The Italian backbone: from talk-show hashtags to coordinated networks
The empirical spine of this topic is a sequence of studies that track Italian electoral communication across three very different media configurations. Iannelli2015-e0818c3e opens the arc in 2013, when Twitter was still parasitic on television: audiences used official talk-show hashtags to comment on and request interaction with broadcast politics, but rarely to reframe the agenda, and politicians largely failed to reciprocate. By 2018 the locus of contestation has shifted to Facebook and the cross-platform news ecology, and Giglietto2019-882f1900 shows how populist parties’ supporters — especially the Five Star Movement — cultivate more insular news-sharing communities whose engagement patterns (shares vs. comments) function strategically, amplifying favorable stories and contesting unfavorable ones. This insularity finding is the first empirical anchor for what the programme comes to call computational political parallelism: rather than measuring outlet ideology in the abstract, these papers infer partisan alignment from the topology of sharing and engagement itself.
The next methodological turn is behavioral rather than attitudinal: Giglietto2020-9d8acdd7 reframes disinformation detection around coordinated link sharing behavior, arguing that near-simultaneous, repeated sharing of the same URLs by networks of pages and groups is a more tractable signal than content veracity or actor identity, and demonstrates its diagnostic power across the 2018 and 2019 Italian campaigns. Giglietto2023-fa71a001 operationalizes this insight as a living workflow for the 2022 snap election, iteratively expanding seed lists of coordinated accounts to surface an M5S-aligned hyperpartisan network, a clickbait operation cloaked in religious pages, and a proselytizing bot network — demonstrating that actor- and behavior-centred detection generalizes across ideological, commercial, and religious motives. Read together, these two papers mark the shift from measuring partisan attention to detecting the infrastructures that manufacture it.
The most recent instalment in the Italian sequence, Giglietto2025-1765bb4f, turns the lens on the platform itself: using the DSA-mandated Meta Content Library, it shows that Meta’s political-content-reduction policy silently suppressed Italian parliamentarians’ Facebook reach months before its announced global rollout, and that mainstream politicians lost visibility asymmetrically relative to extremist accounts, which compensated through posting volume. This paper closes the loop between the earlier partisan-attention and coordinated-network studies and a newer concern of the programme — that platform governance decisions, not just partisan strategy, now co-produce the visibility landscape that structural analyses of “parallelism” must account for.
Building the measurement toolkit on Italian text
A parallel cluster of papers treats the 2018 and 2022 Italian corpora as a testbed for computational methods rather than as substantive findings in themselves. Giglietto2024-cbeb3f70 benchmarks OpenAI’s text-embedding-3-large against the Italian-specific UmBERTo model for unsupervised clustering of political news, finding the general-purpose LLM embedding consistently more semantically coherent — a result that feeds directly into Marino2024-2fbc690f, which documents the practical and epistemic challenges of building a full “LLMs-in-the-loop” pipeline (classification, embedding, cluster labeling) for the same two elections, and argues that validating such pipelines requires expert rather than crowdsourced annotation. Paci2025-ag pushes the linguistic side of this toolkit further, testing whether LLMs can interpret the implicatures and presuppositions pervasive in Italian political speech (via the IMPAQTS corpus) and finding a persistent gap between model and expert interpretation — a sobering check on how much can be automated in the pragmatic reading of political discourse that underlies parallelism measures. Mosca2026-yh complements these text-level tools with a rhetorical-strategic reframing: fake-news accusations in Italian newspapers are analyzed not as misinformation symptoms but as tools of negative campaigning subject to “selective amplification” by aligned outlets, tying the programme’s computational detection work back to a normative theory of media-political strategy. Ducci2022-10cb5d70 and Sbaraini-Fontes2026-cw extend the same gatekeeping logic beyond electoral politics narrowly defined — respectively to Google News’ curation of the health agenda and to Italian audiences’ differential trust in AI-mediated search versus AI-produced news — underscoring that the Italian case functions in this programme as a general laboratory for studying platform-mediated gatekeeping, not solely an electoral one.
Is Italy exceptional? Comparative and cross-national tests
A second, larger cluster asks whether the dynamics documented in Italy — insularity, hyperpartisan amplification, coordinated networks, disproportionate visibility for populist and radical actors — are Italian idiosyncrasies or general features of contemporary electoral communication. Rossi2023-847d5a9f directly triangulates Italy against Germany and France using Meta’s URL Shares Dataset, finding that the share of untrustworthy URLs rose sharply in Italy’s and Germany’s election years while views from untrustworthy sources stayed comparatively stable — a nuance that complicates simple narratives of platform-driven misinformation surges. Balluff2026-bv extends the “media capture” logic underlying Italian party-press parallelism to Austria’s Inseratenaffäre, showing with a difference-in-differences design that covert advertising arrangements manifest as increased visibility and more negative coverage of competitors even in an ostensibly democratic-corporatist system — a mechanism-level parallel to the insularity/amplification dynamics found in Italian populist media ecologies. Balluff2026-ev complements this with a German case on Nord Stream 2 coverage, finding that alternative and legacy media reference largely the same entities but organize them into structurally distinct (more cohesive vs. more modular) networks — a reminder that parallelism can be relational and topological rather than simply lexical or attitudinal, echoing the network-structure findings in the Italian coordinated-behavior papers.
At the EU scale, Iris2026-pg and Darius2026-xl generalize the Italian populist-amplification finding across the whole continent: both document that radical-right and populist actors receive disproportionate media and social-media visibility relative to their electoral strength, with Darius2026-xl pinpointing TikTok as the platform of maximal asymmetry and Iris2026-pg showing the effect intensifies in the final campaign weeks and persists even absent domestic radical-right representation (Ireland). These findings recast the Italian M5S/insularity story as an instance of a broader “populist visibility premium” rather than a national peculiarity. Kalsnes2025-zb and Oprea2025-lf narrow back to platform-level mechanics — Scandinavian parties’ differential success at eliciting “Angry” reactions that drive sharing, and Romanian parties’ reliance on hyperactive/inauthentic Facebook accounts to inflate visibility during the 2024 EP campaign — both of which read as functional analogues to the Italian coordinated-link-sharing and insularity mechanisms, just instantiated through different affordances (emotional reactions, hyperactive sharing) and national contexts. Nizzoli2020-cf and Barbera2015-je supply methodological infrastructure that the Italian papers implicitly draw on: coordination as a continuous rather than binary property of networks (UK 2019 election), and ideal-point estimation from following/sharing structure as a general technique for scaling political actors and audiences (US and European legislatures) — both foundational to the “computational measures of political parallelism” this whole topic is built around. Eady2025-vm pushes this ideology-scaling logic furthest, showing on US Twitter data that ordinary politically engaged users, not politicians, drive the bulk of polarized news sharing, and that electoral competitiveness predicts how extreme legislators’ sharing becomes — a finding that complements the Italian insularity results by locating the demand side of partisan amplification, not just its supply.
Finally, several US and cross-national papers stress-test the causal weight of these platform dynamics. Allcott2025-jb’s large field experiment finds that removing political ads from Facebook/Instagram feeds for six weeks before the 2020 US election produced no detectable attitudinal effects, a useful counterweight suggesting that visibility and exposure findings documented elsewhere in this topic (e.g., Giglietto2025-1765bb4f, Iris2026-pg) should not be over-interpreted as straightforward persuasion effects. Giglietto2026-632ef967 and McNally2025-dn instead locate the causal action upstream, in platform governance itself: the former shows that Facebook’s sharing-to-views amplification is dampened for partisan audiences and boosted for high-quality journalism in ways that fluctuate with known governance interventions (echoing the policy-driven suppression found for Italian MPs), while the latter uses Guardian engagement data to show that News Feed ranking changes produce detectable, lagged effects specifically on hard news — evidence, across both papers, that “the algorithm” is an active, auditable political actor rather than a neutral conduit. Lukito2026-il and Much2026-gu extend the comparative frame to alternative platform ecologies (Truth Social cross-platformization by right-leaning US outlets; podcast-based “gendered media spaces” in the 2024 US election), signaling that the parallelism concept developed for Italian Facebook/Twitter data must eventually travel to newer platforms with very different content logics.
Infrastructure, access, and the future of the programme
A final set of papers turns reflexive, asking whether this entire research programme remains viable as platforms restrict data access. Philipp2026-tl and Lukito2026-nb survey the post-API landscape shaped by the EU Digital Services Act, cataloguing the uneven, often ephemeral access regimes (APIs, portals, vetted-researcher schemes) that now govern what election research is even possible, while Schulte2026-df documents the concrete bottlenecks of multi-platform data collection using the 2025 German election, arguing that platform observability itself should be treated as reusable research infrastructure. Gaisbauer2025-by adds a conceptual critique from within: existing partisan-news-sharing research (including much of the Italian corpus) tends to flatten political leaning into a single dimension and to examine only the outlet level, and proposes a multi-level “political cartography” as a corrective. Methodological papers such as Le-Mens2025-qz (LLM-based ideological scaling), Arora2025-tx (multi-modal framing analysis), and Achmann-Denkler2026-lx (multimodal LLMs for visual political communication) point toward where this toolkit is heading next — beyond text and link-sharing graphs toward images, video, and generative-model-assisted measurement — suggesting that the Italian election studies at this topic’s core are best understood not as a closed case study but as the proving ground for a measurement programme whose methods, and whose data-access preconditions, are still very much in flux.