Skip to Main content Skip to Navigation


Abstract : The detection of communities in social networks has become a very important task. Indeed, as its role consists in partitioning the nodes of a network into subgroups having properties in common, this makes it possible to analyse the behaviour of the entities of the network and to predict the evolution of the latter in time. In social networks, information about nodes, links, and messages may be imperfect. From there, the analysis of such a type of network necessitates the use of a theory of uncertainty. In this thesis, we propose three contributions applied in the framework of the theory of belief functions: First, we were interested in showing the advantage of using evidential attributes in social networks. Indeed, we compared the results of the classi fication of nodes with uncertain attributes (numerical, probabilistic, evidential) generated according to the structure of the network. To do this, we considered two scenarios: attributes generated randomly and others sorted. We also performed the tests in the case of data that was noisy. In order to measure the quality of clustering results, we used normalised mutual information (NMI). The second contribution consists on the correction of noisy information in social networks. To do this, we proposed a model based on the comparison of the calculated distances between the triplets of the network and the coherent triplets defi ned initially. A triplet is composed of two nodes connected to each other by a link. In order to test the proposed approach, we first tested three cases: only the nodes are noisy, only the links are noisy and finally the nodes and the links are noisy simultaneously. Then we tested the method by varying several network parameters. In order to measure the quality of the obtained results, we calculated the accuracy The third contribution is to detect which links are spammed in a social network. A link is considered spammed if its initial class changes according to the types of messages transiting on it. To do this, we used the theory of belief functions to combine the information of links and messages. In order to test our approach, we considered two cases: only the messages are noisy and the messages as well as the links are noisy simultaneously. The quality of the classi cation results was measured using accuracy, precision and recall measurements..
Complete list of metadatas

Cited literature [84 references]  Display  Hide  Download
Contributor : Salma Ben Dhaou <>
Submitted on : Wednesday, May 29, 2019 - 9:38:05 PM
Last modification on : Wednesday, March 11, 2020 - 1:16:40 AM


Files produced by the author(s)


  • HAL Id : tel-02144176, version 1


Salma Ben Dhaou. MODELLING INTERACTIONS BETWEEN NODES IN A CREDIBILIST SOCIAL NETWORK. Artificial Intelligence [cs.AI]. Institut Supérieur de Gestion de Tunis, 2019. English. ⟨tel-02144176⟩



Record views


Files downloads