You can deploy this model and start earning money today!

yazidsupriadi/
indo_gru_bot

Name: indo_gru_bot
Availability: InStock

Views

No views yet

IndoBERT-GRU for Bot Detection on Platform X

This model uses IndoBERT + Bidirectional GRU to classify Twitter/X accounts as bots or humans based on their textual content.

Model Architecture

Base Model: indobenchmark/indobert-base-p1
RNN Type: Bidirectional GRU
Classifier: 2-layer feedforward with ReLU + Sigmoid

Dataset

The dataset includes Indonesian tweets with labels (0 for human, 1 for bot).

Training Details

Loss Function: Binary Cross Entropy (BCELoss)
Optimizer: Adam (lr = 2e-5)
Epochs: 1
Max Token Length: 128
Batch Size: 16

Training Results IndoGRU Bot Detection

Summary per Epoch

Epoch	Train Loss	Train Acc	Train Prec	Train Rec	Train F1	Val Acc	Val Prec	Val Rec	Val F1	ROC AUC
1	0.5365	0.7572	0.7303	0.8178	0.7716	0.5015	0.5015	1.0000	0.6680	0.8852
2	0.4242	0.8073	0.7990	0.8225	0.8106	0.5015	0.5015	1.0000	0.6680	0.9042
3	0.3937	0.8213	0.8184	0.8271	0.8227	0.5015	0.5015	1.0000	0.6680	0.9140
4	0.3732	0.8318	0.8343	0.8294	0.8318	0.5015	0.5015	1.0000	0.6680	0.9205
5	0.3616	0.8363	0.8402	0.8318	0.8360	0.5015	0.5015	1.0000	0.6680	0.9246
6	0.3500	0.8429	0.8496	0.8345	0.8420	0.5015	0.5015	1.0000	0.6680	0.9273
7	0.3394	0.8483	0.8567	0.8376	0.8471	0.5015	0.5015	1.0000	0.6680	0.9290
8	0.3353	0.8523	0.8632	0.8384	0.8506	0.5015	0.5015	1.0000	0.6680	0.9313
9	0.3275	0.8533	0.8641	0.8396	0.8517	0.5015	0.5015	1.0000	0.6680	0.9324
10	0.3229	0.8553	0.8660	0.8417	0.8537	0.5015	0.5015	1.0000	0.6680	0.9335

Notes

Model disimpan di direktori indo_gru_bot/ setiap epoch dengan nama file seperti model_epoch_1.pth.
Hasil validasi menunjukkan akurasi tetap di 0.5015 dengan recall 1.0000, kemungkinan imbalance kelas atau threshold harus diperbaiki.
ROC AUC meningkat dari 0.9163 ke 0.9343 selama 10 epoch.

✅ Semua model sudah disimpan dan diupload ke repository Hugging Face.

Training Log

Epoch	Train Loss	Train Acc	Train Prec	Train Rec	Train F1	Val Acc	Val Prec	Val Rec	Val F1	ROC AUC
10	0.3076	0.8651	0.8773	0.8500	0.8634	0.5015	0.5015	1.0000	0.6680	0.9343

Evaluation Results

Final Validation Accuracy: 0.5015 Final F1-Score: 0.6680 Final ROC AUC: 0.9343

Confusion Matrix

Classification Report

              precision    recall  f1-score   support

       Human       0.85      0.84      0.85      1994
         Bot       0.84      0.85      0.85      2006

    accuracy                           0.85      4000
   macro avg       0.85      0.85      0.85      4000
weighted avg       0.85      0.85      0.85      4000

yazidsupriadi/indo_gru_bot