Sign in

Nir Grinberg

@nirg.bsky.social
185 followers 174 following 28 posts

Assistant prof. at BGU in the field of Computational Social Science.

PostsRepliesMedia
Nir Grinberg @nirg.bsky.social · 07/05/2026
8/8 🎉 Super proud of this computational social science work out of our lab at @bengurionuniv.bsky.social ! Huge thanks to my co-authors Itay Razumenko and Arnon Sturm. Read the full paper at arxiv.org/abs/2604.18041 If you're attending ACL, come say hi 👋 #NLProc #ACL2026 #LegalTech
arxiv.org
JudgeMeNot: Personalizing Large Language Models to Emulate Judicial Reasoning in Hebrew
Despite significant advances in large language models, personalizing them for individual decision-makers remains an open problem. Here, we introduce a synthetic-organic supervision pipeline that trans...
041
Nir Grinberg @nirg.bsky.social · 07/05/2026
7/8 🛤️ An important caveat is that our work successfully emulates sentence-level, granular reasoning, which is considerable easier that modeling the full trajectory of case.
110
Nir Grinberg @nirg.bsky.social · 07/05/2026
6/8 While all data analyzed here is entirely public, by public officials operating in their official capacity for public scrutiny, the sensitivity of the domain + personalization requires further ethical considerations beyond simple regulatory compliance as detailed in Section 9 of the paper.
100
Nir Grinberg @nirg.bsky.social · 07/05/2026
5/8 📊 The results? CoLA consistently outperforms all baselines. RAG captures surface-level style (like POS distributions), but fails in reasoning. CoLA-generated reasoning is so faithful that an LLM trained specifically on each judge cannot tell the difference from the actual judge. 🤯
Personalization performance across BLEU, RougeL, BertScoreF1, and POS-JSD.
Model performance in the Author Discernibility Task.
100
Nir Grinberg @nirg.bsky.social · 07/05/2026
4/8 🧠 For training, we used Chain-of-LoRA (CoLA) in two steps: 1️⃣ Causal LM tuning (adapts to general writing style). 2️⃣ Instruction tuning (focuses on legal reasoning). All using a single GPU and Gemma-4B model, which was super fast.
The ablation study showing performance as a function of training size and LoRA rank
100
Nir Grinberg @nirg.bsky.social · 07/05/2026
3/8 🛠️ The bottleneck for personalization is the lack of reasoning supervision. To fix this, we built a "synthetic-organic" pipeline. We use an agentic workflow to extract raw reasoning sentences from unstructured Israeli verdicts and generate targeted synthetic questions that prompted them.
100
Nir Grinberg @nirg.bsky.social · 07/05/2026
2/8🧑‍⚖️ Judges aren't interchangeable; they have distinct judicial philosophies. Yet, most Legal NLP treats them as one generic voice. The result? AI "mode collapse." Standard models generate long, encyclopedic text that completely misses the human legal fingerprint.
100
Nir Grinberg @nirg.bsky.social · 07/05/2026
🚨 Excited to share our new paper to appear in the Findings of ACL 2026! "JudgeMeNot" asks: Can LLMs learn the specific reasoning signature of an individual judge? We tackle this in the low-resource setting of Hebrew case law. 🔗 arxiv.org/abs/2604.18041 🧵👇
The illustrative diagram shows our approach for training personalized LLMs to emulate Judge A (textualist) vs. Judge B (purposivist) and how those trained models might respond to a legal query.
152
Reposted by Nir Grinberg
Ben-Gurion University of the Negev @bengurionuniv.bsky.social · 09/01/2026
Can social media detect economic shocks before official data does? A new PNAS Nexus study led by @nirg.bsky.social and Samuel Fraiberger shows that AI models tracking job-loss disclosures on social media can predict U.S. unemployment insurance claims up to two weeks early,
101
Nir Grinberg @nirg.bsky.social · 13/01/2026
Credit is also due to @davidlazer.bsky.social for prompting Sam & I to think about this problem 7(!) years ago ;) 12/fin
020
Nir Grinberg @nirg.bsky.social · 13/01/2026
Kudos to my wonderful co-authors Do Lee linkedin.com/in/do-lee and @manueltonneau.bsky.social (both on the job market!), Boris Sobol il.linkedin.com/in/boris-sobol and Sam Fraiberger samuelfraiberger.com. 11/N
100
Nir Grinberg @nirg.bsky.social · 13/01/2026
Yet platform data-access policies increasingly block this potential. Whether platforms or regulators will enable change in the coming years is a core policy question. 10/N
110
Nir Grinberg @nirg.bsky.social · 13/01/2026
There is clear public value here, potentially extending to other countries, especially where official statistical systems are under-developed. 9/N
100
Nir Grinberg @nirg.bsky.social · 13/01/2026
Why this matters? Beyond forecasting, this approach can provide early warnings, surface local labor market stress hidden by national averages, and help flag measurement issues in real time. 8/N
110
Nir Grinberg @nirg.bsky.social · 13/01/2026
Key finding 3: This also works at the state and city (!) level, including "holdout cities" where official UI numbers are sparse or irregularly updated. As expected, accuracy scales with platform penetration and unemployment shocks. 7/N
110
Nir Grinberg @nirg.bsky.social · 13/01/2026
Key finding 2: Our approach consistently outperforms industry consensus forecasts and can improve predictions of US UI claims up to two weeks ahead of official releases. That’s two weeks of additional lead time for policymakers. 6/N
110
Nir Grinberg @nirg.bsky.social · 13/01/2026
Key finding 1: Capturing linguistic diversity matters. Training LLMs with active learning lets us detect many more ways people talk about job loss, producing a far more representative sample of unemployed users than existing approaches. 5/N
110
Nir Grinberg @nirg.bsky.social · 13/01/2026
We combine JoblessBERT (an encoder LLM developed in previous work aclanthology.org/2022.acl-lon... which detects ~3× more employment-related content without sacrificing precision) with post-stratification using inferred demographics to correct for platform bias. 4/N
aclanthology.org
Multilingual Detection of Personal Employment Status on Twitter
Manuel Tonneau, Dhaval Adjodah, Joao Palotti, Nir Grinberg, Samuel Fraiberger. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.
100
Nir Grinberg @nirg.bsky.social · 13/01/2026
So we ask a hard question economic actors and policymakers rightly worry about: Can skewed social media data be turned into trustworthy indicators of unemployment? Can we produce robust predictions across geography ✅, time ✅, demography ✅, and forecasting horizon ✅ ? 3/N
100
Nir Grinberg @nirg.bsky.social · 13/01/2026
Why this matters: In March 2020, weekly unemployment insurance claims jumped from 278K to nearly 6 million in two weeks. As official data lagged, policymakers were flying blind about where the shock was hitting and who was being affected. 2/N
100
Nir Grinberg @nirg.bsky.social · 13/01/2026
New paper out in @pnasnexus.org: We show how skewed social media data can still be used to reliably estimate unemployment, not just nationally but down to the city level. 📈 doi.org/10.1093/pnas... 1/N
doi.org
Can social media reliably estimate unemployment?
Abstract. Digital trace data hold tremendous potential for measuring policy-relevant outcomes in real-time, yet its reliability is often questioned. Here,
1114
Reposted by Nir Grinberg
Kevin Yang @yang3kc.bsky.social · 17/01/2025
Introducing “DomainDemo: a dataset of domain-sharing activities among different demographic groups on Twitter.” Today, we release five derived metrics of over 129,000 domains, quantifying their characteristics such as geographical reach and audience partisanship. 1/3
1155
Reposted by Nir Grinberg
Ross Dahlke @rossdahlke.bsky.social · 17/01/2024
Incels (involuntarily celibates) are increasingly using violent language, particularly non-directed violent language in the largest incel forum, finds @danielmatter.bsky.social @miriamschirmer.bsky.social @nirg.bsky.social @jurgenpfeffer.bsky.social arxiv.org/abs/2401.02001
Close to Human-Level Agreement: Tracing Journeys of Violent Speech in Incel Posts with GPT-4-Enhanced AnnotationsFigure 1: Linear Regression between time and share of violent posts.Figure 2: Linear Regression between time and category of directedness.
0128
Nir Grinberg @nirg.bsky.social · 05/12/2023
Awesome! We’d love to hear what you and your students think about it.
010
Nir Grinberg @nirg.bsky.social · 05/12/2023
We are also grateful for comments received on earlier versions of this work from Diyi Liu, Eran Amsalem @patyrossini.bsky.social Alon Zoizner, and @orentsur.bsky.social & for funding from European Research Council (ERC), Israel Science Foundation (ISF) and BGU's Data Science Center.
041
Nir Grinberg @nirg.bsky.social · 05/12/2023
Big shout-out to the people whose work enabled this research, including @sdmccabe.com @jongreen.bsky.social @davidlazer.bsky.social Magdalena Wojcieszak @jatucker.bsky.social Subhayan Mukerjee @ylelkes.bsky.social @kthorson.bsky.social @chriswells.bsky.social (pls tag others if missing). 6/
160
Nir Grinberg @nirg.bsky.social · 05/12/2023
Finally, looking at the demographic composition of consumption "types", we find that the media-oriented clusters (exc. superconsumers) have older individuals, more women, and more registered Democrats. 5/
Sociodemographic characteristics among different political exposure types. Sample averages are marked in a gray dashed line. Ninety-five percent bootstrapped CIs are shown (mostly occluded due to their small size). CI = confidence interval.
121
Nir Grinberg @nirg.bsky.social · 05/12/2023
Even when putting aside the more extreme "media superconsumers", the two media-oriented clusters (which are ~20% of the population), get half or more of their political content *directly* from media organizations and journalists, without any mediation from peers. 4/
121
Nir Grinberg @nirg.bsky.social · 05/12/2023
Americans also vary in the breakdown of actors that populate their feeds, but interestingly, the bulk of the population gets half or more of their political exposure from *traditional sources*—media organizations, journalists, and politicians. 3/
The composition of political exposure across clusters. The share of politics curated by different actor types (y-axis) across clusters (x-axis). Darker-colored bars represent direct exposure to media organizations, journalists, politicians, OLs, and social peers. Lighter-colored bars represent indirect exposure to media organizations, journalists, politicians, or opinion leaders through social peers. OL = opinion leader.
121
Nir Grinberg @nirg.bsky.social · 05/12/2023
People's political feeds mostly map onto 8 distinct types that vary in the amount of politics they get, both in absolute #'s and as % the feed as a whole. Still, for nearly 90% of the population, about 1 in 12 posts from their network are political. Quite an engaged public! 2/
Prototypical types of individual political exposure. Each point in panel (A) represents the political exposure of a single panel member, reduced to two dimensions using the UMAP algorithm, and colored by the cluster assignment obtained from HDBSCAN. Panel (B) shows the median number of political tweets available to individuals per day (left bars), and their percentage out of all tweets available to them on Twitter (right bars). Cluster labels and their share in the population are specified on the x-axis. Colors are consistent between the two figure panels. Ninety-five percent bootstrapped CIs are omitted from the figure due to their small magnitude, which are upper bounded by twenty-seven exposures to tweets and 0.28 percent, respectively. OL = opinion leader; CI = confidence interval; UMAP = Uniform Manifold Approximation and Projection.
131
Nir Grinberg @nirg.bsky.social · 05/12/2023
🚨New paper🚨 out in the International Journal of Press/Politics w/ Assaf Shamir and @jenny-oser.bsky.social 🎉 Here's what we learned from studying the composition of political content available to 600k+ registered U.S. voters on Twitter during the 2020 election. doi.org/10.1177/1940... 🧵👇
12011