SELP: Semi-supervised evidential label propagation algorithm for graph data clustering

Kuang Zhou (1) , Arnaud Martin (2) , Quan Pan (1) , Zhunga Liu (1)

1 NPU - Northwestern Polytechnical University [Xi'an]
2 DRUID - Declarative & Reliable management of Uncertain, user-generated Interlinked Data

With the increasing size of social networks in the real world, community detection approaches should be fast and accurate. The label propagation algorithm is known to be one of the near-linear solutions which is easy to implement. However, it is not stable and it cannot take advantage of the prior information about the network structure which is very common in real applications. In this paper, a new Semi-supervised clustering approach based on an Evidential Label Propagation strategy (SELP) is proposed to incorporate limited domain knowledge into the community detection model. The main advantage of SELP is that it can effectively use limited supervised information to guide the detection process. The prior information about the labels of nodes in the graph, including the labeled nodes and the unlabeled ones, is initially expressed in the form of mass functions. Then the evidential label propagation rule is designed to propagate the labels from the labeled nodes to the unlabeled ones. The communities of each node can be identified after the propagation process becomes stable. The outliers can be identified to be in a special class. Experimental results demonstrate the effectiveness of SELP on both graphs and classical data sets.

Mots clés: Semi-supervised learning; Label propagation; Theory of belief functions; Uncertainty; Community detection

Domaines: Intelligence artificielle [cs.AI]

Kuang Zhou, Arnaud Martin, Quan Pan, Zhunga Liu. SELP: Semi-supervised evidential label propagation algorithm for graph data clustering. International Journal of Approximate Reasoning, 2018, 92, pp.139-154. ⟨10.1016/j.ijar.2017.09.008⟩. ⟨hal-01668379⟩