Can Generative Agent-Based Modeling Replicate the Friendship Paradox in Social Media Simulations?

Diego Russo; Gian Marco Orlando; Valerio La Gatta; Vincenzo Moscato

REVIEW 4 major objections 6 minor 24 references

Can Generative Agent-Based Modeling Replicate the Friendship Paradox in Social Media Simulations?

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read Simulated social-media agents reproduce the friendship paradox without explicit wiring.

desk verdict A useful but under-controlled demonstration that the friendship paradox appears in a GABM; the missing null model and quantitative real-world comparison leave the 'emergence' claim weaker than the abstract suggests. read the letter →

arxiv 2502.05919 v1 pith:R5LSFNVK submitted 2025-02-09 cs.SI

Gian Marco Orlando , Valerio La Gatta , Diego Russo , Vincenzo Moscato This is my paper · ORCID

classification cs.SI

keywords GenerativeAgentsFriendshipParadoxAgent-BasedModelingSocialMediaSimulationLLMNetworkEmergenceNeighborSuperiorityRecommendationBias

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 asks whether a generative agent-based model (GABM) — a simulation in which LLM-powered agents post, follow, and react like social media users — can reproduce a global network-level pattern known as the Friendship Paradox. The paradox is the well-documented observation that, on average, your friends have more friends than you do, which follows from skewed degree distributions. Using Twitter data from the 2020 US election and the QAnon community to initialize 100 agents per dataset, the authors report that the paradox and its generalizations (activity, virality, susceptibility) emerge in the simulated network without being explicitly programmed. They also find that agents tend to follow users with higher activity or influence, and that the paradox is driven mostly by infrequent, low-contact connections rather than close ties, matching real-network findings. If correct, the result supports GABM as a viable tool for studying whole-network social phenomena, not just local behaviors.

What carries the argument

The machinery is the GABM simulation loop: agents with LLM-generated personality profiles, a short-term and long-term memory of posts and engagement, and a reasoning module that outputs a Choice-Reason-Content triplet each iteration. The load-bearing mechanism for network formation is the 'who to follow' step, where a target agent receives the top-10 other agents ranked by cosine similarity of semantic embeddings of their posts, and the LLM then decides whom to follow and why. This is combined with a recommender feed based on retrieval-augmented generation, and the whole loop repeats until posts become redundant. The outcome is measured by neighbor superiority: the percentage of agents whose average over followers or followees for an attribute exceeds their own value, mirroring the standard definition of the Friendship Paradox and its generalizations.

What would settle it

Run the same simulation with the 'who to follow' candidates chosen uniformly at random instead of by semantic similarity, holding everything else fixed; if the friendship paradox (e.g., 86% mean followee superiority for in-degree in the US Election dataset) drops toward the 50% baseline, the effect is an artifact of the recommender rather than an emergent property of agent behavior.

Watch

Extended reading notes

Core claim

The paper's central claim is that the Friendship Paradox arises naturally in GABM social-media simulations. In both the polarized US Election dataset and the cohesive QAnon dataset, a majority of agents experience mean followee superiority for in-degree, retweets, and other attributes — meaning their followees score higher than they do — while follower superiority is less common. The authors interpret this asymmetry as evidence of an implicit social hierarchy in which agents preferentially follow more active and influential others. They further report that restricting the comparison to the single most-interacted follower or followee weakens the paradox, while expanding to the full neighbor set strengthens it, implying that it is the long tail of infrequent connections that carries the effect. The claim is not that generative agents are better than traditional models at producing the paradox, but that the paradox is an emergent product of the LLM-driven decision process rather than a hand-coded outcome.

Load-bearing premise

The load-bearing assumption is that restricting 'who to follow' candidates to the top-10 most semantically similar agents does not itself create the degree heterogeneity that produces the Friendship Paradox; if the candidate restriction does the work, the results say little about emergent agent behavior.

Editorial extensions

If this is right

If the Friendship Paradox is an emergent outcome of GABM, then GABM can be used to study other global network properties — degree distributions, hierarchy, information asymmetry — without hand-wiring them into the model.
The finding that infrequent connections drive the paradox suggests that in simulated social media, weak ties are the carriers of structural inequality, providing a test bed for interventions aimed at reducing perceived popularity gaps.
The asymmetry between followee and follower superiority offers a quantitative handle on how recommender systems influence the perceived social hierarchy in online platforms.
Because the same pattern appears in two very different communities — polarized mainstream politics and a fringe conspiracy network — the framework generalizes across network contexts.
The framework provides a way to probe how different personality profiles, memory mechanisms, or recommender strategies alter the strength of the paradox.

Reading between the lines

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

The paper does not isolate the role of the semantic-similarity recommender; a natural extension would be to run the same simulation with random candidate suggestions, and if the paradox weakens, the top-10 similarity mechanism — not agent-level reasoning — would be doing much of the work.
A stronger test of emergence would compare the observed neighbor-superiority rates against a null model that randomizes follow decisions while preserving agent activity levels; the paper does not report such a baseline.
The use of only 100 agents per dataset leaves open whether the effect scales to thousands of agents or whether it is an artifact of small-network degree distributions.
The framework could be used to study the temporal dynamics of the paradox — for example, whether it appears early in simulations before the saturation stop condition — which the authors list as future work.

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

Quick read on arXiv:2502.05919. The paper does one new thing: it runs a generative agent-based simulation of Twitter-like behavior and shows that the friendship paradox and its generalizations appear in the resulting network. The setup is concrete -- 100 agents seeded from real US 2020 Election and QAnon Twitter data, LLaMa-3 driven, with memory and a recommender-style feed. They measure neighbor superiority for in-degree, activity, virality, and susceptibility, and report that followee superiority is consistently higher than follower superiority. As a demonstration that GABM can produce a global network-level phenomenon, it is a reasonable starting point.

The soft spots are not minor. The 'who to follow' step restricts candidates to the top-10 agents with highest semantic similarity to the target. That is a strong structural prior. Followers will tend to concentrate on a few semantically central agents regardless of what the LLM does. The paper never applies a null model -- no random followee baseline, no random candidate lists, no ablation of the top-10 restriction. Without that, you cannot attribute the observed degree heterogeneity, and therefore the FP, to the generative choices of the agents. The phrase 'emerges naturally' overstates what is shown.

The second soft spot is the real-world comparison. They cite that their findings align with real networks, but never report the actual FP percentages or neighbor-superiority rates from the source Twitter datasets. So 'consistent with real-world observations' is supported only qualitatively. For RQ2, the k-restriction analysis is descriptive; no statistical test establishes the increasing trend, and 17/21 vs 10/21 is a pattern but not a demonstrated regularity.

What the paper does well is transparency. The pipeline is described in enough detail to replicate, they report standard deviations over three runs, and they don't hide the manual threshold choices. The self-citation for RAG is appropriate -- that component is not the source of the FP claim. This is not a circular or dishonest paper; it's an under-controlled empirical study.

Who is this for? Researchers working on GABM validation and social simulation. They will learn where the field stands and get a clear example of the kind of control that needs to accompany an 'emergence' claim. The paper deserves a serious referee, but it needs major revision before publication: a random-followee null model, quantitative comparison to the real networks, and significance testing on RQ2.

I'd bring it to reading group as a cautionary example, but I wouldn't build on it as it stands.

Recommendation: send to peer review with the expectation of substantial revision.

Referee Report

4 major / 6 minor

Summary. This paper proposes a generative agent-based modeling (GABM) framework for social media, using Llama 3 agents with profiles inferred from Twitter users, a memory unit, and a semantic-similarity-based recommendation mechanism. It evaluates whether the Friendship Paradox and its generalizations (activity, virality, susceptibility) emerge by measuring neighbor superiority across in-degree, out-degree, and several content-metrics in two simulated Twitter corpora. The authors report that followee superiority exceeds follower superiority in 12 of 14 cases in Table 1 and that paradox prevalence generally increases when expanding from the closest connections to the full follower/followee sets, concluding that GABM can replicate this real-world network phenomenon.

Significance. If the result is robust, the paper would be a valuable step in showing that LLM-driven agents can reproduce a global network-level phenomenon without explicit wiring to create the paradox. The measurement approach is reasonable as a first step: the authors use a formal neighbor-superiority definition, report three independent runs with standard deviations, and seed agents from two real Twitter datasets. However, the central claim currently rests on an unablated recommender mechanism and lacks quantitative comparison with the real-world paradox magnitudes, so the significance is not yet established. The paper does not provide code or a reproducibility statement, which limits verification, but the clarity of the proposed workflow is a strength.

major comments (4)

[§2.2 / Table 1] The top-10 semantic-similarity candidate pool is an unablated confound that is load-bearing for the paper's central claim. For each target agent, the 'who to follow' mechanism restricts candidate followees to the ten agents with the highest average BERT cosine similarity to the target's posts. This structural prior can concentrate followership on semantically central agents even if the LLM's follow decisions are essentially arbitrary, because such agents will appear in many candidate lists. To support the claim that the Friendship Paradox 'emerges naturally,' the authors should add a null model that selects followees uniformly at random from the same top-10 candidate lists, or one that uses random candidate lists, and compare the neighbor-superiority percentages in Table 1. Without this baseline, the 86% in-degree followee superiority for the US Election dataset, and the analogous percentages for other attributes, may reflect the recommender rather than emergent generative agent behavior.
[§3.1] The claim that the simulations are 'consistent with real-world observations' is not quantitatively tested. Table 1 reports only simulated neighbor-superiority percentages; the paper does not report the corresponding percentages computed on the original Twitter datasets, nor does it perform any statistical comparison between simulated and empirical values. A direct comparison using the same neighbor-superiority metric on the source data is necessary to support the abstract's statement that the findings are consistent with real-world networks, rather than only qualitatively aligned with the existence of the paradox.
[§3.2 / Figure 3] The RQ2 conclusion that infrequent connections primarily drive the paradoxes is supported only by descriptive trends. For instance, the US Election in-degree followee superiority increases from 72.33% to 82.66% to 86.33% as the connection set expands from k=1 to k=3 to the full followee set, but no standard deviations or significance tests are reported for these differences, and the trend is not monotone in many of the other cases (the authors note 17 of 21 for follower means and only 10 of 21 for followee means). The authors should report per-condition standard deviations, as they do in Table 1, and provide a statistical test or effect-size measure for the claimed monotone pattern, or temper the causal language in Section 4.
[§2.1 / §2.2] Several ad-hoc thresholds are fixed without sensitivity analysis: the LTM transfer threshold tau = 0.5, the top-10 candidate-pool size, and the stop-condition cosine-similarity threshold of 0.99. Since the paper's contribution is that the Friendship Paradox emerges naturally rather than through explicit wiring, the authors should demonstrate stability of the main results under reasonable variation of these thresholds, at least for a small range of settings. This would also address the paper's claim that the framework avoids extensive parameter calibration.

minor comments (6)

[§2.2] The description of the semantic similarity computation is ambiguous: 'the average cosine similarity between the semantic embeddings of ai's posts and those of other agents' does not specify whether the average is over all posts of each agent, and whether the target agent's own posts are excluded from the candidate set; please clarify.
[Figure 3] Figure 3 does not include error bars or a legend identifying which attribute corresponds to each line; the caption says 'for each attribute' but the lines are not labeled in the figure or caption. Adding standard-deviation bars, as in Table 1, would substantially improve interpretability.
[§3] The Influence-driven Adoption Rate (IAR) is defined verbally as 'the ratio of content shared by the agent to the total content exposure' but no formula is given; a formal definition would make the metric reproducible.
[§2.1] Reference [8] is cited for the neighbor-superiority operationalization, but it is a thesis; consider citing a more accessible or peer-reviewed source for this definition.
[§2.2 / §3] The stop-condition description states that the simulation halts when 'L original posts' exhibit cosine similarity greater than 0.99 with prior content, where L is the number of agents; it is unclear whether each agent must produce at least one redundant post or whether any L posts suffice. Please clarify the exact stopping criterion.
[Abstract / §4] The phrase 'infrequent connections' is used in the abstract and conclusion, but the operationalization in §3.2 defines this only as connections outside the top-k most interactive set; this relational definition should be stated in the abstract or the relevant section to avoid ambiguity.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the FP is measured directly from the generated network, not fitted; the only self-citation ([7]) supports the RAG component, not the FP conclusion.

full rationale

The paper's central claim is that the Friendship Paradox and its generalizations appear in the simulated networks. This is a direct measurement (Table 1) from the edge set and nodal attributes produced by the generative agents. No parameter is fitted to reproduce the FP; the 'who to follow' mechanism restricts candidates to the top-10 semantically similar agents, but this is a structural choice, not a value fitted to the FP outcome. The self-citation [7] (Ferraro et al.) is used for the Retrieval-Augmented Generation component and for the background claim about echo chambers; it is not the basis for the FP result. No uniqueness theorem, ansatz, or renaming is imported to force the conclusion. The absence of a random-followee ablation or a quantitative comparison to real-world FP magnitudes is a methodological limitation (external validity), not circularity. Therefore no circular step can be exhibited.

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

No parameters were fitted to reproduce the Friendship Paradox, but several design constants are hand-chosen and untested. The most consequential is the top-10 semantically similar candidate pool, which restricts the network's possible edges and may artificially generate the degree heterogeneity that yields the FP.

free parameters (4)

Short-term memory to long-term memory threshold tau = 0.5
Hand-chosen engagement threshold for transferring posts to LTM; no sensitivity analysis provided (Section 2.1).
Candidate pool size (top-k semantically similar agents) = 10
The 'who to follow' mechanism restricts potential new connections to the 10 most similar agents, directly shaping the network topology that FP is measured on (Section 2.2).
Stop-condition similarity threshold = 0.99
Simulation halts when L original posts have cosine similarity greater than 0.99 to prior content; the threshold is arbitrary and affects the final network state (Section 2.2).
Number of agents per simulation run = 100
Only 100 users are instantiated from datasets with hundreds of thousands to millions of users; small sample size may limit generalizability (Section 3, Experimental setup).

assumptions (4)

domain assumption LLM-generated agents produce follower decisions that resemble real Twitter user decisions
The validation claim assumes agent behavior is a proxy for real users; the paper does not validate this proxy against the real Twitter networks it uses (Section 2.1, Section 3.1).
ad hoc to paper The top-10 semantic similarity candidate pool is a neutral setup that does not predetermine the Friendship Paradox
This design choice restricts who can be followed and may create the degree heterogeneity that produces FP, independent of any emergent agent preference (Section 2.2, Interaction Phase).
standard math The neighbor superiority metrics adequately operationalize the Friendship Paradox
The metric is drawn from prior literature [8,16] and is mathematically uncontroversial.
domain assumption The stop condition marks a meaningful terminal state of the simulation
The saturation heuristic (cosine similarity greater than 0.99) is not validated against behavior in longer runs (Section 2.2, Stop Condition).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Can Generative Agent-Based Modeling Replicate the Friendship Paradox in Social Media Simulations?." pith.science (2026). https://pith.science/paper/R5LSFNVK

@misc{pith2026250205919,
  author       = {Pith},
  title        = {Pith review of: Can Generative Agent-Based Modeling Replicate the Friendship Paradox in Social Media Simulations?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/R5LSFNVK}},
  note         = {Machine review of arXiv:2502.05919}
}

read the original abstract

Generative Agent-Based Modeling (GABM) is an emerging simulation paradigm that combines the reasoning abilities of Large Language Models with traditional Agent-Based Modeling to replicate complex social behaviors, including interactions on social media. While prior work has focused on localized phenomena such as opinion formation and information spread, its potential to capture global network dynamics remains underexplored. This paper addresses this gap by analyzing GABM-based social media simulations through the lens of the Friendship Paradox (FP), a counterintuitive phenomenon where individuals, on average, have fewer friends than their friends. We propose a GABM framework for social media simulations, featuring generative agents that emulate real users with distinct personalities and interests. Using Twitter datasets on the US 2020 Election and the QAnon conspiracy, we show that the FP emerges naturally in GABM simulations. Consistent with real-world observations, the simulations unveil a hierarchical structure, where agents preferentially connect with others displaying higher activity or influence. Additionally, we find that infrequent connections primarily drive the FP, reflecting patterns in real networks. These findings validate GABM as a robust tool for modeling global social media phenomena and highlight its potential for advancing social science by enabling nuanced analysis of user behavior.

Figures

Figures reproduced from arXiv: 2502.05919 by the authors.

**Figure 3.** Proportion of agents experiencing mean follower [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 12 canonical work pages

[1]

Amarnath Amarasingam and Marc-André Argentino. 2020. The QAnon conspir- acy theory: A security threat in the making. CTC Sentinel 13, 7 (2020), 37–44

work page 2020
[2]

James Bagrow, Christopher Danforth, and Lewis Mitchell. 2017. Which friends are more popular than you? Contact strength and the friendship paradox in social networks. (03 2017). https://doi.org/10.48550/arXiv.1703.06361

work page Pith review arXiv doi:10.48550/arxiv.1703.06361 2017
[3]

E. Chen, A. Deb, and E. Ferrara. 2022. #Election2020: the first public Twitter dataset on the 2020 US Presidential election. J. Comput. Soc. Sci. 5, 1 (2022), 1–18. https://doi.org/10.1007/S42001-021-00117-9

work page doi:10.1007/s42001-021-00117-9 2022
[4]

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 4171–4186

2019
[5]

Abhimanyu Dubey et al . 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

arXiv 2024
[6]

Scott L Feld. 1991. Why your friends have more friends than you do. American journal of sociology 96, 6 (1991), 1464–1477

work page 1991
[7]

Antonino Ferraro, Antonio Galli, Valerio La Gatta, Marco Postiglione, Gian Marco Orlando, Diego Russo, Giuseppe Riccio, Antonio Romano, and Vincenzo Moscato
[8]

Babak Fotouhi. 2014. Complex networks: dynamism of connectivity and opinion. (2014)

work page 2014

Show all 24 references

[9]

C. Gao, X. Lan, Z. Lu, J. Mao, J. Piao, H. Wang, D. Jin, and Yong Li. 2023. S3: Social- network Simulation System with Large Language Model-Empowered Agents. arXiv:2307.14984 (2023)

2023 arXiv
[10]

Navid Ghaffarzadegan, Aritra Majumdar, Ross Williams, and Niyousha Hos- seinichimeh. 2023. Generative agent-based modeling: Unveiling social system dynamics through coupling mechanistic models with generative artificial intelli- gence. arXiv preprint arXiv:2309.11456 (2023)

2023 arXiv
[11]

Matthew N. Hannah. 2022. Collection of tweets related to QAnon hashtags. https://doi.org/doi:/10.4231/32MD-DB04

2022 doi
[12]

Nathan Hodas, Farshad Kooti, and Kristina Lerman. 2013. Friendship paradox re- dux: Your friends are more interesting than you. InProceedings of the International AAAI Conference on Web and Social Media , Vol. 7. 225–233

2013
[13]

Matthew O. Jackson. 2019. The Friendship Paradox and Systematic Biases in Perceptions and Social Norms. Journal of Political Economy 127, 2 (2019), 777–818. https://doi.org/10.1086/701031 arXiv:https://doi.org/10.1086/701031

2019 doi
[14]

Tse Yang Lim, Erin J Stringfellow, Celia A Stafford, Catherine DiGennaro, Jack B Homer, Wayne Wakeland, Sara L Eggers, Reza Kazemi, Lukas Glos, Emily G Ewing, et al. 2022. Modeling the evolution of the US opioid crisis for national policy development. Proceedings of the Nation...

2022
[15]

Luca Luceri, Jinyi Ye, Julie Jiang, and Emilio Ferrara. 2024. The Susceptibility Paradox in Online Social Influence. arXiv preprint arXiv:2406.11553 (2024)

2024 arXiv
[16]

Naghmeh Momeni and Michael Rabbat. 2016. Qualities and inequalities in online social networks through the lens of the generalized friendship paradox. PloS one 11, 2 (2016), e0143633

2016
[17]

Vishakh Padmakumar and He He. 2023. Does Writing with Language Models Reduce Content Diversity? arXiv preprint arXiv:2309.05196 (2023). WEBSCI ’25, May 20-24, 2025, New Brunswick, New Jersey, USA Orlando et al

2023 arXiv
[18]

Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S Bernstein. 2023. Generative agents: Interactive simulacra of human behavior. In Proceedings of the 36th annual acm symposium on user interface software and technology . 1–22

2023
[19]

Javier Pastor-Galindo, Pantaleone Nespoli, and José A Ruipérez-Valiente. 2024. Large-Language-Model-Powered Agent-Based Framework for Misinformation and Disinformation Research: Opportunities and Open Challenges. IEEE Security & Privacy (2024)

2024
[20]

Petter Törnberg, Diliara Valeeva, Justus Uitermark, and Christopher Bail. 2023. Simulating social media using large language models to evaluate alternative news feed algorithms. arXiv preprint arXiv:2310.05984 (2023)

2023 arXiv
[21]

Lei Wang, Jingsen Zhang, Hao Yang, Zhiyuan Chen, Jiakai Tang, Zeyu Zhang, Xu Chen, Yankai Lin, Ruihua Song, Wayne Xin Zhao, et al. 2023. User behavior sim- ulation with large language model based agents. arXiv preprint arXiv:2306.02552 (2023)

2023 arXiv
[22]

Ross Williams, Niyousha Hosseinichimeh, Aritra Majumdar, and Navid Ghaf- farzadegan. 2023. Epidemic modeling with generative agents. arXiv preprint arXiv:2307.04986 (2023)

2023 arXiv
[23]

Q. Wu, G. Bansal, J. Zhang, Y. Wu, B. Li, E. Zhu, L. Jiang, X. Zhang, S. Zhang, J. Liu, A. H. Awadallah, R. W. White, D. Burger, and C. Wang. 2023. AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation. arXiv:2308.08155 (2023)

2023 arXiv
[2024]

arXiv:2411.16031 [cs.SI] https://arxiv.org/abs/2411.16031

Agent-Based Modelling Meets Generative AI in Social Network Simulations. arXiv:2411.16031 [cs.SI] https://arxiv.org/abs/2411.16031

arXiv

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [1]

Amarnath Amarasingam and Marc-André Argentino. 2020. The QAnon conspir- acy theory: A security threat in the making. CTC Sentinel 13, 7 (2020), 37–44

work page 2020

[2] [2]

James Bagrow, Christopher Danforth, and Lewis Mitchell. 2017. Which friends are more popular than you? Contact strength and the friendship paradox in social networks. (03 2017). https://doi.org/10.48550/arXiv.1703.06361

work page Pith review arXiv doi:10.48550/arxiv.1703.06361 2017

[3] [3]

E. Chen, A. Deb, and E. Ferrara. 2022. #Election2020: the first public Twitter dataset on the 2020 US Presidential election. J. Comput. Soc. Sci. 5, 1 (2022), 1–18. https://doi.org/10.1007/S42001-021-00117-9

work page doi:10.1007/s42001-021-00117-9 2022

[4] [4]

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 4171–4186

2019

[5] [5]

Abhimanyu Dubey et al . 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

arXiv 2024

[6] [6]

Scott L Feld. 1991. Why your friends have more friends than you do. American journal of sociology 96, 6 (1991), 1464–1477

work page 1991

[7] [7]

Antonino Ferraro, Antonio Galli, Valerio La Gatta, Marco Postiglione, Gian Marco Orlando, Diego Russo, Giuseppe Riccio, Antonio Romano, and Vincenzo Moscato

[8] [8]

Babak Fotouhi. 2014. Complex networks: dynamism of connectivity and opinion. (2014)

work page 2014