Opinion Mining relays on various techniques such as natural language processing, information retrieval, and structured and unstructured data mining. Opinion Mining process involve three main steps which are opinion retrieval, opinion classification and opinion study of opinion dynamics is crucial in nowadays-ubiquitous on-line system context. Researchers with different background have presented several models to study the opinion formation, propagation and aggregation from different points of view.

We explore the combination of classical opinion mining methods with complex networks analysis and its impact on opinion formation and propagation in order to build a consistent opinion model.In order to study the impact of influential users (influential nodes) on the opinion formation and propagation, we firstly integrate several influencing factors extracted from the network in the opinion mining process. These factors are usually computed by using different centralities measures as degree, closeness, betweenness and page rank centralities, etc. In addition and in order to understand how opinions interact and are influenced by each other’s through the structure and how the properties of complex network contribute to this process, we propose a preliminary method for aggregating opinions in order to converge to a consistent opinion model for  groups of users/actors that belong to particular types of subnetworks. These subnetworks are either egocentric networks around influencers or communities obtained by applying several algorithms of community detection.We then define and study the notion of opinion stability inside the egocentric networks around influencers and inside the communities, and we target to detect the opinion modification for both types of subnetworks. The opinion stability means the fact of sharing a majority of common preferences concerning a given topic in a group of users.

We aim first to observe how the opinion can be stable in egocentric networks and what are the factors that help to detect a modification of the opinion. We thus study the variation of opinion according to several nodal and topological measures as the influencers (egos) centralities, the link weights and the geodesic distance from the influencer, etc.Likewise, we study the opinion stability in communities in function of several global and local measures as the density, the coefficient of clustering and the average of centralities measures, etc. We observe the opinion modification specially for boundary nodes between the communities. This helps us to appreciate how the community structure of the network contributes to expand or limit the opinion propagation. We experiment our study on online social networks as twitter and on different blogging systems. For example, in online shopping websites, the users' preferences are created from their connections through evaluating their opinions and then aggregate them to a compact opinion to deduce a common rate. In this case, the entities nodes are messages or users and the interactions among the entities are relationships as reply-to ones.