tiki-review-helpfulness-phobert-berf

DPhO05/tiki-review-helpfulness-phobert-berf

tiki-review-helpfulness-phobert-berf is a text classification model from DPhO05. Use it when you need a label for a piece of text. The card lists the license as mit.

This repository contains training outputs for a Vietnamese review helpfulness detection pipeline.

Downloads · 30 days

9% of all-time downloads

All-time downloads

Public

Repo size

35.9 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.npy25.3 MB · 67%

At a glance

Task: Text Classification
License: mit
Access: Public
Created: May 18, 2026
Updated: Jun 6, 2026
SHA: ac9df828

Model card

From the Hugging Face model README

Tiki Review Helpfulness Detection

This repository contains training outputs for a Vietnamese review helpfulness detection pipeline.

Task

Binary classification:

Label	Meaning
0	not helpful
1	helpful

Pipeline

review title + content
-> PhoBERT embedding
-> Random Forest class probability features
-> metadata features
-> final ML model, LightGBM if available
-> k-fold evaluation

Embedding Model

vinai/phobert-base-v2

Dataset

label	meaning	count	percent
0	not helpful	2050	0.249028
1	helpful	6182	0.750972

Total trainable rows: 8232

Best Model

Best model by mean F1: phobert_logreg

{
  "accuracy_mean": 0.974124904738613,
  "accuracy_std": 0.0032638868870474725,
  "precision_mean": 0.9836382870495802,
  "precision_std": 0.0016924503386736296,
  "recall_mean": 0.9818818626335245,
  "recall_std": 0.004810572605942585,
  "f1_mean": 0.9827526561321027,
  "f1_std": 0.0022188370355954554,
  "roc_auc_mean": 0.9897551528556232,
  "pr_auc_mean": 0.9956131460074612
}

K-Fold Model Comparison

model	accuracy_mean	accuracy_std	precision_mean	precision_std	recall_mean	recall_std	f1_mean	f1_std	roc_auc_mean	pr_auc_mean
phobert_logreg	0.974125	0.003264	0.983638	0.001692	0.981882	0.004811	0.982753	0.002219	0.989755	0.995613
phobert_metadata_logreg	0.973032	0.003061	0.985979	0.001732	0.978000	0.004145	0.981968	0.002082	0.993541	0.997467
berf_rfprob_metadata_final_ml	0.972181	0.004537	0.971367	0.005955	0.992236	0.001467	0.981682	0.002945	0.971489	0.978750
metadata_logreg	0.920797	0.002363	0.989222	0.003441	0.904401	0.003468	0.944904	0.001666	0.984218	0.993569

Fold-Level Metrics

fold	model	accuracy	precision	recall	f1	roc_auc	pr_auc	tn	fp	fn	tp
1	metadata_logreg	0.918640	0.989352	0.901374	0.943316	0.982959	0.993784	398	12	122	1115
1	phobert_logreg	0.978749	0.984677	0.987065	0.985870	0.987385	0.994535	391	19	16	1221
1	phobert_metadata_logreg	0.975713	0.984615	0.983023	0.983819	0.990262	0.995932	391	19	21	1216
1	berf_rfprob_metadata_final_ml	0.970249	0.970681	0.990299	0.980392	0.969412	0.970365	373	37	12	1225
2	metadata_logreg	0.920461	0.991119	0.902183	0.944562	0.987092	0.993396	400	10	121	1116
2	phobert_logreg	0.973285	0.981437	0.983023	0.982229	0.989185	0.995432	387	23	21	1216
2	phobert_metadata_logreg	0.974499	0.988553	0.977365	0.982927	0.994331	0.997557	396	14	28	1209
2	berf_rfprob_metadata_final_ml	0.979964	0.981600	0.991916	0.986731	0.984478	0.988542	387	23	10	1227
3	metadata_logreg	0.922236	0.985114	0.910194	0.946173	0.977728	0.990951	393	17	111	1125
3	phobert_logreg	0.974484	0.982229	0.983819	0.983023	0.989648	0.995213	388	22	20	1216
3	phobert_metadata_logreg	0.973269	0.984553	0.979773	0.982157	0.993504	0.997479	391	19	25	1211
3	berf_rfprob_metadata_final_ml	0.970231	0.969170	0.991909	0.980408	0.963716	0.977172	371	39	10	1226
4	metadata_logreg	0.918591	0.986749	0.903722	0.943412	0.986730	0.995037	395	15	119	1117
4	phobert_logreg	0.969623	0.985270	0.974110	0.979658	0.988957	0.995352	392	18	32	1204
4	phobert_metadata_logreg	0.967801	0.985234	0.971683	0.978411	0.993443	0.997685	392	18	35	1201
4	berf_rfprob_metadata_final_ml	0.968408	0.966142	0.992718	0.979250	0.964708	0.975092	367	43	9	1227
5	metadata_logreg	0.924058	0.993778	0.904531	0.947056	0.986580	0.994679	403	7	118	1118
5	phobert_logreg	0.974484	0.984578	0.981392	0.982982	0.993601	0.997534	391	19	23	1213
5	phobert_metadata_logreg	0.973876	0.986939	0.978155	0.982527	0.996164	0.998681	394	16	27	1209
5	berf_rfprob_metadata_final_ml	0.972053	0.969243	0.994337	0.981629	0.975134	0.982577	371	39	7	1229

OOF BERF-Style Evaluation

accuracy	precision	recall	f1	roc_auc	pr_auc	tn	fp	fn	tp
0.972182	0.971338	0.992236	0.981676	0.970106	0.978583	1869	181	48	6134

OOF Confusion Matrix

actual	pred_0	pred_1
true_0	1869	181
true_1	48	6134

OOF Classification Report

              precision    recall  f1-score   support

           0     0.9750    0.9117    0.9423      2050
           1     0.9713    0.9922    0.9817      6182

    accuracy                         0.9722      8232
   macro avg     0.9731    0.9520    0.9620      8232
weighted avg     0.9722    0.9722    0.9719      8232

Artifacts

phobert_embeddings.npy
kfold_metrics.csv
kfold_metrics_summary.csv
oof_predictions.csv
final_model/
config.json

Notes

The current results are based on manually labeled gold data. Validate on a manually checked holdout set before pseudo-labeling larger unlabeled data.

Task: Text Classification
License: mit
Languages: vi
Created: May 18, 2026
Updated: Jun 6, 2026