.tajeuna,shengrui.wang}@usherbrooke.ca Department of Computer Science, University of Quebec at Montreal, Montreal, QC, Canada [email protected]
Abstract. This poster paper presents an approach for tracking community structures. In contrast to the vast majority of existing methods, which are based on time-to-time consecutive evaluation, the proposed approach uses a similarity measure that involves the global temporal aspect of the network under investigation. A notable feature of our approach is that it is able to preserve the generated content across different time points. To demonstrate the suitability of the proposed method, we conducted experiments on real data extracted from the DBLP.
Keywords: Community evolution structure
1
·
Similarity measure
·
Topological
Introduction
To understand the evolution of communities over time, several approaches have been proposed. Most of these approaches investigate the common nodes of two communities at consecutive time stamps ti and ti+1 using a Jaccard or modified Jaccard measure [1], [2]. However, as demonstrated in [3], at the end of lifespan such an approach may yield a community that does not share any nodes with the initially observed community. In fact, a tracking approach that considers only consecutive time points may not necessarily capture the overall temporal evolution of a community. For purposes of clarification, let’s look at the the evolution of community Ct11 from t1 to t4 in two different cases as presented in Fig. 1. In the first case (Fig. 1 (First case)) we have an evolution obtained from a simple one-to-one investigation of nodes, with the corresponding evolution of the bag of topics from Bt11 to Bt14 . As time evolves from time t1 to t4 , we can see how nodes initially observed in Ct11 gradually disappear as the topics are gradually change from the computer science field to the mathematic field. However, we can not say that the main topic of community Ct11 has gradually changed from social network analysis to boolean algebra due to the fact that individuals found in Ct14 may not share the same c Springer International Publishing Switzerland 2016 P. Perner (Ed.): MLDM 2016, LNAI 9729, pp. 341–345, 2016. DOI: 10.1007/978-3-319-41920-6 25
342
E.G. Tajeuna et al.
Fig. 1. An example of community and topic evolution.
interest as the individuals initially observed in Ct11 . In the second case (Fig. 1 (Second case)), we have an evolution from Ct11 to Ct24 , against the corresponding evolution of the bag of topics from Bt11 to Bt24 . We can see how some nodes initially observed in Ct11 persist over time, as topics remain in the computer science field due to the fact that individuals found in this evolution are all interested on social network analysis. The example in Fig. 1 suggests that consecutive tracking yields an inappropriate sequence which may inappropriately reflect the temporal evolution of a community for which the main topics may not be pre
Data Loading...
