Social Cyber Geographical Worldwide Inventory of Bots

Kathleen M. Carley; Lynnette Hui Xian Ng

REVIEW 4 major objections 4 minor 46 references

Social Cyber Geographical Worldwide Inventory of Bots

T0 review · 4 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Across countries, roughly one in five pandemic-related X accounts is a bot, and bots mostly write in the dominant language of the country they claim to be from.

desk verdict A useful multilingual bot inventory, but the uniform ~20% bot-per-country claim is not supported: the bot classifier is validated on only four machine-translated languages and applied to dozens, with no error bars on the headline number. read the letter →

arxiv 2501.18839 v1 pith:LOZ5OLCT submitted 2025-01-31 cs.SI

Lynnette Hui Xian Ng , Kathleen M. Carley This is my paper · ORCID

classification cs.SI

keywords socialbotsmultilingualbotdetectiongeolocationcybergeographyCOVID-19misinformationTwitter/Xdataproportionlanguagedistribution

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper sets out to map where social-media bots are active worldwide by analyzing about 100 million posts from roughly 31 million X (Twitter) accounts discussing the coronavirus pandemic in 2021. Its central finding is that the proportion of bot accounts is about 20% in country after country, regardless of economic size or population, so automated activity looks like a global baseline rather than a problem confined to specific states. The paper also finds that most bots write in the dominant language of the country they claim to be from, that bots can shift their claimed country over time while keeping the same language, and that only about 47% of bots write in English. If this is right, English-only bot detection misses more than half of automated activity, and efforts to regulate or counter bots need to take the multilingual, geographically distributed pattern seriously.

What carries the argument

The carrying machinery is Multilingual BotBuster, a bot classifier built by extending the feature-based BotBuster approach and using a multilingual BERT model as tokenizer and pretrained model; it was trained and tested on English bot datasets machine-translated into Chinese, Russian, and Arabic, with reported accuracy 82.79 ± 4.23% across those languages. The companion Geolocation Identifier extracts location words from a user's self-written description with a named-entity recognizer, fuzzy-matches them by Levenshtein distance against a gazetteer of cities and countries, and returns a country and coordinates. Together they convert raw tweets into a country-by-language grid on which bot proportions can be computed.

What would settle it

Build human-annotated bot/human test sets for Thai, Japanese, Tamil, Greek, and Latvian from the same pandemic corpus, run Multilingual BotBuster on them, and compare per-language accuracy and bot percentages. If the model performs at or near chance on those languages, or if the bot percentages it assigns diverge sharply from human judgments, the uniform ~20% country-level result and the high language-specific bot percentages are not established.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central finding is that the median bot proportion is approximately 20% across the countries represented in the dataset, and that this figure is stable even as claimed geographic affiliations shift over time. The paper further reports that at least 80% of bots affiliated with a country post in that country's dominant language, that bot percentages bear no significant correlation with GDP or population, and that bot-authored hashtags cluster around locally relevant narratives such as SputnikV in Russia and anti-China blame hashtags in China-affiliated accounts. The paper frames this as a 'social cyber geography': the combined space of bots' self-declared locations and the languages they write in.

Load-bearing premise

The analysis assumes that a bot-detection model trained only on English text machine-translated into Chinese, Russian, and Arabic correctly identifies bots in every other language in the data, including Thai, Japanese, Tamil, Greek, and Latvian.

Editorial extensions

If this is right

If the ~20% figure is correct, bot presence during the pandemic was a near-universal phenomenon, so regulators cannot treat automation as a problem of any one region or platform segment.
Because only 47% of bots write in English, English-only bot-detection pipelines would miss the majority of automated accounts and the narratives they spread.
Because bots mostly use the dominant language of their claimed country, bot operators are effectively targeting language diasporas, not just residents of that country.
Since bot affiliation shifts over time while language stays constant, tracking bots by claimed location alone is misleading; language is the more stable signal of target audience.
The absence of correlation between bot proportion and GDP or population implies that economic or demographic characteristics do not explain where bots are deployed.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A direct test of the uniformity claim would be to repeat the same country-level bot accounting on other high-volume global topics, such as a major election or a natural-disaster response, and see whether the median bot proportion again lands near 20%.
The paper's language-consistency result suggests an audience-reach model: bot operators pick the language with the largest diaspora of the target country. This could be tested by regressing the bot language mix against diaspora population estimates from migration statistics.
Because the classifier was validated only on English, Chinese, Russian, and Arabic, the high bot percentages reported for Thai, Japanese, Tamil, Greek, and Latvian rest on extrapolation; human-labeled test sets for those languages would show whether those numbers are real or an artifact of the multilingual model.
The gazetteer identifies where a bot claims to be, not where it operates; temporal 'movement' between countries could be profile editing rather than operator relocation, so pairing the method with VPN or IP inference would separate declared affiliation from actual origin.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

You asked about 2501.18839. Here's my take: the paper does something genuinely useful—builds a multilingual bot detection pipeline and applies it to a large 2021 COVID-19 corpus (about 100M tweets, 31M users) to produce a worldwide bot inventory—but it oversells the central empirical claim. The headline that 'regardless of the country, the average proportion of bots is approximately 20%' is not supported by the evidence as presented.

What's new and worth crediting: they extend BotBuster to a multilingual setting by machine-translating three English bot/human datasets into Chinese, Russian, and Arabic, and show that a multilingual BERT model beats the English-only baselines on those four languages. The worldwide bot inventory itself, along with the observation that bots mostly write in the dominant language of their claimed country, is a useful descriptive contribution. The regional hashtag analysis is also interesting.

The soft spots are load-bearing. First, the bot classifier is validated only on those four machine-translated languages; it is then applied to Thai, Japanese, Tamil, Greek, Latvian, and many others—exactly the languages that produce the highest bot percentages. If the model does not transfer, those numbers are noise. Second, the geolocation identifier has no accuracy evaluation; the authors compare methods on speed and coverage, but never report how often the gazetteer assigns the correct country on a labeled set. Third, the ~20% figure has no error bars, per-country sample sizes, or sensitivity analysis, even though the aggregation uses median and the claim is about averages. Fourth, the 'no correlation' conclusion rests on R^2 = 0.021 and 0.022 from linear regressions, with no significance tests or confidence intervals; with this many countries, a small R^2 could still be statistically significant, and substantively it just means GDP/population explain almost none of the variance, which is a different statement from 'no correlation.' The 0.5 probability threshold also comes from prior English-language work without being re-evaluated for the multilingual setting.

None of these flaws destroy the paper's descriptive value, but they do undercut the uniform-20% claim. I'd send this to peer review, because the dataset and the multilingual detection effort deserve expert scrutiny, and a good reviewer could push the authors to either add the missing validation or reframe the contribution as a pipeline/inventory rather than a universal empirical regularity. If I had to decide today with the evidence in the paper, I'd reject the central claim but encourage a resubmission after the missing pieces are added.

Referee Report

4 major / 4 minor

Summary. The paper presents a worldwide inventory of social media bots on X (formerly Twitter) during the 2021 coronavirus pandemic. The authors combine a multilingual bot detection model (Multilingual BotBuster) with a gazetteer-based geolocation identifier to estimate the proportion of bot accounts affiliated with each country. The central claim is that, regardless of country, the average proportion of bots is approximately 20% (Results, Figure 1), and that bots predominantly write in the dominant language of the country they claim to be from. The paper also examines bot percentages by language, temporal stability of affiliations, dominant hashtags per region, and correlations with GDP and population.

Significance. If the central claim is correct, this would be a significant empirical finding for social cybersecurity, suggesting a near-uniform global prevalence of automated accounts during a major public-health event. The paper leverages a very large dataset (~100 million posts, ~31 million users) and makes an explicit effort to move beyond English-centric bot detection. The proposed Multilingual BotBuster and the geolocation tool are potentially reusable resources. However, the validity of the headline result hinges entirely on whether the bot classifier generalizes to the many languages in the corpus that were not part of its validation set, and on the accuracy of the unvalidated geolocation step. At present, the evidence does not support the strong uniformity claim.

major comments (4)

[Methodology, Multilingual Bot Detection] The Multilingual BotBuster model is trained and validated only on three English-language datasets machine-translated into Chinese, Russian, and Arabic (Supplementary Material). The reported accuracy of 82.79 ± 4.23% is explicitly 'across the languages tested' — only those four. The model is then applied without further validation to all languages in the corpus. This is load-bearing because the Languages with the highest reported bot percentages — Thai (59.5%), Japanese (28.3%), Tamil (22.6%), Greek (28.9%), Latvian (37.5%) — are all outside the validation set. If the classifier's decision behavior differs on these unseen languages (e.g., due to machine-translation artifacts, script differences, or domain shift), every country-level bot proportion, including the headline 'approximately 20%' figure, is potentially an artifact. The authors need to provide per-language validation results, or at minimum a hand-validated sample from the unseen languages, before the central claim can be assessed.
[Results, Figure 1] The claim 'regardless of the country, the average proportion of bots is approximately 20%' is presented without any uncertainty quantification. No confidence intervals, standard errors, or per-country sample sizes are reported. The heatmap uses the median as the averaging function, but the distribution of bot proportions across countries is not shown. Without such information, the apparent uniformity could be driven by a few countries with very small sample sizes or by wide variance masked by the median. The claim should be accompanied by error bars or a statistical test for equality of proportions across countries.
[Bot Percentage vs Country Indicators] The paper states that there is 'no significant correlation between bot activity and economic activity or population' based on R² = 0.021 and R² = 0.022 from linear regressions. R² values alone do not establish the absence of a statistically significant correlation; the authors do not report p-values, confidence intervals, or effect sizes. Moreover, the number of accounts per country varies widely, so an unweighted regression may be misleading. The authors should use an appropriate correlation or regression framework that accounts for sampling variability and report the associated test statistics. As written, the conclusion that 'people from all countries are equally likely to create bots' is not supported by the evidence presented.
[Methodology, Geolocation Identifier] The gazetteer-based Geolocation Identifier is never validated for accuracy. The paper reports that it returns a location for 62% of users with descriptions (Table 4), but it does not report precision, recall, or any comparison against ground-truth locations. Since the entire country-level bot proportion analysis depends on assigning each user to a country through this tool, an unvalidated geolocation step could systematically misclassify users (e.g., due to fuzzy matching errors or ambiguous place names) and thereby distort the per-country bot percentages. The authors should provide at least a small human-annotated evaluation of the geolocation results.

minor comments (4)

[Abstract and Results] There are several typographical errors: 'Lativian' should be 'Latvian', 'Portugese' should be 'Portuguese', 'Myammar' should be 'Myanmar', and 'gazetter' should be 'gazetteer' in several places. The abstract contains an awkward sentence fragment: 'Since creating a bot uses universal programming technology, bots, how prevalent are these bots throughout the world?'
[Table 2 and Table 7] The hashtags listed in Table 2 and Table 7 contain apparent spacing artifacts (e.g., 'V oteFordOut2022', 'AW ANIpagi', 'NormaBaharu'). These should be corrected or explained, as they appear to be rendering errors of non-English hashtags.
[Introduction and Methodology] The references contain placeholder question marks, e.g., 'that characterize a bot?' and 'influencing online narratives?, 27–29'. These should be resolved to proper citations.
[Figure 1 and Figure 3 captions] The captions state that white areas indicate 'no bots present in the data we collected.' It would be helpful to also indicate whether these are countries with no tweets at all or no bots among the tweets collected, and to provide the number of users per country in the supplementary material.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the headline bot proportion is an empirical aggregation of a publicly trained classifier, not a quantity derived from its own inputs.

full rationale

The paper's central result — that roughly 20% of pandemic-related users per country are bots — is a measurement produced by applying the Multilingual BotBuster classifier to the collected corpus and aggregating the outputs by inferred country and language. There is no step in which a target quantity is defined in terms of another target quantity, nor any parameter fitted to the 2021 corpus and then renamed as a prediction. The classifier was trained on three public annotated English bot datasets (cresci-rtbust-2019, botometer-feedback-2019, cresci-stock-2018), machine-translated into Chinese, Russian, and Arabic, and evaluated with five-fold cross-validation against independent baselines including Botometer and BotHunter; the reported accuracy of 82.79 ± 4.23% is an external validation, not a self-referential fit. The 0.5 decision threshold is inherited from the authors' earlier published BotBuster system (ref. 19), which is a self-citation, but the threshold is a stated convention from a published, separately evaluated detector rather than a parameter fitted to this dataset or to the ~20% result. The paper also acknowledges the limitation that affiliations are self-declared and validated zero-bot regions against the independent GeoCOV19 dataset. The concern that the classifier may not generalize to languages outside its four-language validation set is a validity and external-generalization risk, not a circular derivation: the labels are not constructed to force the country-level proportions, so the central claim retains independent empirical content.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper's quantitative claims rest on thresholds chosen by the authors (bot score 0.5, gazetteer match 0.80) and on unvalidated assumptions: Twitter language tags are trusted, machine-translated English labels stand in for real non-English bot behavior, declared locations are treated as intended affiliations, and a four-language classifier is assumed to work for dozens of untested languages. No new entities are introduced; the 'social cyber geography' is a framing device, not an entity.

free parameters (2)

Bot classification decision threshold = 0.5
Probability threshold from previous BotBuster longitudinal studies (ref 19) to separate bot/human; the uniform ~20% proportion is directly sensitive to this choice.
Geolocation fuzzy match threshold = >0.80 (Levenshtein)
Gazeetteer match cut-off chosen by the authors; no sensitivity analysis; affects which users get assigned a country and thus per-country bot proportions.

assumptions (4)

domain assumption X's per-tweet language tag is accurate enough for grouping.
The language of each tweet is taken from metadata tags assigned by X's internal algorithms (Methodology, 'Bot Percentage by Language'); incorrect tags would misclassify bot language distributions.
domain assumption English bot/human labels remain valid after machine translation to Chinese, Russian, and Arabic.
The multilingual training sets are Google-Translated versions of English annotated datasets (Supplementary Material 'Multilingual BotBuster'); translation artifacts could shift label semantics.
domain assumption A user's self-declared location in their description corresponds to the affiliation the bot wants to project.
The paper acknowledges this, stating the affiliation is where bots claim to be from, not their true origin (Discussion, Limitations). The entire geospatial analysis uses this proclaimed affiliation.
domain assumption The fine-tuned multilingual BERT model generalizes to languages not in the training set.
Accuracy is reported only for English, Chinese, Russian, Arabic; the model is applied to Thai, Japanese, Tamil, Greek, Latvian, etc. This is the weakest assumption and is load-bearing for the per-language and per-country results.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Social Cyber Geographical Worldwide Inventory of Bots." pith.science (2026). https://pith.science/paper/LOZ5OLCT

@misc{pith2026250118839,
  author       = {Pith},
  title        = {Pith review of: Social Cyber Geographical Worldwide Inventory of Bots},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LOZ5OLCT}},
  note         = {Machine review of arXiv:2501.18839}
}

read the original abstract

Social Cyber Geography is the space in the digital cyber realm that is produced through social relations. Communication in the social media ecosystem happens not only because of human interactions, but is also fueled by algorithmically controlled bot agents. Most studies have not looked at the social cyber geography of bots because they focus on bot activity within a single country. Since creating a bot uses universal programming technology, bots, how prevalent are these bots throughout the world? To quantify bot activity worldwide, we perform a multilingual and geospatial analysis on a large dataset of social data collected from X during the Coronavirus pandemic in 2021. This pandemic affected most of the world, and thus is a common topic of discussion. Our dataset consists of ~100 mil posts generated by ~31mil users. Most bot studies focus only on English-speaking countries, because most bot detection algorithms are built for the English language. However, only 47\% of the bots write in the English language. To accommodate multiple languages in our bot detection algorithm, we built Multilingual BotBuster, a multi-language bot detection algorithm to identify the bots in this diverse dataset. We also create a Geographical Location Identifier to swiftly identify the countries a user affiliates with in his description. Our results show that bots can appear to move from one country to another, but the language they write in remains relatively constant. Bots distribute narratives on distinct topics related to their self-declared country affiliation. Finally, despite the diverse distribution of bot locations around the world, the proportion of bots per country is about 20%. Our work stresses the importance of a united analysis of the cyber and physical realms, where we combine both spheres to inventorize the language and location of social media bots and understand communication strategies.

Figures

Figures reproduced from arXiv: 2501.18839 by the authors.

**Figure 1.** Geographic Heat Map of the average (median) percentage of bots affiliated with each country, across the entire data. White areas indicates that there are no bots present in the data we collected [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

**Figure 2.** Languages that have the highest proportion of bot users. The Worldwide Inventory present in this paper has broad coverage of geographical regions and languages, which we believe can foster downstream multidimensional research. Future work involves characterizing these bots into different types of bots and analyzing the geographical distribution between bot types. Such an analysis provides insight towards the affilia… view at source ↗

**Figure 3.** Geographic Heat Map of the average (mean) percentage of bots affiliated with a country that authored posts in the country’s dominant language, across the entire data. White areas means that there are no bots present in the data we collected. Country Languages United States English, Spanish, Chinese, French, Undefined Russia English, Russian, Thai China English, Chinese, Thai, Spanish Antarctica English, Thai, Undefi… view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Scatter Map of Mean Bot Proportion vs. (a) GDP and (b) Population of country Results aggregated by month [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

**Figure 5.** Figure 5: Bot proportion against commonly used languages by month 12/19 [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

**Figure 6.** Figure 6: Geographical heatmap of bot proportion affiliated to each country by month 13/19 [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]

**Figure 7.** Figure 7: Geographical heatmap of bot proportion authoring posts in the country’s dominant language by month 14/19 [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 37 canonical work pages

[1]

Ratti, C. et al. Redrawing the map of great britain from a network of human interactions. PloS one 5, e14248 (2010). 17/19

work page 2010
[2]

Integrating social network data into gisystems

Andris, C. Integrating social network data into gisystems. Int. J. Geogr. Inf. Sci. 30, 2009–2031 (2016)

work page 2016
[3]

& Ofli, F

Qazi, U., Imran, M. & Ofli, F. Geocov19: a dataset of hundreds of millions of multilingual covid-19 tweets with location information. SIGSPATIAL Special12, 6–15 (2020)

work page 2020
[4]

Real-time geospatial surveillance of localized emotional stress responses to covid-19: a proof of concept analysis

Edry, T.et al. Real-time geospatial surveillance of localized emotional stress responses to covid-19: a proof of concept analysis. Heal. & Place 70, 102598 (2021)

work page 2021
[5]

Carley, K. M. Social cybersecurity: an emerging science. Comput. mathematical organization theory 26, 365–381 (2020)

work page 2020
[6]

Shen, F. et al. Examining the differences between human and bot social media accounts: A case study of the russia-ukraine war. First Monday 28 (2023)

work page 2023
[7]

Ng, L. H. X., Zhou, W. & Carley, K. M. Exploring cognitive bias triggers in covid-19 misinformation tweets: A bot vs. human perspective. arXiv preprint arXiv:2406.07293 (2024)

work page Pith review arXiv 2024
[8]

& Tammi, T

Unlu, A., Truong, S., Sawhney, N. & Tammi, T. Unveiling the veiled threat: the impact of bots on covid-19 health communication. Soc. Sci. Comput. Rev. 08944393241275641 (2024)

work page 2024

Show all 46 references

[9]

Bots, elections, and social media: a brief overview

Ferrara, E. Bots, elections, and social media: a brief overview. Disinformation, misinformation, fake news social media: Emerg. research challenges opportunities95–114 (2020)

2020
[10]

Ng, L. H. X., Bartulovic, M. & Carley, K. M. Tiny-botbuster: Identifying automated political coordination in digital campaigns. In International Conference on Social Computing, Behavioral-Cultural Modeling and Prediction and Behavior Representation in Modeling and Simulation, ...

2024
[11]

Duan, Z. et al. Algorithmic agents in the hybrid media system: Social bots, selective amplification, and partisan news about covid-19. Hum. Commun. Res. 48, 516–542 (2022)

2022
[12]

Ng, L. H. X. & Carley, K. M. Pro or anti? a social influence model of online stance flipping. IEEE Transactions on Netw. Sci. Eng. 10, 3–19 (2022)

2022
[13]

Tan, Z. et al. Botpercent: estimating twitter bot populations from groups to crowds. arXiv preprint arXiv:2302.00381 (2023)

2023 arXiv
[14]

Public diplomacy: The missing component in israel’s foreign policy

Gilboa, E. Public diplomacy: The missing component in israel’s foreign policy. In Israel’s Strategic Agenda, 102–134 (Routledge, 2013)

2013
[15]

Ng, L. H. X., Robertson, D. C. & Carley, K. M. Cyborgs for strategic communication on social media. Big Data & Soc. 11, 20539517241231275 (2024)

2024
[16]

H., Chen, E., Zhang, M., Muric, G

Chang, H.-C. H., Chen, E., Zhang, M., Muric, G. & Ferrara, E. Social bots and social media manipulation in 2020: The year in review. In Handbook of Computational Social Science, Volume 1, 304–323 (Routledge, 2021)

2020
[17]

& Shirazi, F

Hajli, N., Saeed, U., Tajvidi, M. & Shirazi, F. Social bots and the spread of disinformation in social media: the challenges of artificial intelligence. Br. J. Manag. 33, 1238–1253 (2022)

2022
[18]

& Chitungo, I

Dzinamarira, T., Dzobo, M. & Chitungo, I. Covid-19: A perspective on africa’s capacity and response. J. medical virology 92, 2465–2472 (2020)

2020
[19]

Ng, L. H. X. & Carley, K. M. Botbuster: Multi-platform bot detection using a mixture of experts. In Proceedings of the international AAAI conference on web and social media, vol. 17, 686–697 (2023)

2023
[20]

& Menczer, F

Yang, K.-C., Ferrara, E. & Menczer, F. Botometer 101: Social bot practicum for computational social scientists. J. computational social science 5, 1511–1528 (2022)

2022
[21]

Crises narratives defining the covid-19 pandemic: Expert uncertainties and conspiratorial sensemaking

Nadesan, M. Crises narratives defining the covid-19 pandemic: Expert uncertainties and conspiratorial sensemaking. Am. Behav. Sci. 00027642221085893 (2022)

2022
[22]

Himelein-Wachowiak, M. et al. Bots and misinformation spread on social media: Implications for covid-19. J. medical Internet research 23, e26933 (2021)

2021
[23]

Ng, L. H. X., Robertson, D. C. & Carley, K. M. Stabilizing a supervised bot detection algorithm: How much data is needed for consistent predictions? Online Soc. Networks Media 28, 100198 (2022)

2022
[24]

& Tesconi, M

Mazza, M., Cresci, S., Avvenuti, M., Quattrociocchi, W. & Tesconi, M. Rtbust: Exploiting temporal patterns for botnet detection on twitter. In Proceedings of the 10th ACM conference on web science, 183–192 (2019)

2019
[25]

Yang, K.-C. et al. Arming the public with artificial intelligence to counter social bots. Hum. Behav. Emerg. Technol.1, 48–61 (2019). 18/19

2019
[26]

& Tesconi, M

Cresci, S., Lillo, F., Regoli, D., Tardelli, S. & Tesconi, M. Cashtag piggybacking: Uncovering spam and bot activity in stock microblogs on twitter. ACM Transactions on Web (TWEB)13, 1–27 (2019)

2019
[27]

S., Ng, L

Jacobs, C. S., Ng, L. H. X. & Carley, K. M. Tracking china’s cross-strait bot networks against taiwan. In International conference on social computing, behavioral-cultural modeling and prediction and behavior representation in modeling and simulation, 115–125 (Springer, 2023)

2023
[28]

Bail, C. A. et al. Assessing the russian internet research agency’s impact on the political attitudes and behaviors of american twitter users in late 2017. Proc. national academy sciences 117, 243–250 (2020)

2020
[29]

Simchon, A., Brady, W. J. & Van Bavel, J. J. Troll and divide: the language of online polarization. PNAS nexus 1, pgac019 (2022)

2022
[30]

Ng, L. H. X. & Carley, K. M. Assembling a multi-platform ensemble social bot detector with applications to us 2020 elections. Soc. Netw. Analysis Min. 14, 45 (2024)

2024
[31]

Beskow, D. M. & Carley, K. M. Bot-hunter: a tiered approach to detecting & characterizing automated activity on twitter. In Conference paper. SBP-BRiMS: International conference on social computing, behavioral-cultural modeling and prediction and behavior representation in mod...

2018
[32]

& Toutanova, K

Devlin, J., Chang, M., Lee, K. & Toutanova, K. BERT: pre-training of deep bidirectional transformers for language understanding. CoRR abs/1810.04805 (2018). 1810.04805

2018 arXiv
[33]

Ng, L. H. X. & Carley, K. M. Deflating the chinese balloon: types of twitter bots in us-china balloon incident. EPJ Data Sci. 12, 63 (2023)

2023
[34]

& Camber, W

Haupt, A. & Camber, W. Conflict prediction through geo-spatial interpolation of radicalization in syrian social media. Tradoc Analysis Cent. (2015)

2015
[35]

R., Grenager, T

Finkel, J. R., Grenager, T. & Manning, C. D. Incorporating non-local information into information extraction systems by gibbs sampling. In Proceedings of the 43rd annual meeting of the association for computational linguistics (ACL’05), 363–370 (2005)

2005
[36]

& Wolf, T

Sanh, V ., Debut, L., Chaumond, J. & Wolf, T. Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter. ArXiv abs/1910.01108 (2019)

2019 arXiv
[37]

Roberta: A robustly optimized BERT pretraining approach

Liu, Y .et al. Roberta: A robustly optimized BERT pretraining approach. CoRR abs/1907.11692 (2019). 1907.11692

2019 arXiv
[38]

google-bert/bert-base-chinese · Hugging Face — huggingface.co

Google. google-bert/bert-base-chinese · Hugging Face — huggingface.co. https://huggingface.co/google-bert/ bert-base-chinese (2022). [Accessed 11-09-2024]

2022
[39]

Pre-training with whole word masking for chinese bert

Cui, Y .et al. Pre-training with whole word masking for chinese bert. arXiv preprint arXiv:1906.08101 (2019)

2019 arXiv
[40]

& Habash, N

Inoue, G., Alhafni, B., Baimukan, N., Bouamor, H. & Habash, N. The interplay of variant, size, and task type in Arabic pre-trained language models. In Proceedings of the Sixth Arabic Natural Language Processing Workshop(Association for Computational Linguistics, Kyiv, Ukraine ...

2021
[41]

& Yuret, D

Safaya, A., Abdullatif, M. & Yuret, D. KUISAIL at SemEval-2020 task 12: BERT-CNN for offensive speech identification in social media. In Proceedings of the Fourteenth Workshop on Semantic Evaluation, 2054–2059 (International Committee for Computational Linguistics, Barcelona (...

2020
[42]

Conneau, A. et al. Unsupervised cross-lingual representation learning at scale. CoRR abs/1911.02116 (2019). 1911.02116

2019 arXiv
[43]

& Conneau, A

Lample, G. & Conneau, A. Cross-lingual language model pretraining. arXiv preprint arXiv:1901.07291 (2019)

2019 arXiv
[44]

& Poria, S

Bhardwaj, R. & Poria, S. Red-teaming large language models using chain of utterances for safety-alignment.arXiv preprint arXiv:2308.09662 (2023)

2023 arXiv
[45]

Radford, A. et al. Language models are unsupervised multitask learners. OpenAI blog 1, 9 (2019)

2019
[46]

& Savitri, I

Pohan, S., Pohan, H. & Savitri, I. N. Digital diplomacy-maximizing social media in indonesia’s economic and cultural diplomacy. In 1st International Conference on Social and Political Development (ICOSOP 2016), 372–390 (Atlantis Press, 2016). 19/19

2016

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [1]

Ratti, C. et al. Redrawing the map of great britain from a network of human interactions. PloS one 5, e14248 (2010). 17/19

work page 2010

[2] [2]

Integrating social network data into gisystems

Andris, C. Integrating social network data into gisystems. Int. J. Geogr. Inf. Sci. 30, 2009–2031 (2016)

work page 2016

[3] [3]

& Ofli, F

Qazi, U., Imran, M. & Ofli, F. Geocov19: a dataset of hundreds of millions of multilingual covid-19 tweets with location information. SIGSPATIAL Special12, 6–15 (2020)

work page 2020

[4] [4]

Real-time geospatial surveillance of localized emotional stress responses to covid-19: a proof of concept analysis

Edry, T.et al. Real-time geospatial surveillance of localized emotional stress responses to covid-19: a proof of concept analysis. Heal. & Place 70, 102598 (2021)

work page 2021

[5] [5]

Carley, K. M. Social cybersecurity: an emerging science. Comput. mathematical organization theory 26, 365–381 (2020)

work page 2020

[6] [6]

Shen, F. et al. Examining the differences between human and bot social media accounts: A case study of the russia-ukraine war. First Monday 28 (2023)

work page 2023

[7] [7]

Ng, L. H. X., Zhou, W. & Carley, K. M. Exploring cognitive bias triggers in covid-19 misinformation tweets: A bot vs. human perspective. arXiv preprint arXiv:2406.07293 (2024)

work page Pith review arXiv 2024

[8] [8]

& Tammi, T

Unlu, A., Truong, S., Sawhney, N. & Tammi, T. Unveiling the veiled threat: the impact of bots on covid-19 health communication. Soc. Sci. Comput. Rev. 08944393241275641 (2024)

work page 2024