Arefeh Kazemi

—

Unclaimed · AI-structured profileOpenAlex

Arefeh Kazemi: NLP researcher auditing whether LLM-generated dialogues preserve authentic social dynamics

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-21) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.

Research profile

Output—

Insufficient data: the catalog holds only one paper for this name, so no stable publication-volume or venue-distribution measure exists yet. This axis would count total research outputs across venues and years once a fuller bibliographic record is linked.

Impact—

Not scored: no citation counts, h-index, or downstream-usage signals are available in this snapshot. This axis would quantify scholarly influence — citations, uptake, and reuse of the cataloged work — once impact data is attached to the profile.

Focus100/100

The focus dimension scores 100, the maximum. The underlying field distribution is perfectly concentrated: the single cataloged paper is classified entirely under natural language processing / computational social science, giving that field a 100% share of the record. With only one reviewed publication, this is a clean but narrow signal — it establishes a clearly defined NLP-and-computational-social-science identity rather than a demonstrated, sustained specialization across many outputs.

Collaboration32/100

The collaboration dimension scores 32, computed from 4 distinct coauthors in the record. The network is compact: Hamza Qadeer, Sinan Asci, Joachim Wagner, and Brian Davis recur as frequent collaborators, and the cataloged 2026 paper is first-authored, suggesting a small stable team rather than a large rotating consortium. The low score reflects limited breadth of co-authorship visible so far, not weak teamwork.

Activity—

Cannot be scored: only one publication year (2026) appears in the record, so publishing cadence and recent-output trends cannot be computed. This axis would measure yearly output rhythm and research momentum across a longer publication history.

Reproducibility55/100

The reproducibility dimension scores 55, derived from a reproducibility mean of 55 over 1 assessed paper. That paper — the 2026 cyberbullying synthesis study — carries a medium reproducibility readiness rating. Evolve's review notes a credible evaluation framework but also that only 10 authentic conversations anchor the analysis, a data-scale constraint that plausibly shapes the moderate score. With a single-paper basis, treat this as an indicative, not settled, signal.

Research summary

Arefeh Kazemi is an early-career researcher working at the intersection of natural language processing and computational social science. The Evolve catalog currently contains one first-authored paper: a 2026 arXiv preprint asking whether the social patterns observed in authentic cyberbullying conversations survive when dialogues are generated by large language models. Evolve's review finds that synthetic dialogues preserve coarse conversational structure but distort magnitude, participant roles, and escalation dynamics, while judging the paper's evaluation framework credible — with the caveat that only ten authentic conversations anchor the analysis. The paper's keywords span cyberbullying detection, synthetic data generation, LLM evaluation, social realism, conversation analysis, and toxicity detection, and its reproducibility readiness is rated medium (reproducibility dimension 55, computed from a single paper). Profile metrics show extreme topical concentration: the focus dimension scores 100, with all cataloged output classified under one field. The collaboration dimension scores 32, reflecting a compact network of four distinct coauthors — Hamza Qadeer, Sinan Asci, Joachim Wagner, and Brian Davis. Output, impact, and activity dimensions lack sufficient data and are not scored. Overall this is an early-stage profile built from a single reviewed publication plus bibliometric aggregates.

Collaboration pattern

The visible collaboration footprint is a compact four-person core: Hamza Qadeer, Sinan Asci, Joachim Wagner, and Brian Davis, the coauthors of the single cataloged 2026 paper, which Kazemi led as first author. With 4 distinct coauthors and a collaboration dimension score of 32, the pattern suggests a small, stable working team rather than a broad, rotating network; nothing larger can be inferred from the current record.

Research keywords

cyberbullying detectionsynthetic data generationLLM evaluationsocial realismconversation analysistoxicity detectionnatural language processingcomputational social scienceLLM-generated dialoguecyberbullying dynamics

Highlights

Synthetic cyberbullying dialogues keep the shape, lose the dynamics
Evolve's review of the 2026 first-author arXiv preprint finds LLM-generated cyberbullying conversations preserve coarse conversational structure yet distort magnitude, participant roles, and escalation compared with authentic dialogues.
A credible synthetic-data audit framework, anchored on ten real conversations
The framework comparing LLM-generated and authentic dialogues is judged credible by the review, but the evidence base is thin: only 10 authentic conversations anchor the analysis, a limitation the verdict explicitly flags.
First-author arXiv preprint rated medium for reproducibility
The single cataloged output (2026, first author, arXiv) carries a medium reproducibility readiness rating and is the sole basis for the profile's reproducibility score of 55 (reproPapers = 1).

Timeline

2026First-author arXiv preprint comparing cyberbullying dynamics in LLM-generated and authentic dialogues enters the Evolve catalog.
2026Evolve review concludes synthetic dialogues keep coarse structure but distort magnitude, roles, and escalation; reproducibility readiness is rated medium.
2026Four coauthors on the paper — Hamza Qadeer, Sinan Asci, Joachim Wagner, Brian Davis — constitute the profile's entire visible collaboration network (4 distinct coauthors).

Papers in the Evolve catalog

2026-09-17first author
Do Social Patterns Hold in Synthetic Data? Analyzing Cyberbullying Dynamics in LLM-Generated and Authentic Dialogues

Frequent co-authors

Hamza Qadeer×1 Sinan Asci×1 Joachim Wagner×1 Brian Davis×1

About this profile

This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.

All bibliometric dimensions rest on a very small base: one reviewed paper, one field, and four coauthors; treat scores as indicative only.
Institution, country, and ORCID are absent from the source data; the profile is name-resolved from public bibliographic records and may require identity confirmation.
No citation counts, h-index, or venue-tier data are available, so the output, impact, and activity dimensions are unscored rather than low.
The reproducibility score (55) and focus score (100) derive from a single 2026 arXiv paper and will shift as the catalog grows.