Over a period of analysis from January 21st, 2019 to March 7th, 2019, Constella’s team of data scientists conducted an analysis of the public, digital sociopolitical conversation in Italy across multiple digital sources such as Twitter, Facebook, YouTube, Instagram, blogs, forums, and other digital communities. The goals of the analysis were:
- To identify key communities of opinion, the users that make up these communities, and the issues addressed by these users as a means to identify any abnormal behavior that might signal or point to disinformation or disinformation strategies with a special focus on those strategies aimed at artificially increasing the propagation of polarizing messages and content.
- To identify users who demonstrated abnormally high activity during the period of analysis and the impact of their activity across the broader network of users.
Through this analysis, our team attained the following key insights:
- Our analysts continue to observe a high level of polarization in the public digital conversation related to the EU elections. Detailed content analyses yield several examples of sites propagating disinformation aimed at influencing the public debate around specific issues.
- An analysis of abnormalities helped to identify 240 (0.12%) accounts with abnormally high activity that generate at least 42 and up to 168 posts per day. In aggregate, this group produced a total of 718,332 (11%) posts over the period analyzed. This finding confirms the previously identified patterns of a small number of users (around 0.1%) creating over 10% of the observed public results. These accounts, as identified in other analyses across Europe, demonstrate a high affinity of interaction with specific themes and communities of users — in this case the Right and Anti-Immigration, Partito Democratico, and Movimento 5 Stelle.
- These abnormal activity users are concentrated primarily in the following communities: Right and Anti-Immigration (42% of users, 44% of comments), Partito Democratico (34% of users, 31% of comments), and Movimento 5 Stelle (22% of users, 23% of comments).
- Although the Right and Anti-Immigration community makes up 18% of users in the overall network analyzed, their presence among the abnormal high activity users is considerably greater, representing 42% of abnormal high activity users. This pattern is also visible in the case of Movimento 5 Stelle – although 8% of users in the overall network are identified in this community, users from this community make up 22% of abnormal high activity users.
- Media plays a key role in the digital debate as it feeds and reinforces the narratives of the different communities. Consistent with patterns seen across other countries analyzed, relevant content in the general debate in addition to content most shared by abnormal high activity users bear a xenophobic, anti-immigration discourse. A considerable amount of this content is produced and propagated by emerging sites with highly segmented editorial strategies, heavily focusing on matters such as Islam and alleged cases of violence by immigrants. Some of the identified sites became inaccessible in early March and although they were taken offline, they did achieve significant relevance over the period analyzed.
Communities Detection Analysis
Constella’s data scientists used Analyzer, our proprietary cloud-based analytics platform, to build and identify the largest communities of users interacting in connection with the Italian public debate. This community was identified in Twitter, and their interactions were filtered for retweets only, in order to perform a topological analysis of the propagation of their messages and content. This resulted in a unimodal network of authors (giant component, edges being re-tweets). To determine the different communities and visually represent the resulting network, Constella’s data scientists applied proprietary clustering algorithms, similar to the Louvain Method for community detection. The visualization below represents the most relevant 195,504 users and the 4,587,724 interactions produced by these users.
Network topology is characterized by network structures with nodes (profiles) and edges (lines) that connect them. Our algorithms cluster nodes and edges in order to calculate proximities and communities of users.
The following five key communities emerged from the community detection analysis:
- Partito Democratico (46% of users, 44% of retweets): Criticism of the current government and its policies. Another main theme involves conversations regarding electing a new Secretary for the party.
- Criticism against Salvini (20% of users, 4% of retweets): Community based around the criticism of Salvini with respect to immigration and the debate surrounding the Sanremo music festival in which the winner of the festival was part-Egyptian.
- Right and Anti-Immigration Group (18% of users, 35% of retweets): Different immigration-related topics in Italy with a focus on illegal immigration and the alleged crimes perpetrated by immigrants.
- Movimento 5 Stelle (M5s) (8% of users, 15% of retweets): Focuses on the work of M5S in the government and supporting the party’s objectives. There are also some conversations focused on immigration issues, infrastructure development, and reddito di cittadinanza.
- Venezuelan Political Crisis (8% of users, 2% of retweets): Debate around the political party, Forza Italia, which criticized the government of Italy for not taking a strong stance against Nicolas Maduro, the President of Venezuela.
Abnormality Detection – High Activity Users
Our analysts identified a group of 240 (0.12%) users demonstrating abnormally high activity. These users produced a total of 718,332 (11%) posts over the period of analysis – meaning at least 42 and up to 168 posts per day. These abnormal high activity users are found to be concentrated primarily in the following three communities: Right and Anti-Immigration (42% of users, 44% of comments), Partito Democratico (34% of users, 31% of comments), and Movimento 5 Stelle (22% of users, 23% of comments).
High Activity Users Generate Significant Activity
Our team used the 240 abnormal high activity accounts detected as an initial seed to recreate the resulting extended network which includes any users that interacted with them over the period analyzed. Interactions can be replies, retweets, or mentions between any of the users.
The abnormal high activity users manage to attract and engage a significant volume of other users. The resulting extended network includes 71,351 users interacting with the abnormal high activity users (through retweets, replies, or mentions). These 71,351 users produce 1,308,540 total interactions over the period analyzed, demonstrating that the high activity users manage an effective pattern of driving the conversation beyond solely retweets and thus amplifying engagement across the public debate. In each view, the network below represents unique retweets, mentions, and replies among users interacting with the abnormal high activity users in the analysis. Users interacting via multiple forms of engagement with the abnormal high activity users, or any combination of retweets, mentions, or replies, make up the remainder of the interactions, or those not counted as unique interactions.
Abnormality Detection – Account Creation Dates
In order to evaluate whether there is any pattern between the abnormal high activity users and the dates on which their accounts were created, Our analysts mapped the creation dates of the high activity accounts. In other analyses ahead of the EU Elections, our team identified a trend of relatively recent creation among the high activity accounts, with a notable proportion of such accounts being created within the past one to three years. In the case of Italy, our analysts identify no such trend present over the period analyzed, although there does appear to be an uptick in the creation dates of high activity accounts behaving with affinity to the Partito Democratico since 2018.
The Role of Emerging Sites – Disinformation Content from the Top 50 Most Relevant Domains Shared by High Activity Users
Our analysis identified a growing network of emerging digital media with highly segmented and specific editorial strategies that ranked high in influence among the high activity users identified. Among the top 50 most relevant sites shared by the high activity users, several were identified to have produced or published potential disinformation content over the period analyzed. Several of these domains demonstrate characteristics identified in Constella’s previous analyses that are indicative of sites sharing disinformation. These characteristics include but are not limited to high penetration in their key communities of interest and extremely segmented content focused on specific polarizing themes. Domains such as imolaoggi.it (8th), voxnews.info (11th), ilprimatonazionale.it (16th), and riscattonazionale.org (27th) are examples of relevant sites that were shared with high frequency and achieved substantial impact over the period analyzed. Sites such as cronacapiu.it (34th) and cronacaora.it (49th), for example, became inaccessible in early March, plausibly because they were taken offline – although they did achieve substantial relevance over the period analyzed. The table below shows some of the relevant sites identified in the analysis. Rank indicates the level of influence of the site or domain, the percentage represents to which point the content hosted on that site propagates through the different communities earlier identified and described. A feature of sites with an increased probability of being part of a disinformation campaign, although not the only or definitive feature, is that their content tends to propagate with very high intensity across only one or two communities of users.
High Activity Users – Examples of Most Shared Content
Below are examples of some of the most frequently shared content among the identified abnormal activity users. A consistent narrative across this content by the abnormal high activity users is immigration and the alleged criminality of immigrants. Several of these pieces of content cite specific cases which involve immigrants, violence, and unlawfulness in a clear effort to reframe the conversation on immigration along the lines of public safety and crime.
Over a period of analysis from December 2018 to January 2019, our team conducted a previous public digital sphere analysis of the Italian sociopolitical conversation. Read the analysis here: Italy: Digital Public Debate Ahead of EU Parliamentary Elections
Interested in our work? Please contact us at firstname.lastname@example.org. To learn more about Constella, subscribe to our newsletter below.