Aleksandar Tomašević

c, et al.: Operational Validation of Large-Language-Model Agent Social Simulation: Evidence from Voat v/technology https://arxiv.org/abs/2508.21740 https://arxiv.org/pdf/2508.21740 https://arxiv.org/html/2508.21740

Operational Validation of Large-Language-Model Agent Social Simulation: Evidence from Voat v/technology
https://arxiv.org/abs/2508.21740

Operational Validation of Large-Language-Model Agent Social Simulation: Evidence from Voat v/technology Large Language Models (LLMs) enable generative social simulations that can capture culturally informed, norm-guided interaction on online social platforms. We build a technology community simulation m...

Rough first version of the paper is on ArXiv: arxiv.org/abs/2508.21740

Full paper with robustness analysis up in a few weeks.

Fails: Identity persistence, long-term coherence, true preferential attachment. ❌

Memoryless setup prevents identity formation; network core is more difuse, no agent differentiation in stylistic alignment.

Can LLM agents reproduce online community dynamics?

We simulated Voat's v/technology community with YSocial and uncensored LLM.

Simulation reproduces: activity patterns, network topology, toxicity distributions + decent topic coverage ✅

100% AI generated :)

MADOC paper accepted at #ICWSM 25!

Looking forward to share more insights from the dataset in Copenhagen.

The dataset will be updated soon: bug fixes and additional sentiment scores.

🚨New method in studying old relations! 🚨
Are Dark Triad traits differently related to self-reported and task-performance domains of emotional functioning?
👇
t.co/Wxi9MdrMWW

Thanks to Beata Grabovac and @atomashevic.bsky.social for the fruitful collaboration

Apply to the "Software and Data for Supporting Network Science" satellite workshop at @netsciconf.bsky.social before Feb. 7th! Got some cool network data or software to share? This is the satellite for you! Submission form at netsci.nascol.net/cfa

PostDoc at Central European University in
#datasci #netsci #compsocsci #inclusive 🏳️‍🌈🇪🇺
@weareceu.bsky.social

We have released the first version of MADOC!

Huuge dataset with ~9M posts, 236M comments, and 23M users from Reddit, Voat, Bluesky, and Koo.

MADOC mixes benign and controversial/banned communities, ideal for the study of hate speech, user migrations, and community moderation.

pymadoc Python package to download and combine parts of MADOC dataset

If you want to take a quick look at (subset) of MADOC, Python and R packages are available:

pypi.org/project/pyma...

github.com/atomashevic/...

Dataset is available on Zenodo:
zenodo.org/records/1463...

Preprint describing MADOC is on ArXiv:
arxiv.org/abs/2501.12886

We have released the first version of MADOC!

Huuge dataset with ~9M posts, 236M comments, and 23M users from Reddit, Voat, Bluesky, and Koo.

MADOC mixes benign and controversial/banned communities, ideal for the study of hate speech, user migrations, and community moderation.

The Machiavellian Approach and Avoidance Questionnaire: Further Validation and Evidence of Cross-National Validity Researchers on antagonistic personality traits debate about an appropriate measurement approach to Machiavellianism. One measure intended to resolve this discourse, the Machiavellian Approach and A...

🎉Thrilled to share our new paper!
We tested cross-national validity of the Machiavellian Approach and Avoidance Questionnaire (MAAQ) across Germany, Canada, United Kingdom, and Serbia 👇
www.tandfonline.com/doi/full/10....

Coming soon: Pilet: Opensource, Modular, Portable Mini Computer Where retro meets modern, powered by Raspberry Pi 5 with 7-hour battery life

this looks perfect 😍

ssh into server and you can do meaningful work + have strong, strong retro vibes!

So many new people on Bluesky! Who should I follow? | Johannes B. Gruber If there is one development at the moment which I full heartedly enjoy reading about it’s that the remains of what was once called Twitter is seeing a large E𝕏odus. Since a certain billionaire has tak...

So many new people on here! Who should I follow?

I tried to narrow it down at least a little with #rstats

johannesbgruber.eu/post/2024-11...

The success of every online social platform depends on the quality of cat content.

Zlatko says hello to Bluesky 🐈

go.bsky.app/MMeFFH6

GitHub - atomashevic/transforEmotion: Sentiment Analysis for Text, Image and Video Using Transformer Models Sentiment Analysis for Text, Image and Video Using Transformer Models - atomashevic/transforEmotion

transforEmotion 0.1.5 in live on Github and CRAN.

github.com/atomashevic/...

- Resolved Youtube download issues by switching to pytubefix library

- Added rag function for retrieval-augmented generation using various transformer models.

GitHub - atomashevic/transforEmotion: Sentiment Analysis for Text, Image and Video Using Transformer Models Sentiment Analysis for Text, Image and Video Using Transformer Models - atomashevic/transforEmotion

transforEmotion 0.1.5 in live on Github and CRAN.

github.com/atomashevic/...

- Resolved Youtube download issues by switching to pytubefix library

- Added rag function for retrieval-augmented generation using various transformer models.

New version of "Dynamic Exploratory Graph Analysis of Emotions in Politics" is up on PsyArXiv!

We switch from CNN to CLIP with transforEmotion and now we have a clear two-dimensional network of emotion co-occurence.

Preprint: osf.io/preprints/ps...

Github: github.com/atomashevic/...

rollama: An R package for using generative large language models... rollama is an R package that wraps the Ollama API, which allows you to run different Generative Large Language Models (GLLM) locally. The package and learning material focus on making it easy to...

📢New preprint📢

As researchers embrace the next revolution in computational social science, the arrival of GLLMs (generative AI), there is a critical need for open-source alternatives. We introduce one with (r)Ollama and explain why it is important!

arxiv.org/abs/2404.07654

Lost in Transformation: The Horror and Wonder of Logit | Robert Kubinec Perhaps no other subject in applied statistics and machine learning has caused people as much trouble as the humble logit model. Most people who learn either subject start off with some form of linear...

🚨 Blog post:

Lost in Transformation: The Horror and Wonder of Logit

I use the analogy of 🗺️↔️🌏 to clarify why logit is such a good model for binary outcomes.

tl; dr: OLS works on flat maps; logit works on globes.

statsky polisky econsky academicsky

www.robertkubinec.com/post/flat_ea...

But package is primarily built for #Rstats users with limited or zero knowledge of #Python.

The package mimics and simplifies the usage of transformers library in Python, so you can perform inference, run small tasks in a few lines of R code without worrying what is happening in the background.

I'm mostly working on emotion detection in videos, so I'm looking to enhance the package with more transformer-based model: speech to text, emotional recognition in speech...

The idea is to have a complete pipeline to process video and extract multiple time series of emotion scores.

GitHub - atomashevic/transforEmotion: Sentiment Analysis for Text, Image and Video Using Transformer Models Sentiment Analysis for Text, Image and Video Using Transformer Models - atomashevic/transforEmotion

Haven't posted here in a while, so here's a small update.

I've spent last few months developing and mantaining transforEmotion #Rstats package

github.com/atomashevic/...

It's a package for sentiment analysis for text, image and video in R using HuggingFace #transformers API

My body is not a temple its a deep space station occupying both a strategically pivotal location and also near an actual (celestial) temple.

Replication games: how to make reproducibility research more systematic In some areas of social science, around half of studies can’t be replicated. A new test-fast, fail-fast initiative aims to show what research is hot — and what’s not. In some areas of social sci...

Systematic replications across multiple social sciences www.nature.com/articles/d41... #openscience

☀️ It was really sunny in Novi Sad yesterday. We enjoyed a photoshoot for the IDIOMATIC project and had our last get-together before I depart for the US.