Daejeon, July 27–31
In the recent book “Zerstörungslust” (“Urge for Destruction”), Carolin Amlinger and Oliver Nachtwey describe the global rise to power of the new right in Europe. Individuals unhappy with liberal democracy and with the dismantlement of traditional hierarchies are developing “an affective negation of inclusive liberalism”, which is based on “aggressive nostalgia” for an uncertain time in the past when everything was still intact (Amlinger and Nachtwey 2025, 9, own translation). (Amlinger and Nachtwey 2025, 9, own translation).
The MAGA movement exemplifies this form of nostalgia. Political philosophers Cíbik and Lukić highlight the temporal logic embedded in its central slogan: whereas “Make America Great” implies a forward-looking political project, “Make America Great Again” “relies on the shining attributes of past greatness, real or imagined. Its rhetorical power does not come from a future vision but from present discontent” (2025, 2). This illustrates how nostalgic political discourse selectively uses collective memory as a tool to motivate and unite a community in the present.(2025, 2). This illustrates how nostalgic political discourse selectively uses collective memory as a tool to motivate and unite a community in the present.
The connection of recent right-wing terrorist attacks to social media usage (Collins 2025) highlights the urgency of examining forms of nostalgia and their circulation in so-called alt-tech platforms, in particular, the platform Truth Social (TS). Alt-right nostalgia is a form contested digital memory (Hoskins 2018), especially in this case: TS was launched in February 2022, following Donald Trump’s exclusion from Twitter and other social media platforms after the January 6, 2021 attack on the U.S. Capitol (Bhuiyan 2022). It is part of a group of platforms, which also include microblogging platforms Parler, Gettr and Gab, as well as video hosting sites such as Rumble, Odysee and BitChute. These spaces have been explored so far, among other things, in regard to their understandings of safety (Otero and Scharlach 2025) and their role in cultivating emotional belonging (Collins 2024).
The current study evaluates different proposed markers of nostalgia. We do so on two existing Truth Social datasets (Gerard et al. 2023; Shah et al. 2024), one centered on Donald Trump's follower network in 2022 and one organized around 2024 election hashtags. We expect that nostalgic content will be more prevalent in the dataset focused on presidential elections, as it is considered a mobilizing force. The contribution focuses on pronoun-based and temporal-restitutive indicators of nostalgia, with implications for how computationally tractable proxies should be combined and validated in future work on contested digital memory.
Research on nostalgia in political contexts distinguishes between individual and collective forms of nostalgia. While individual nostalgia is typically associated with personal memory and individual emotion, collective nostalgia is oriented toward a shared past and a collective identity. Empirical work suggests that it is this group-focused form of nostalgia, rather than individual longing, that is systematically connected to right-wing political narratives (Versteegen 2024; Reyna et al. 2025). Collective nostalgia becomes politically salient when it is combined with perceptions of relative deprivation, that is, the belief that a valued social position or status has been unjustly lost. In this configuration, nostalgia functions not merely as remembrance, but as a framework for interpreting present grievances and legitimizing political claims (Versteegen 2024; Reyna et al. 2025). Collective nostalgia becomes politically salient when it is combined with perceptions of relative deprivation, that is, the belief that a valued social position or status has been unjustly lost. In this configuration, nostalgia functions not merely as remembrance, but as a framework for interpreting present grievances and legitimizing political claims (Versteegen 2024). (Versteegen 2024).
A key mechanism underlying the political effects of collective nostalgia lies in its selective orientation toward the past. Nostalgia can lead to outgroup intolerance, “because nostalgia is often rooted in selective memories that focus on positive aspects, omitting negative aspects of this longed-for time” (Reyna et al. 2025). Scholars distinguish between different narrative structures of nostalgia, such as redemptive narratives, which portray the past as a period of hardship that has been overcome. By contrast, contamination narratives depict the past as originally stable but subsequently corrupted by external or internal outgroups. Concepts such as grievance and blame therefore play a central role in group-serving attributions of loss associated with nostalgic discourse (Reyna et al. 2025).
Collective nostalgia can be anchored in different identity dimensions, including gender, religion, race, and nation. While the specific content of nostalgic longing varies across these domains, comparative research shows that the core psychological and narrative features of nostalgia remain remarkably stable - even across cultures (Hepper et al. 2014). This combination of contextual specificity and structural similarity makes collective nostalgia a particularly productive object for comparative and computational analysis.(Hepper et al. 2014). This combination of contextual specificity and structural similarity makes collective nostalgia a particularly productive object for comparative and computational analysis.
A phenomenon characterized through its ambivalence cannot be operationalized as a single, directly observable linguistic category. In this contribution, nostalgia is not treated as a clearly identifiable category, but as a multilayered phenomenon that manifests through a variety of recurring, sometimes measurable features associated with group identity, temporal dynamics, and perceived loss.
For this preliminary contribution, we evaluate the following indicators:
Following the FAIR principle of Reusability, we examine two already-existing datasets on Truth Social. The first dataset was collected between February and September 2022 and originally encompasses "the content of 823,927 truths posted by 454,458 users including the full history of the 65,536 most active users" (Gerard et al. 2023); after filtering to English-language posts with non-empty content, our working subset comprises 721,505 posts. The second dataset is focused on the 2024 Presidential Elections and originally comprises ~1.5 million posts published between February 2024 and October 2024 that contained election-related posts as well as hashtags related to everyday trending topics (e.g. "JoeBiden", "DonaldTrump", "2024USElections") (Shah et al. 2024); our working subset contains 1,262,907 posts.
The datasets were collected using different scraping logics: while the 2022 dataset proceeded in concentric circles around Donald Trump, scraping data from his followers, the 2024 dataset is a hashtag-oriented dataset that scraped posts containing election-related hashtags at regular intervals. This sampling difference is itself a methodological constraint: the 2022 corpus is centered on Trump's social network, while the 2024 corpus is event-centered around election hashtags. We return to its implications in the Outlook.
To assess whether the patterns are platform-specific rather than period-specific, we add a third corpus: the USC X-24 dataset of 2024 U.S. presidential election discourse on Twitter/X (Balasubramanian et al. 2024). We use part_1 of the public release, restricted to English-language tweets (n = 883,111; May-July 2024), which overlaps the Truth Social 2024 window (see Table 1).
| corpus | source | window | posts | content+function tokens |
| 2022 Truth Social | Gerard et al. 2023 (follower-network) | Feb-Sep 2022 | 721,505 | 13,148,853 |
| 2024 Truth Social | Shah et al. 2024 (election hashtags) | Feb-Oct 2024 | 1,262,907 | 55,702,431 |
| 2024 Twitter | USC X-24 part_1 English (Balasubramanian et al. 2024) | May-Jul 2024 | 883,111 | 14,096,898 |
Table 1. Corpus overview.
All preprocessing and analysis were carried out in RStudio. Texts were first cleaned with regular expressions in stringr: URLs were replaced with the placeholder URL, @-mentions with USER, and emoji tokens of the form <emoji: …> with EMOJI; hashtag tokens (#word) were removed in full to avoid inflating topical-keyword counts, and whitespace was normalised. The cleaned text was then tokenised, lemmatised and part-of-speech tagged with spacyr (R interface to spaCy), using the en_core_web_sm model.
We operationalised the two indicators introduced above directly on the spaCy parse. Collective self-reference was extracted by filtering tokens with POS tag PRON (and the fine-grained Penn tags PRP / PRP$), then grouping by lemma so that surface variants collapse into a single type. Per-corpus pronoun frequencies were computed both as absolute counts and as relative frequencies normalised per one million tokens, to make the differently-sized 2022 and 2024 corpora comparable. Temporal constructions were extracted in two passes: an adverbial pass matching the lemmas “again” and “back” on tokens tagged ADV, and a verbal pass matching the lemmas {reestablish, reclaim, restore, recover, return} on tokens tagged VERB. Counts were aggregated per corpus and again normalised per million tokens.
At the level of pronoun families (Table 2), the two Truth Social corpora show opposite profiles. In the 2022 corpus, first-person singular forms (I/me/my/mine) dominate at 19,106 per million tokens, clearly outweighing first-person plural forms (we/us/our/ours) at 13,540 per million. In the 2024 election corpus, this relationship inverts: 1sg drops sharply to 6,305 per million while 1pl remains comparatively high at 8,120 per million, making collective self-reference the more frequent framing. The 2024 Twitter comparison corpus sits between the two TS profiles (1sg = 13,141; 1pl = 7,521 per million), with 1sg still dominant, suggesting that the 1pl prominence in TS 2024 is not a generic feature of 2024 election discourse but specific to the Truth Social election corpus (TSm/TW ratio = 1.08 for 1pl vs. 0.48 for 1sg).
| Marker (lemma) | Category | 2022 TS / 1M | 2024 TS (May–Jul) / 1M | 2024 Twitter / 1M | 2024 TS / 2024 TW |
| 1sg (I / me / my / mine) | pronoun family | 19,106.1 | 6,304.7 | 13,140.5 | 0.48 |
| 1pl (we / us / our / ours) | pronoun family | 13,540.0 | 8,120.0 | 7,520.9 | 1.08 |
| Temporal adverbs | |||||
| again | ADV | 956.1 | 635.0 | 902.9 | 0.70 |
| back | ADV | 1,293.3 | 512.5 | 799.8 | 0.64 |
| aggregate (again + back) | ADV | 2,249.5 | 1,147.5 | 1,702.7 | 0.67 |
| —— Restitutive verbs —— | |||||
| reestablish | VERB | 0.3 | 0.6 | 1.1 | 0.56 |
| reclaim | VERB | 6.2 | 7.4 | 2.9 | 2.55 |
| restore | VERB | 45.4 | 79.1 | 28.9 | 2.73 |
| recover | VERB | 27.8 | 80.1 | 28.1 | 2.85 |
| return | VERB | 116.1 | 46.0 | 77.3 | 0.59 |
| aggregate (5 stems) | VERB | 195.7 | 213.2 | 138.3 | 1.54 |
Table 2. Nostalgia-marker rates across corpora
This shift supports the hypothesis insofar as the election context appears to require a different framing and a more explicitly collective language. At the same time, the results need to be interpreted with caution. Close-reading of several randomized samples demonstrated that many of the examples are mobilizing or sentimental without being clearly nostalgic. Prior research shows that right-wing social media strongly promotes in-group superiority, which is reflected in elevated use of first-person plural pronouns (Collins 2025).
This can be observed in constructions such as “Creating disastrous trade deals has sucked our country dry!” Here, “our country” invokes a collective in-group, but the nostalgic dimension remains weak and implicit, relying only on the assumption that trade deals were once not disastrous. This limitation illustrates why pronoun usage alone is insufficient as a nostalgia indicator and why temporal constructions (e.g. “go back”) are a valuable complementary marker, as in “I wish to God that we could go back to a society that believes in the content of our character and not the color of our skin.”
A further insight, and a construction that is easier to follow through the corpora, concerns the restitutive component of verbs such as "restore" or "reestablish". Table 2 shows that broadly applicable markers ("again", "back") account for the majority of temporal occurrences across all three corpora, whereas semantically more specific restorative verbs ("reclaim", "restore", "recover") remain comparatively rare. Their distribution, however, is the more telling signal: while the generic adverbs "again" and "back" are in fact less frequent on Truth Social 2024 than on Twitter 2024 (TS/TW ratio ≈ 0.67), the restitutive verbs "restore", "recover" and "reclaim" appear roughly 2.5–3 times more often on Truth Social than on the contemporaneous Twitter corpus. Within Truth Social, "restore" and "recover" also rise markedly from 2022 to 2024 (45.4 → 79.1 and 27.8 → 80.1 per million), even as the more neutral "return" declines. This combination — restitutive verbs elevated relative to both Twitter and the earlier TS corpus — points to a platform- and context-specific intensification of restorative framing in the election period, while still requiring contextual interpretation.
In the final part of this contribution, we synthesize the preceding analyses by situating individual indicators within a corpus-level perspective. Even though nostalgia does not manifest itself in a single form, corpus analyses in this field are fruitful. At the corpus level, nostalgia emerges as an overlap between political mobilization, moral evaluation and selective references to the past. The preliminary analysis shows that different operationalizations foreground different dimensions of this overlap, and that no single indicator captures the phenomenon on its own. The comparison across the three corpora illustrates this complexity: while collective self-reference and generic temporal adverbs do not rise uniformly in the election-focused Truth Social corpus, restitutive verbs are markedly more frequent on Truth Social 2024 than on the contemporaneous Twitter corpus, suggesting a platform- and context-specific intensification of restorative framing.
Building on the analyses presented above, we therefore examine how multiple criteria interact, including sentiment-based indicators, and explore their interaction through topic modeling, similarity networks and clustering approaches. Clustering is particularly productive because it provides a corpus-level overview of recurring patterns, allowing nostalgic framings to be situated within the broader discourse rather than classified in isolation.