Bots don’t Vote, but They Surely Bother! A Study of Anomalous Accounts in a National Referendum Eduardo Graells-Garrido Data Science Institute, Universidad del Desarrollo Santiago, Chile egraells@udd.cl Ricardo Baeza-Yates Institute for Experiential AI, Northeastern University California, USA rbaeza@acm.org ABSTRACT The Web contains several social media platforms for exchange of ideas and content publishing. These platforms are used by people, but also by distributed agents known as bots. Bots have existed for decades, with many of them being benevolent, although their infuence in propagating and generating deceptive information has increased recently. Here we present a characterization of the discus- sion on Twitter about the 2020 Chilean constitutional referendum. Through a profle-oriented analysis that enables the isolation of anomalous content using machine learning, we obtain a character- ization that matches national vote turnout, and we measure how anomalous accounts (some of which are automated bots) produce content and interact promoting (false) information. CCS CONCEPTS · Information systems → Social networks. KEYWORDS Social networks, bot detection, political polarization. ACM Reference Format: Eduardo Graells-Garrido and Ricardo Baeza-Yates. 2022. Bots don’t Vote, but They Surely Bother!: A Study of Anomalous Accounts in a National Referendum. In 14th ACM Web Science Conference 2022 (WebSci ’22), June 26–29, 2022, Barcelona, Spain. ACM, New York, NY, USA, 5 pages. https: //doi.org/10.1145/3501247.3531576 1 INTRODUCTION Social media platforms have acquired a crucial role in meaning- making processes within communities [11]. In the context of social changes and worldwide events, such processes have acquired more importance than ever. As technology evolves, the łsocialž has be- come more than just people: social platforms provide a myriad of services ranging from news, health, business, games, among oth- ers. The entities in these platforms are people, but also companies, political parties, and media sources of all sizes and credibility. Yet, not all accounts that pretend to be people are actual persons. Some of them are automated accounts. Although sometimes bots are benevolent [1], the last several years the focus of bots has been Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for components of this work owned by others than the author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specifc permission and/or a fee. Request permissions from permissions@acm.org. WebSci ’22, June 26–29, 2022, Barcelona, Spain © 2022 Copyright held by the owner/author(s). Publication rights licensed to ACM. ACM ISBN 978-1-4503-9191-7/22/06. . . $15.00 https://doi.org/10.1145/3501247.3531576 deceiving people by manipulating and amplifying social media content. This situation has promoted methods to detect and char- acterize bots [5, 21], as well as to understand their role in social interactions mediated by these platforms [18, 19]. In this paper we study the political discussion around the Chilean Constitutional Referendum, held on October 25th, 2020. This event was one of the consequences of the fercest social outburst in the last decades [20]. It started on October 18, 2019, and it is considered an important event that has impacted Chile’s well-being, due to a łperfect stormž of situations, including the recent pandemic [15]. One of the main demands of the social movements involved was a referendum to draft a new constitution for the country, because the core of the current constitution was drafted during Pinochet’s dictatorship. Thus, the plebiscite enticed strong and polarizing discussions on social media, particularly in the micro-blogging platform Twitter. Being publicly accessible, the trending topics of Twitter are part of everyday conversations and media reports. Given how social media can shape people’s perception, and how this perception can be tied to voting turnout, here we aim to understand the role of bots in the discussion. Mainly, we focused on the volume of content published by bots, their potential synchronization, and their political leaning. We applied an existing methodology for stance detection [9], which enabled us to classify Twitter accounts into in-favor or against a new constitution. Then, we applied an existing anomaly detection method, Isolation Forest [13], to quantify how anomalous each account was with respect to their behavior in the platform. We interpreted the global patterns of anomalous behavior, and then established a criteria to defne a bot. As a result, we observed that the stance classifcation produced results aligned with the election turnout; that the fraction of bots is small (0.66%) but their impact is much larger; and that, in terms of interaction and information difusion, there are bot communities in both sides of the political spectrum, yet the larger communities were right-leaning, against the drafting of a new constitution. 2 DATA We connected to the Twitter Streaming API using a system de- signed to crawl Chilean tweets. The query parameters were key- words related to mainstream political discussion in Chile, including keywords related to the two stances of the referendum: to approve (Apruebo in Spanish) the drafting of a new constitution, or to reject it (Rechazo in Spanish). We studied the period between August 1st, 2020, and October 25th, 2020. In total, we obtained 2.3M tweets from 251K users (see Figure 1) after a cleaning process. This repre- sents about 10% of all Twitter users in Chile 1 and about 1.3% of the 1 https://datareportal.com/reports/digital-2020-chile, p. 38. 302