Xue-Yong Fu
—
NLP researcher advancing fine-tuning strategy selection and evaluation of LLM agents for customer support
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-14) 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—
No publication-volume data is in scope, so output is unscored. This axis would measure the count and cadence of indexed publications over a career; with only two cataloged papers available, the sample is too small to compute a score.
Impact—
Citation and influence metrics are not available in this snapshot, so impact is unscored. It would measure citation counts, citing works, and field-weighted influence; the two arXiv preprints reviewed here carry no citation data in scope.
Focus0/100
Focus scores 0 because the field distribution splits evenly — 0.5 general natural language processing and 0.5 LLM agent evaluation — across the two cataloged papers. With so small a sample, this may reflect label granularity rather than genuine breadth: both papers target the same applied customer-support LLM setting, so the axis should be read as unresolved, not as scattered interests.
Collaboration56/100
Collaboration scores 56 based on 16 distinct coauthors in the available records, with twelve recurring collaborators including Shashi Bhushan TN, Md Tahmid Rahman Laskar, and Gundeep Singh. On both cataloged papers Fu appears as a middle-listed, non-corresponding author, consistent with contributing inside larger applied teams rather than leading small groups. The moderate score reflects a mid-sized network centered on one product-oriented research line rather than a broad cross-institution portfolio.
Activity—
Recent-activity signals (publication recency and yearly cadence) are insufficient, so activity is unscored. It would measure how consistently the researcher publishes over recent years; both cataloged papers are dated 2026, giving no multi-year basis.
Reproducibility38/100
Reproducibility scores 38, the mean readiness across two reviewed papers rated medium and low. Evolve's verdicts flag shared limitations: reliance on proprietary customer-support data, small evaluation sets (one study reports 0/77 holistic success), and no statistical significance testing. The controlled study designs are a strength, but public artifacts would be needed to lift this axis.
Research summary
Xue-Yong Fu is a natural language processing researcher whose profile is name-resolved from public bibliographic data; the current snapshot lists no institution, country, or ORCID. The Evolve catalog contains two 2026 arXiv papers with Fu as a middle-listed, non-corresponding author, both applied to customer-support LLMs. The first is a controlled comparison of fine-tuning strategies across 13 support models, finding multi-task full fine-tuning the strongest single adapted model across settings spanning LoRA, parameter-efficient fine-tuning, multi-task learning, model merging, and catastrophic forgetting. The second diagnoses why supervised fine-tuning can raise next-turn dialogue metrics without improving end-to-end workflow success, reporting 0/77 holistic success and analyzing tool calling under gold-history evaluation. The field distribution splits evenly at 0.5 general NLP and 0.5 LLM agent evaluation, and the reproducibility score of 38 averages two papers rated medium and low, with recurring reviewer concerns about proprietary data, small evaluation sets, and missing significance tests. Collaboration records list 16 distinct coauthors, twelve of them recurring. Given the small sample, output, impact, and activity remain unscored.
Collaboration pattern
The available records show 16 distinct coauthors and a stable recurring group of twelve, including Shashi Bhushan TN, Md Tahmid Rahman Laskar, Gundeep Singh, Kevin Sanders, and Shayna Gardiner. Fu sits in the middle of the author list on both cataloged papers without a corresponding-author role, fitting a pattern of contributing to larger applied teams on customer-support LLM research rather than leading independent small-team projects.
Research keywords
Highlights
Fine-tuning strategy benchmark across 13 customer-support LLMs
Next-turn gains do not make better agents
Timeline
- 2026Published a controlled comparison of fine-tuning strategies across 13 customer-support LLMs on arXiv
- 2026Co-authored a diagnostic study showing next-turn metric gains fail to translate into workflow success (0/77 holistic)
- 2026Both papers entered the Evolve catalog, covering fine-tuning strategy selection and LLM agent evaluation
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.
- Name resolved from public bibliographic data; no institution, country, or ORCID is available, and overlap with other similarly named researchers cannot be excluded.
- All bibliometric scores derive from a very small sample: two reviewed 2026 arXiv papers in the Evolve catalog.
- Output, impact, and activity axes are unscored because citation counts and publication-volume data are not in scope.
- Both cataloged papers are arXiv preprints; no peer-reviewed venue or affiliation data appears in this snapshot.