Learning Methods and Algorithms for Semantic Text Classification across Multiple Domains

Information is nowadays a key resource: machine learning and data mining techniques have been developed to extract high-level information from great amounts of data. As most data comes in form of unstructured text in natural languages, research on text mining is currently very active and dealing with practical problems. Among these, text categorization deals with the automatic organization of large quantities of documents in priorly defined taxonomies of topic categories, possibly arranged in large hierarchies. In commonly proposed machine learning approaches, classifiers are automatically trained from pre-labeled documents: they can perform very accurate classification, but often require a consistent training set and notable computational effort. Methods for cross-domain text categorization have been proposed, allowing to leverage a set of labeled documents of one domain to classify those of another one. Most methods use advanced statistical techniques, usually involving tuning of parameters. A first contribution presented here is a method based on nearest centroid classification, where profiles of categories are generated from the known domain and then iteratively adapted to the unknown one. Despite being conceptually simple and having easily tuned parameters, this method achieves state-of-the-art accuracy in most benchmark datasets with fast running times. A second, deeper contribution involves the design of a domain-independent model to distinguish the degree and type of relatedness between arbitrary documents and topics, inferred from the different types of semantic relationships between respective representative words, identified by specific search algorithms. The application of this model is tested on both flat and hierarchical text categorization, where it potentially allows the efficient addition of new categories during classification. Results show that classification accuracy still requires improvements, but models generated from one domain are shown to be effectively able to be reused in a different one.

Essays in Empirical Political Economics

This thesis takes two perspectives on political institutions. From the one side, it examines the long-run effects of institutions on cultural values. From the other side, I study strategic communication, and its determinants, of politicians, a pivotal actor inside those institutions. The first chapter provides evidence for the legacy of feudalism - a set of labor coercion and migration restrictions -, on interpersonal distrust. I combining administrative data on the feudal system in the Prussian Empire (1816 – 1849) with the geo-localized survey data from the German Socio-Economic Panel (1980 – 2020). I show that areas with strong historical exposure to feudalism have lower levels of inter-personal trust today, by means of OLS- and mover specifications. The second chapter builds a novel dataset that includes the Twitter handles of 18,000+ politicians and 61+ million tweets from 2008 – 2021 from all levels of government. I find substantial partisan differences in Twitter adoption, Twitter activity and audience engagement. I use established tools to measure ideological polarization to provide evidence that online-polarization follows similar trends to offline-polarization, at comparable magnitude and reaches unprecedented heights in 2018 and 2021. I develop a new tool to demonstrate a marked increase in affective polarization. The third chapter tests whether politicians disseminate distortive messages when exposed to bad news. Specifically, I study the diffusion of misleading communication from pro-gun politicians in the aftermath of mass shootings. I exploit the random timing of mass shootings and analyze half a million tweets between 2010 – 2020 in an event-study design. I develop and apply state-of-the-art text analysis tools to show that pro- gun politicians seek to decrease the salience of the mass shooting through distraction and try to alter voters’ belief formation through misrepresenting the causes of the mass shootings.

A double chance to investigate populist influence: The Five Star Movement.

In the last decade, new kinds of European populist parties and movements characterized by a left wing, right wing or “eclectic” attitude have succeeded in entering in governments where they could exert a direct populist influence on their coalition partners or, conversely, become victims themselves of the influence of the institutional background. Such a scenario brought this research to formulate two questions: (i) “To what extent did populist parties succeed in influencing their government coalition partners, leading them to adopt populist rhetoric and change their policy positions?” and (ii) “Have populist parties been able to retain their populist “outside mainstream politics” identity, or have they been assimilated to mainstream parties?”. As a case study this project chose the Italian Five Star Movement. Since 2018 this eclectic populist actor has experienced three different governments first with the radical right wing populist League (2018-2019) and then with the mainstream center left Democratic Party (2019-2021). In addition to this, currently the Five Star Movement is a coalition partner of the ongoing Draghi’s government. Theoretically based on the ideological definition of populism (Mudde, 2004), on a new “revised” model of the inclusionary - exclusionary framework to classify populist parties and on a novel definition of “populist influence”,this research made use of both quantitative (bidimensional and text analysis) and qualitative methods (semi-structured interviews) and mainly focuses on the years 2017- 2020.The importance of this study is threefold. First it contributes to the study of populist influence in government in relation to the ideological attachment of the political actors involved. Second, it contributes to understand if populists in power necessarily need to tone down their anti-system character in order to survive. Third, this study introduces conceptual and methodological novelties within the study of populism and populist influence in government.