Benchmarking aspect-conditioned sentiment classification of Uzbek healthcare feedback with auxiliary medical entity supervision

Abstract Background A healthcare comment can express different sentiments about waiting time, staff interactions, and other service aspects. We benchmarked sentiment classification for a supplied aspect in Uzbek online healthcare feedback and examined how scores differed with auxiliary medical entity supervision. Methods The main comparison covered 13 configurations under a reported group-stratified five-fold protocol, with text–aspect pairs from the same normalized comment kept in one fold. BERTbek and XLM-R were compared with lexical, subword, and transformer baselines. Two additional review-only controls supported ablations of aspect conditioning, aspect fusion, and sequential and joint medical named entity recognition training. Macro-averaged F1 on six service aspects was the primary endpoint. The corresponding score on the full medical-domain benchmark was secondary. Results BERTbek with aspect fusion and joint medical entity supervision had the highest reported macro-averaged F1 on the service subset (84.91) and full benchmark (82.74), on a 0–100 scale. The differences from standard paired-input BERTbek were 0.84 and 1.83 points, respectively. The direct differences from BERTbek with aspect fusion alone were 0.50 and 0.78 points, respectively. The largest descriptive differences from the paired-input baseline among service aspects were for waiting time and price. Conclusions The reported point estimates support further evaluation of medical entity supervision for aspect-conditioned sentiment classification in Uzbek healthcare feedback. Matched test predictions are needed to estimate uncertainty. Independent institutional data are needed to assess usefulness in routine review of service feedback.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-17
DOI
https://doi.org/10.21203/rs.3.rs-11748609/v1
Primary Topic
Sentiment Analysis and Opinion Mining
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Benchmarking aspect-conditioned sentiment classification of Uzbek healthcare feedback with auxiliary medical entity supervision

Elov Botir Boltayevich, Ruhillo Alaev
3 links
Research Square
Sentiment Analysis and Opinion Mining
preprint

Benchmarking aspect-conditioned sentiment classification of Uzbek healthcare feedback with auxiliary medical entity supervision

Elov Botir Boltayevich, Ruhillo Alaev
preprint en

Abstract

Abstract Background A healthcare comment can express different sentiments about waiting time, staff interactions, and other service aspects. We benchmarked sentiment classification for a supplied aspect in Uzbek online healthcare feedback and examined how scores differed with auxiliary medical entity supervision. Methods The main comparison covered 13 configurations under a reported group-stratified five-fold protocol, with text–aspect pairs from the same normalized comment kept in one fold. BERTbek and XLM-R were compared with lexical, subword, and transformer baselines. Two additional review-only controls supported ablations of aspect conditioning, aspect fusion, and sequential and joint medical named entity recognition training. Macro-averaged F1 on six service aspects was the primary endpoint. The corresponding score on the full medical-domain benchmark was secondary. Results BERTbek with aspect fusion and joint medical entity supervision had the highest reported macro-averaged F1 on the service subset (84.91) and full benchmark (82.74), on a 0–100 scale. The differences from standard paired-input BERTbek were 0.84 and 1.83 points, respectively. The direct differences from BERTbek with aspect fusion alone were 0.50 and 0.78 points, respectively. The largest descriptive differences from the paired-input baseline among service aspects were for waiting time and price. Conclusions The reported point estimates support further evaluation of medical entity supervision for aspect-conditioned sentiment classification in Uzbek healthcare feedback. Matched test predictions are needed to estimate uncertainty. Independent institutional data are needed to assess usefulness in routine review of service feedback.

Research Square
Alisher Navo'i Tashkent State University of Uzbek Language and Literature (UZ), National University of Uzbekistan (UZ)
Sentiment Analysis and Opinion Mining
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon