Yiwen Jiang

Monash University · Australia

Unclaimed · AI-structured profileOpenAlex

Monash NLP researcher building CBT-aligned counseling datasets and benchmarks for LLM safety and personalization

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-17) 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—

Not yet scored: publication-volume data is insufficient in the current OpenAlex snapshot, where h-index and i10-index are recorded as 0 and no cumulative works series is available. This axis would measure total paper count and its yearly trajectory once the record fills in.

Impact—

Not yet scored: the snapshot contains no citation accumulation, so influence cannot be assessed. This axis would measure citation counts, citing works, and downstream uptake in mental-health NLP and LLM evaluation once citations to the 2026 preprints begin to accrue.

Focus100/100

Scores 100/100. The topic distribution is tightly concentrated: speech and dialogue systems (3 works), topic modeling (2), and ethics and social impacts of AI (2), with the profiled field pinned to mental-health conversational AI. All three reviewed 2026 papers converge on conversational LLMs — CBT counseling alignment, dialogue stigma, and personalization — a deliberately narrow, clinically oriented line rather than scattered output.

Collaboration67/100

Scores 67/100, from 28 distinct coauthors in the snapshot and 12 recurring partners including Stephanie Fong, Zimu Wang, Hongxi Yang, Zongyuan Ge, and Dominic Dwyer. All three reviewed 2026 papers are multi-author, with the researcher first author once (VIBE-Bench) and middle author twice, never corresponding — a team-embedded contributor pattern rather than solo-led projects.

Activity57/100

Scores 57/100. The snapshot records 6 works in the recent three-year window, and the counts-by-year series places all 6 in 2026, so visible activity is recent and front-loaded: three arXiv preprints in the Evolve catalog (CBT counseling dataset, stigma benchmark, personalization benchmark) plus other work. No earlier years appear in the series, which caps this axis.

Reproducibility55/100

Scores 55/100, a mean (reproMean 55) computed from 1 paper assessed for reproducibility. The three catalogued preprints are each tagged medium readiness: datasets and benchmarks are released as artifacts, but Evolve verdicts flag fully LLM-synthetic sessions, GPT-4o-judged evaluation, synthetic self-labeled data, and a 128-sample gold set without significance tests.

Research summary

Yiwen Jiang is a natural language processing researcher at Monash University (Australia) whose work centers on conversational LLMs in clinically and socially sensitive settings. The Evolve catalog holds three 2026 arXiv preprints. StratCBT ("Steering the Compass") introduces the first large strategy-aligned CBT dataset of 9.7K counseling sessions, though it is fully LLM-synthetic and GPT-4o-judged. SDARE-Bench is presented as the first benchmark for LLM conversational stigma detection and response in dyadic and group dialogue, with striking group-pressure experiments. VIBE-Bench, first-authored, proposes PRCM and shows personalized LLMs failing cross-concept preference mapping across Big Five, Holland RIASEC, and emotion-regulation profiles. Bibliometrically, the OpenAlex snapshot records 6 works in the recent three-year window — all in 2026 — 28 distinct coauthors, and topics concentrated on speech and dialogue systems (3), topic modeling (2), and ethics and social impacts of AI (2), yielding a focus score of 100/100. Collaboration (67) and activity (57) sit mid-range; reproducibility averages 55 from one assessed paper. Output and impact are unscored because no citation record has accumulated yet. The overall pattern is an early-stage but sharply focused agenda: building datasets and benchmarks that probe whether LLMs behave safely, faithfully, and personally in dialogue.

Collaboration pattern

The snapshot lists 28 distinct coauthors, with 12 recurring partners including Stephanie Fong, Zimu Wang, Hongxi Yang, Xiangyu Zhao, Zhongxing Xu, Jiahe Liu, Zongyuan Ge, Dominic Dwyer, Yaling Shen, Hiu Weh Naomi Chow, Heung Ying Lai, and Qingyang Xu. Across the three reviewed 2026 papers, the researcher is first author once (VIBE-Bench) and a middle author twice, never corresponding — a team-embedded contributor who leads specific benchmark efforts within larger groups.

Research keywords

Mental health NLPCognitive Behavioral TherapySynthetic dialogue generationStrategy-aligned generationLLM-as-judge evaluationLLM safetyStigma detectionMulti-party conversationSycophancyPersonalized LLMsPreference reasoningBenchmark construction

Highlights

StratCBT: a 9.7K-session strategy-aligned CBT counseling dataset
First large strategy-aligned CBT dataset for dynamic psychological counseling conversations, built with 9.7K sessions via synthetic dialogue generation and evaluated with LLM-as-judge; Evolve notes it is fully LLM-synthetic and GPT-4o-judged, so realism claims need caution.
SDARE-Bench: stigma detection and response in dyadic and group dialogue
First benchmark targeting LLM conversational stigma detection and response in dyadic and group dialogue; its multi-party group-pressure tests yield striking results, though synthetic self-labeled data and moderate classifier F1 weaken the conclusions.
VIBE-Bench (first author): when profiles don't mean preferences
First-authored benchmark introducing PRCM, showing personalized LLMs fail cross-concept preference mapping across Big Five personality, Holland RIASEC, and emotion-regulation profiles; the design is compelling but the 128-sample gold set lacks significance tests.

Timeline

2026Posted "Steering the Compass" (StratCBT) on arXiv: a 9.7K-session strategy-aligned CBT counseling dataset for dynamic psychological counseling conversations.
2026Released SDARE-Bench on arXiv, the first benchmark evaluating LLM stigma detection and response in dyadic and group dialogue.
2026First-authored VIBE-Bench on arXiv, introducing PRCM and testing personalized LLMs when profiles don't mean preferences.
2026OpenAlex snapshot records 6 works in 2026 with 28 distinct coauthors; the focus dimension reaches 100/100.

Papers in the Evolve catalog

Frequent co-authors

Stephanie Fong×3 Zimu Wang×2 Hongxi Yang×2 Xiangyu Zhao×2 Zhongxing Xu×2 Jiahe Liu×2 Zongyuan Ge×2 Dominic Dwyer×2 Yaling Shen×1 Hiu Weh Naomi Chow×1 Heung Ying Lai×1 Qingyang Xu×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.

Bibliometrics come from an OpenAlex snapshot only; no ORCID is linked, and the profile is name-resolved from public bibliographic data — "Yiwen Jiang" is a common name.
h-index and i10-index are recorded as 0 and no citation counts exist, so the output and impact dimensions are unscored rather than low.
The counts-by-year series contains only 2026 (6 works); no earlier publication years are available, so the activity axis reflects a very short observable window.
Reproducibility averages 55 from 1 assessed paper; all three catalogued items are arXiv preprints and peer-reviewed venues are not yet recorded.