Analysis of dynamics and credibility in social networks

According to the view of many researchers, today one of the greatest challenges on the net is the ability to discriminate the truth from falsehood, while moving along the mazy world of the Web and Online Social Networks. Many attempts have been made with this goal in mind; however, there are no shared solutions yet. The aim of this thesis is to propose a methodology to tackle this problem, starting from an in-depth analysis of the main interactions that occur within Online Social Networks. In fact, the analysis starts with a description of the different ways of interaction in Online Social Networks, to better explain why people massively use these instruments today. The focus is mainly on social network dynamics, with particular attention to the models of participation and to the reasons that push people to be connected. It is also very important to understand how these dynamics help the dissemination of information. A relevant part of this work is related to describing the information spreading techniques and phenomena. Then, the concept of credibility on the Web is furthermore investigated, with specific focus on Online Social Networks. Among the questions to be addressed, the most relevant ones are: What is true and what is false? Why do we trust some information instead of others? How much is our social network significant for trusting or untrusting some news? After explaining this social background and demonstrating how the social context can influence the perception of truth on the net, a model is proposed, with the aim to help users to estimate to what extent a piece of information can be considered true. The estimation is based on four aspects: (i) the credibility of the source that publishes it and the users who share it, including the reliability of the social relations of the source; (ii) the structure of the source site; (iii) the text used to spread the information, including the sentence structure and the used words; and (iv) the use of images. All these aspects have been dealt with, using different machine learning tech- niques. At a first stage, each aspect has been analyzed independently of the others. These different modules lead all to very promising results. A further step of analysis, which is modeled in this work, requires a composite system to put all the results together.