NLP and Public Engagement: The Case of the Italian School Reform

NLP and Public Engagement: the Case of the Italian School Reform Tommaso Caselli∗, Giovanni Moretti3, Rachele Sprugnoli3♮, Sara Tonelli3, Damien Lanfrey♣, Donatella Solda Kutzmann♣ Vrije Universiteit Amsterdam∗, FBK3, University of Trento♮, MIUR♣ De Boelelaan 1105 1081 HV Amsterdam NL, Via Sommarive, 18 38122 Povo, Viale Trastevere, 76/a 00153 Roma t.caselli@vu.nl, {moretti;sprugnoli;satonelli}@fbk.eu, {damien.lanfrey,donatella.soldakutzmann}@istruzione.it Abstract In this paper we present PIERINO (PIattaforma per l’Estrazione e il Recupero di INformazione Online), a system that was imple- mented in collaboration with the Italian Ministry of Education, University and Research to analyse the citizens’ comments given in #labuonascuola survey. The platform includes various levels of automatic analysis such as key-concept extraction and word co-occurrences. Each analysis is displayed through an intuitive view using different types of visualizations, for example radar charts and sunburst. PIERINO was effectively used to support shaping the last Italian school reform, proving the potential of NLP in the context of policy making. Keywords: Text Mining, Information visualization, E-government 1. Introduction The engagement of citizens in policy making has become an important challenge for public administrations in the last years since current digital technologies could be adapted to promote forms of direct democracy by means of online sur- veys, asking for the citizens’ opinion on key issues related to public life. However, the full potential of state-of-the- art technologies in collecting and analyzing such data has not been exploited yet. Online surveys limited to yes/no questions are somehow easily processed but more elabo- rated questionnaires, involving idea gathering and propos- als collection from citizens, even if easily implemented through online forms, raise costs for an effective analysis. Indeed, successful public consultations - especially on top- ics high on the agenda, can end up with thousands of free- text answers, which cannot be analyzed manually. To make the engagement process effective, reward citizens’ efforts, and produce useful outcomes for the administration, such amount of information needs to be automatically processed to extract content that can support civil servants and politi- cians in taking decisions. Besides, the extracted content must be displayed in an intuitive way, since it should be also accessed by non-technical users. Research in the automatic analysis of open answers has mainly focused on data from the marketing and advertis- ing fields (Yamanishi and Li, 2002) with few notable stud- ies in the area of e-government (Ku and Leroy, 2014). The literature reports the use of different techniques. For ex- ample, supervised approaches (Giorgetti et al., 2003) and active learning (Patil and Ravindran, 2015) are employed to perform survey coding, that is matching open-ended an- swers with a short description associated to a code, thus converting qualitative information (text) into a quantitative format (code). In addition, text clustering and topic model- ing methods are applied to explore information contained in open answers, e.g. to summarize and classify them (Wang and Mulrow, 2014) and to identify the intentions of survey participants (Rosell and Velupillai, 2008). In this work, we present the first attempt to apply state- of-the-art Italian Natural Language Processing (NLP) …

Content truncated to 500 words