Analysis of political influence from Italian Senate voting data

Over the recent years the interest in the study of social systems by means of techniques drawn from the areas of control theory and machine learning domain has risen significantly. In particular, the analysis of social influence and opinion dynamics is one of the most representative in this field of study. Inspired by an existing and solid research active on the U.S. Congress, this thesis aims to apply these techniques to the voting data of the XVII legislature of the Italian Senate. This has been proven challenging for the higher number of both political members and parties involved in the Italian case. The goal is to derive an index of political influence exerted by the political groups composing the Senate on each individual Senator active during the legislature, which has been defined her Political DNA (Political Data-aNalytic Affinity). This is achieved firstly by acquiring all the relevant data from the public platform Openpolis. A set of learning techniques is then applied on this dataset to extract the most significative information, focusing in particular on the sparsity requirement which allows to also obtain a certain degree of interpretability. A probabilistic framework is then adopted to derive what we have previously defined as the Political DNA of a Senator. Finally, a set of clustering techniques is proposed to obtain a political grouping of the Italian Senators which is compared to the nominal one and the most significative cases of outliers are illustrated.