First-author researcher at UWA: NLP mining of GenAI user feedback plus security measurement of long-horizon LLM agents
2
papers here
2
reviewed by Evolve
2
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) 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—
Output is not yet scored: the bibliometric feed behind this profile lacks publication-volume measures for this name. This axis would normally count total papers, venue types, and first- or corresponding-author share over time.
Impact—
Impact is unscored because citation counts and derivative indicators are not available in the data behind this profile. The axis would measure citations, citing works, and uptake of the two 2026 preprints once they accrue citations.
Focus0/100
Focus scores 0 because the catalogued output is split evenly across two distinct fields: human-computer interaction / computational social science (NLP) at 0.5 and AI security (LLM agent security) at 0.5. Neither area dominates the record, so the distribution shows no concentration. This is a two-track pattern — mining generative-AI user feedback with BERTopic and RoBERTa on one track, measuring LLM agent security exposure on the other — rather than a single-line research agenda.
Collaboration28/100
Collaboration scores 28, on the low end. The record lists only 3 distinct coauthors — Umme Nusrat Jahan, Shouvaggo Sharif Shammo, and Nur Al Hasan Haldar — across the two first-author 2026 preprints, and neither paper carries a corresponding-author role. This points to small-team or early-stage work rather than large multi-institution consortia; the collaboration footprint should widen as the publication record grows.
Activity—
Activity is not scored: the underlying feed has no yearly publication or submission cadence for this name beyond the two 2026 preprints. The axis would track output per year and recency of contributions to gauge ongoing research pace.
Reproducibility55/100
Reproducibility averages 55 across the 2 papers with reproducibility assessments, a mid-range score. Evolve rates both preprints medium for repro readiness: the BERTopic+RoBERTa review-mining pipeline is validated but relies on app-store data subject to self-selection bias, while the influence-distance analysis on AgentDojo logs is careful and honest though confined to a single benchmark.
Research summary
Md Jafrin Hossain is a researcher at The University of Western Australia whose Evolve-catalogued output comprises two first-author arXiv preprints dated 2026, splitting evenly (0.5/0.5) between human-computer interaction / computational social science (NLP) and AI security for LLM agents. The first mines 17,012 generative-AI app-store reviews with a validated BERTopic plus RoBERTa pipeline, surfacing friction and polarization patterns around user trust that Evolve judges credible, while flagging self-selection bias in review data. The second introduces influence distance on provenance graphs to measure security exposure in long-horizon LLM agents, showing on AgentDojo logs that step count overstates attack proximity for 96.9% of pairs. Evolve rates both preprints medium for reproducibility, and the reproducibility axis averages 55 across those two assessed papers. Collaboration metrics are thin: only 3 distinct coauthors appear, the collaboration score is 28, and neither paper lists the researcher as corresponding author. Output, impact, and activity axes remain unscored for insufficient data. The result is an early-stage, arXiv-centered profile of a two-field researcher rather than a complete publication record.
Collaboration pattern
Both catalogued papers are first-author and non-corresponding, drawn from a small pool of 3 distinct coauthors: Umme Nusrat Jahan, Shouvaggo Sharif Shammo, and Nur Al Hasan Haldar. The collaboration score of 28 sits low, consistent with small-team arXiv preprints rather than multi-institution consortia. The recurring trio appears alongside the researcher's own affiliation at The University of Western Australia; deeper co-author affiliation data is out of scope.
Research keywords
Highlights
Mining 17,012 generative-AI app reviews with a validated BERTopic+RoBERTa pipeline
Friction and polarization in how users talk about generative AI
Influence distance: step count overstates attack proximity in LLM agents
Timeline
- 2026Posts first-author arXiv preprint 'What Users Think of Generative AI', an NLP analysis of 17,012 app-store reviews using BERTopic and RoBERTa.
- 2026Posts first-author arXiv preprint 'Structurally Close, Temporally Distant' on measuring security exposure in long-horizon LLM agents via influence distance.
- 2026Catalogued output for the year totals two first-author preprints split evenly across NLP user studies and LLM agent security, with 3 distinct coauthors.
Papers in the Evolve catalog
Frequent co-authors
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.
- Bibliometric dimensions are computed from an OpenAlex-based snapshot; the output, impact, and activity axes lack sufficient data and are left unscored.
- Catalog coverage is limited to two 2026 arXiv preprints reviewed by Evolve; peer-reviewed venues, grants, and non-paper outputs are out of scope.
- The focus score of 0 reflects an even 0.5/0.5 split across two fields, not a lack of research activity.
- Common-name risk: the profile is name-resolved from public bibliographic data and anchored to ORCID 0000-0002-3149-9460; country-level affiliation metadata was not provided.