balinese-ner 0.1.0__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- balinese_ner-0.1.0/LICENSE +21 -0
- balinese_ner-0.1.0/PKG-INFO +262 -0
- balinese_ner-0.1.0/README.md +234 -0
- balinese_ner-0.1.0/pyproject.toml +46 -0
- balinese_ner-0.1.0/setup.cfg +4 -0
- balinese_ner-0.1.0/src/balinese_ner/__init__.py +22 -0
- balinese_ner-0.1.0/src/balinese_ner/cli.py +82 -0
- balinese_ner-0.1.0/src/balinese_ner/decoding.py +180 -0
- balinese_ner-0.1.0/src/balinese_ner/loader.py +202 -0
- balinese_ner-0.1.0/src/balinese_ner/model.py +348 -0
- balinese_ner-0.1.0/src/balinese_ner/registry.py +150 -0
- balinese_ner-0.1.0/src/balinese_ner.egg-info/PKG-INFO +262 -0
- balinese_ner-0.1.0/src/balinese_ner.egg-info/SOURCES.txt +23 -0
- balinese_ner-0.1.0/src/balinese_ner.egg-info/dependency_links.txt +1 -0
- balinese_ner-0.1.0/src/balinese_ner.egg-info/entry_points.txt +2 -0
- balinese_ner-0.1.0/src/balinese_ner.egg-info/requires.txt +6 -0
- balinese_ner-0.1.0/src/balinese_ner.egg-info/top_level.txt +1 -0
- balinese_ner-0.1.0/tests/test_balinese_ner_package.py +251 -0
- balinese_ner-0.1.0/tests/test_hf_balibert.py +74 -0
- balinese_ner-0.1.0/tests/test_improvement_1_empirical.py +349 -0
- balinese_ner-0.1.0/tests/test_improvement_2_empirical.py +962 -0
- balinese_ner-0.1.0/tests/test_venv_imports.py +22 -0
- balinese_ner-0.1.0/tests/test_verify_imp2_test_xlsx.py +822 -0
- balinese_ner-0.1.0/tests/test_viterbi_adversarial.py +301 -0
- balinese_ner-0.1.0/tests/test_web_ui.py +165 -0
|
@@ -0,0 +1,21 @@
|
|
|
1
|
+
MIT License
|
|
2
|
+
|
|
3
|
+
Copyright (c) 2026 BaliNER Project Contributors
|
|
4
|
+
|
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
|
7
|
+
in the Software without restriction, including without limitation the rights
|
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
|
10
|
+
furnished to do so, subject to the following conditions:
|
|
11
|
+
|
|
12
|
+
The above copyright notice and this permission notice shall be included in all
|
|
13
|
+
copies or substantial portions of the Software.
|
|
14
|
+
|
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
|
21
|
+
SOFTWARE.
|
|
@@ -0,0 +1,262 @@
|
|
|
1
|
+
Metadata-Version: 2.4
|
|
2
|
+
Name: balinese-ner
|
|
3
|
+
Version: 0.1.0
|
|
4
|
+
Summary: Hybrid Balinese Named Entity Recognition (BaliNER) package with Hugging Face Hub, Inference API, local/VPS caching, mirror fallback, and grammar-constrained Viterbi decoding.
|
|
5
|
+
Author: Dramaniako
|
|
6
|
+
License: MIT
|
|
7
|
+
Project-URL: Homepage, https://github.com/Dramaniako/BaliNER
|
|
8
|
+
Project-URL: Repository, https://github.com/Dramaniako/BaliNER
|
|
9
|
+
Project-URL: ModelHub, https://huggingface.co/Dramaniako/BaliBERT
|
|
10
|
+
Keywords: balinese,ner,nlp,indobert,transformers,satua-bali,viterbi
|
|
11
|
+
Classifier: Development Status :: 4 - Beta
|
|
12
|
+
Classifier: Intended Audience :: Developers
|
|
13
|
+
Classifier: Intended Audience :: Science/Research
|
|
14
|
+
Classifier: License :: OSI Approved :: MIT License
|
|
15
|
+
Classifier: Programming Language :: Python :: 3
|
|
16
|
+
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
|
17
|
+
Classifier: Topic :: Text Processing :: Linguistic
|
|
18
|
+
Requires-Python: >=3.9
|
|
19
|
+
Description-Content-Type: text/markdown
|
|
20
|
+
License-File: LICENSE
|
|
21
|
+
Requires-Dist: torch>=2.2.0
|
|
22
|
+
Requires-Dist: transformers>=4.38.0
|
|
23
|
+
Requires-Dist: huggingface_hub>=0.24.0
|
|
24
|
+
Requires-Dist: safetensors>=0.4.0
|
|
25
|
+
Requires-Dist: numpy>=1.24.0
|
|
26
|
+
Requires-Dist: requests>=2.31.0
|
|
27
|
+
Dynamic: license-file
|
|
28
|
+
|
|
29
|
+
# 🦁 BaliNER: Named Entity Recognition untuk Teks Bahasa Bali Berbasis Transformer
|
|
30
|
+
|
|
31
|
+
Sistem ekstraksi entitas bernama (*Named Entity Recognition*) mutakhir untuk mendeteksi karakter dan tokoh pada sastra cerita rakyat Bali (*Satua Bali*) berbasis fine-tuning model Transformer (IndoBERT / Multilingual). Dilengkapi antarmuka web interaktif profesional minimalis (**BaliNER Studio**) serta REST API siap produksi.
|
|
32
|
+
|
|
33
|
+
---
|
|
34
|
+
|
|
35
|
+
## 📊 Hasil Evaluasi & Akurasi Model (IndoBERT Base)
|
|
36
|
+
|
|
37
|
+
Pengujian performa model dievaluasi secara ketat pada data uji buta (*blind test set*) dari berkas `test.xlsx` yang sepenuhnya terisolasi dari proses pelatihan dan validasi (*mencegah data leakage*). Evaluasi menggunakan standar akademik **chunk-level CoNLL** melalui pustaka `seqeval` dengan strategi *first-subword alignment*.
|
|
38
|
+
|
|
39
|
+
### 1. Laporan Klasifikasi Rinci per Kelas Entitas (Data Uji `test.xlsx`)
|
|
40
|
+
|
|
41
|
+
| Kategori Entitas | Deskripsi Karakter | Precision | Recall | F1-Score | Support (Sampel) |
|
|
42
|
+
| :--- | :--- | :---: | :---: | :---: | :---: |
|
|
43
|
+
| **`ADJ`** | Penanda Sifat Karakter Tokoh | **0.88** | **0.93** | **0.91** | 132 |
|
|
44
|
+
| **`PNAME`** | Tokoh Manusia (*Person Name*) | **0.81** | **0.85** | **0.83** | 845 |
|
|
45
|
+
| **`ANM`** | Tokoh Binatang / Fabel (*Animal*) | **0.79** | **0.82** | **0.81** | 872 |
|
|
46
|
+
| **`GODS`** | Dewa & Bhatara (*Deities & Spiritual*) | **0.63** | **0.70** | **0.66** | 162 |
|
|
47
|
+
| **`OBJ`** | Benda Pusaka / Keramat (*Sacred Object*) | **0.60** | **0.60** | **0.60** | 577 |
|
|
48
|
+
| **Micro Average** | Rata-rata Agregat Global | **0.75** | **0.78** | **0.77** | **2.588** |
|
|
49
|
+
| **Macro Average** | Rata-rata Tak Berbobot per Kelas | **0.74** | **0.78** | **0.76** | **2.588** |
|
|
50
|
+
| **Weighted Average** | Rata-rata Berbobot Proporsi Kelas | **0.75** | **0.78** | **0.77** | **2.588** |
|
|
51
|
+
|
|
52
|
+
> **Catatan Analisis**:
|
|
53
|
+
> - Kategori **ADJ** memperoleh skor F1 tertinggi (**0.91**) karena pola leksikal sifat tokoh yang sangat konsisten (*ririh*, *sugih*, *lacur*).
|
|
54
|
+
> - Kategori **PNAME** (**0.83**) dan **ANM** (**0.81**) memiliki rekognisi yang sangat kuat berkat pemanfaatan partikel budaya (*I*, *Men*, *Pan*, *Sang*, *Ni*) dan normalisasi batas kata (*boundary snapping*).
|
|
55
|
+
> - Kategori **OBJ** (**0.60**) dan **GODS** (**0.66**) memiliki tantangan ambiguitas konteks yang berhasil ditingkatkan melalui penyelarasan fitur POS tag.
|
|
56
|
+
|
|
57
|
+
---
|
|
58
|
+
|
|
59
|
+
### 2. Riwayat Pelatihan & Validasi per Epoch
|
|
60
|
+
|
|
61
|
+
Fine-tuning dilakukan selama 5 epoch pada model backbone serumpun Austronesia `indobenchmark/indobert-base-p1` dengan learning rate `3e-5`, batch size `16`, dan optimizer AdamW:
|
|
62
|
+
|
|
63
|
+
| Epoch | Training Loss | Validation Loss | Precision | Recall | F1-Score |
|
|
64
|
+
| :---: | :---: | :---: | :---: | :---: | :---: |
|
|
65
|
+
| **1** | 0.2262 | 0.2221 | 0.7169 | 0.7096 | 0.7133 |
|
|
66
|
+
| **2** | 0.1437 | 0.1908 | 0.7305 | 0.7873 | 0.7579 |
|
|
67
|
+
| **3** | 0.0731 | 0.1836 | 0.7174 | 0.7996 | 0.7563 |
|
|
68
|
+
| **4** | 0.0380 | 0.2147 | 0.7281 | 0.8160 | 0.7695 |
|
|
69
|
+
| **5** | **0.0191** | 0.2262 | **0.7383** | **0.8078** | **0.7715** |
|
|
70
|
+
|
|
71
|
+
Model terbaik pada checkpoint optimal secara otomatis diekspor ke direktori `./best_balinese_ner_model_base`.
|
|
72
|
+
|
|
73
|
+
---
|
|
74
|
+
|
|
75
|
+
## 📑 Daftar Diagram Arsitektur & Rekayasa Perangkat Lunak UML
|
|
76
|
+
|
|
77
|
+
Dokumentasi lengkap analisis sistem, kebutuhan fungsional (11 use cases), kebutuhan non-fungsional, dan visualisasi arsitektur UML tersimpan dalam berkas-berkas markdown terpisah:
|
|
78
|
+
|
|
79
|
+
- 📐 **[RINGKASAN_DIAGRAM.md](file:///d:/Project/BaliNER/docs/RINGKASAN_DIAGRAM.md)**: **Versi Sederhana / Ringkasan Seluruh Diagram** (Use Case, Activity, Class, Flowchart, dan Sequence) untuk pemahaman cepat.
|
|
80
|
+
1. 🎯 **[USE_CASE_DIAGRAM.md](file:///d:/Project/BaliNER/docs/USE_CASE_DIAGRAM.md)**: Diagram kasus penggunaan UML untuk 11 fungsi sistem serta matriks kepatuhan kebutuhan operasional, keamanan, informasi, dan kinerja.
|
|
81
|
+
2. 🔄 **[ACTIVITY_DIAGRAM.md](file:///d:/Project/BaliNER/docs/ACTIVITY_DIAGRAM.md)**: Diagram aktivitas UML dengan jalur tanggung jawab (*swimlanes*) antarmodul dari input teks hingga evaluasi.
|
|
82
|
+
3. 🏛️ **[CLASS_DIAGRAM.md](file:///d:/Project/BaliNER/docs/CLASS_DIAGRAM.md)**: Diagram kelas berorientasi objek memodelkan modularitas pipeline inferensi, model manager, dan peladen web API.
|
|
83
|
+
4. 📊 **[FLOWCHART.md](file:///d:/Project/BaliNER/docs/FLOWCHART.md)**: Diagram alur menyeluruh (*end-to-end*) dari verifikasi bahasa, ekstraksi entitas, hingga pencatatan riwayat.
|
|
84
|
+
5. ⏱️ **[SEQUENCE_DIAGRAM.md](file:///d:/Project/BaliNER/docs/SEQUENCE_DIAGRAM.md)**: Diagram urutan waktu interaksi kronologis antarkomponen saat inferensi dan pencatatan umpan balik.
|
|
85
|
+
6. 🎨 **[UI_UX_DOCUMENTATION.md](file:///d:/Project/BaliNER/docs/UI_UX_DOCUMENTATION.md)**: Dokumentasi komprehensif desain antarmuka pengguna (UI/UX), dilengkapi 19 tangkapan layar asli, pemetaan 11 kebutuhan fungsional, dan evaluasi heuristik.
|
|
86
|
+
|
|
87
|
+
---
|
|
88
|
+
|
|
89
|
+
## 🏷️ Taksonomi Label Entitas Karakter (11 Kelas BIO)
|
|
90
|
+
|
|
91
|
+
Sistem mengidentifikasi 5 kategori utama karakter budaya Bali dengan skema anotasi *Beginning-Inside-Outside* (BIO):
|
|
92
|
+
|
|
93
|
+
| Tag BIO | Kategori Karakter | Penjelasan & Contoh |
|
|
94
|
+
| :--- | :--- | :--- |
|
|
95
|
+
| `B-PNAME` / `I-PNAME` | **Person Name** | Tokoh manusia (contoh: *I Pekak Dukuh*, *Men Sugih*, *I Tiwas*). |
|
|
96
|
+
| `B-ANM` / `I-ANM` | **Animal** | Tokoh binatang / fabel (contoh: *Sang Lutung*, *I Kekua*, *I Macan*). |
|
|
97
|
+
| `B-OBJ` / `I-OBJ` | **Object** | Benda magis / pusaka bertuah (contoh: *Gada*, *Pecut Sakti*). |
|
|
98
|
+
| `B-GODS` / `I-GODS` | **Gods & Deities** | Dewa, bhatara, atau entitas spiritual (contoh: *Bhatara Guru*). |
|
|
99
|
+
| `B-ADJ` / `I-ADJ` | **Adjective Modifier** | Penanda sifat karakter tokoh (contoh: *anak ririh*, *tualen*). |
|
|
100
|
+
| `O` | **Outside** | Kata umum non-entitas. |
|
|
101
|
+
|
|
102
|
+
---
|
|
103
|
+
|
|
104
|
+
## 📂 Struktur Direktori Proyek
|
|
105
|
+
|
|
106
|
+
Proyek ini telah ditata secara modular dan rapi sesuai standar rekayasa perangkat lunak dan pembelajaran mesin (*Machine Learning*):
|
|
107
|
+
|
|
108
|
+
```text
|
|
109
|
+
BaliNER/
|
|
110
|
+
├── data/ # Dataset anotasi aksara/teks Latin Satua Bali
|
|
111
|
+
│ ├── train.xlsx # Dataset latih (60.627 token, 4.398 kalimat)
|
|
112
|
+
│ ├── test.xlsx # Dataset uji blind test (28.986 token, 2.052 kalimat)
|
|
113
|
+
│ └── all.xlsx # Dataset gabungan lengkap
|
|
114
|
+
├── docs/ # Dokumentasi arsitektur sistem & rekayasa perangkat lunak
|
|
115
|
+
│ ├── images/ # Tangkapan layar antarmuka & aset visual sistem
|
|
116
|
+
│ ├── ACTIVITY_DIAGRAM.md # Activity diagram & alur swimlanes
|
|
117
|
+
│ ├── CLASS_DIAGRAM.md # Class diagram & pemodelan komponen OOP
|
|
118
|
+
│ ├── FLOWCHART.md # Flowchart alur inferensi & verifikasi bahasa
|
|
119
|
+
│ ├── GITHUB_SETUP.md # Panduan unggah & konfigurasi Git LFS
|
|
120
|
+
│ ├── RINGKASAN_DIAGRAM.md # Rangkuman komprehensif diagram sistem
|
|
121
|
+
│ ├── SEQUENCE_DIAGRAM.md # Sequence diagram interaksi inferensi & umpan balik
|
|
122
|
+
│ ├── UI_UX_DOCUMENTATION.md # Dokumentasi UI/UX & evaluasi heuristik
|
|
123
|
+
│ └── USE_CASE_DIAGRAM.md # Use case diagram 11 kebutuhan fungsional
|
|
124
|
+
├── tests/ # Pengujian unit & integrasi otomatis
|
|
125
|
+
│ ├── __init__.py # Inisialisasi paket tes
|
|
126
|
+
│ ├── test_venv_imports.py # Pengujian kompatibilitas environment & dependensi
|
|
127
|
+
│ └── test_web_ui.py # Suite 9 unit test Flask & pipeline inferensi
|
|
128
|
+
├── notebooks/ # Koleksi Jupyter Notebook varian model Transformer
|
|
129
|
+
│ ├── README.md # Panduan lengkap tiap model eksperimen
|
|
130
|
+
│ ├── BaliNER_IndoBERT_base.ipynb # Model IndoBERT Base (IndoBenchmark)
|
|
131
|
+
│ ├── BaliNER_IndoBERT_large.ipynb # Model IndoBERT Large (IndoBenchmark)
|
|
132
|
+
│ ├── BaliNER_NusaBERT.ipynb # Model NusaBERT Base (LazarusNLP)
|
|
133
|
+
│ ├── BaliNER_IndoRoBERTa.ipynb # Model IndoRoBERTa Base (Cahya)
|
|
134
|
+
│ ├── BaliNER_RoBERTa.ipynb # Model RoBERTa Base standar
|
|
135
|
+
│ ├── BaliNER_XLM_RoBERTa.ipynb # Model XLM-RoBERTa Base multilingual
|
|
136
|
+
│ ├── BaliNER_mDeBERTa_v3.ipynb # Model mDeBERTa-v3 Base (Disentangled Attention)
|
|
137
|
+
│ └── BaliNER_mBERT.ipynb # Model mBERT Base (Google Multilingual)
|
|
138
|
+
├── reports/ # Hasil komparasi performa model (CSV, JSON, Markdown, PNG)
|
|
139
|
+
│ ├── MODEL_COMPARISON.md # Laporan analisis komparatif publikasi
|
|
140
|
+
│ ├── model_comparison_summary.csv # Ringkasan matriks leaderboard
|
|
141
|
+
│ ├── model_comparison_per_class.csv # Rincian metrik per kategori entitas
|
|
142
|
+
│ ├── model_comparison_overall_f1.png # Grafik bar chart F1-score keseluruhan
|
|
143
|
+
│ ├── model_comparison_per_class.png # Grafik grouped bar chart 5 kategori entitas
|
|
144
|
+
│ └── model_comparison_latency_vs_f1.png# Scatter plot trade-off akurasi vs latensi
|
|
145
|
+
├── submission/ # Arsip berkas pengumpulan tugas akademik (diabaikan Git)
|
|
146
|
+
│ ├── 2408561071_Kelompok_BaliNER.pdf # Laporan dokumentasi tugas
|
|
147
|
+
│ ├── 2408561071_Kelompok_BaliNER.zip # Arsip penyerahan mandiri
|
|
148
|
+
│ └── 2408561071_Kelompok_BaliNER/ # Folder build penyerahan
|
|
149
|
+
├── best_balinese_ner_model_base/ # Model IndoBERT Base terlatih (konfigurasi & tokenizer)
|
|
150
|
+
├── static/ # Aset antarmuka web (CSS & JS)
|
|
151
|
+
│ ├── css/style.css # Gaya CSS profesional minimalis
|
|
152
|
+
│ └── js/app.js # Logika interaktif antarmuka klien
|
|
153
|
+
├── templates/ # Template Jinja2 Flask
|
|
154
|
+
│ └── index.html # Template HTML BaliNER Studio
|
|
155
|
+
├── app.py # Peladen web Flask & backend REST API
|
|
156
|
+
├── infer_ner.py # Utilitas CLI inferensi teks kustom
|
|
157
|
+
├── train_ner.py # Skrip CLI pelatihan & evaluasi Transformer
|
|
158
|
+
├── compare_models.py # Program evaluasi & komparasi benchmark seluruh model
|
|
159
|
+
├── BaliNER.ipynb # Jupyter Notebook eksperimen & analisis data
|
|
160
|
+
├── run_web_ui.bat # Launcher 1-klik untuk sistem operasi Windows
|
|
161
|
+
├── run_web_ui.sh # Launcher 1-klik untuk Linux / macOS
|
|
162
|
+
├── requirements.txt # Daftar dependensi Python
|
|
163
|
+
├── CONTRIBUTING.md # Panduan kontribusi pengembangan
|
|
164
|
+
├── LICENSE # Lisensi proyek (MIT)
|
|
165
|
+
└── README.md # Dokumentasi utama proyek
|
|
166
|
+
```
|
|
167
|
+
|
|
168
|
+
---
|
|
169
|
+
|
|
170
|
+
## 💻 Panduan Penggunaan Sistem
|
|
171
|
+
|
|
172
|
+
### Opsi 0: Python Package Hybrid (`balinese-ner`) — *Untuk Integrasi Developer & Redundansi Server*
|
|
173
|
+
Paket Python yang dapat di-`pip install` dengan dukungan **cache lokal/VPS**, **Hugging Face Hub (`Dramaniako/BaliBERT`)**, **Hugging Face Inference API**, **backup mirror fallback (GitHub Releases)**, serta **Constrained Viterbi BIO Decoding** (0 pelanggaran BIO):
|
|
174
|
+
|
|
175
|
+
```bash
|
|
176
|
+
pip install git+https://github.com/Dramaniako/BaliNER.git
|
|
177
|
+
```
|
|
178
|
+
|
|
179
|
+
```python
|
|
180
|
+
from balinese_ner import PretrainedModelNER
|
|
181
|
+
|
|
182
|
+
# Mode Hybrid (Default): cek cache lokal/VPS -> unduh HF Hub -> fallback mirror jika HF down
|
|
183
|
+
ner = PretrainedModelNER.load("BaliBERT", mode="hybrid")
|
|
184
|
+
print(ner.predict("I Pekak Dukuh lantas ngaukin Sang Lutung ane sedeng ngamah biu."))
|
|
185
|
+
|
|
186
|
+
# Mode API dengan Fallback Lokal (hemat RAM VPS, otomatis fallback ke lokal jika API HF gangguan)
|
|
187
|
+
ner_api = PretrainedModelNER.load("BaliBERT", mode="api_fallback")
|
|
188
|
+
|
|
189
|
+
# Mode Offline Murni (menggunakan cache/bobot lokal tanpa internet)
|
|
190
|
+
ner_offline = PretrainedModelNER.load("BaliBERT", mode="local")
|
|
191
|
+
```
|
|
192
|
+
|
|
193
|
+
Untuk mengisi cache pada server VPS sebelum demo (*pre-warming*):
|
|
194
|
+
```bash
|
|
195
|
+
balinese-ner --prewarm --model BaliBERT
|
|
196
|
+
```
|
|
197
|
+
|
|
198
|
+
---
|
|
199
|
+
|
|
200
|
+
### Opsi 1: Antarmuka Web Interaktif (BaliNER Studio Web UI) — *Direkomendasikan*
|
|
201
|
+
Aplikasi web modern bergaya *profesional minimalis* yang terhubung langsung ke model `best_balinese_ner_model_base`:
|
|
202
|
+
- **Cara 1 (Windows 1-Click Launcher)**: Cukup klik ganda berkas **[run_web_ui.bat](file:///d:/Project/BaliNER/run_web_ui.bat)**.
|
|
203
|
+
- **Cara 2 (Terminal CLI)**:
|
|
204
|
+
```powershell
|
|
205
|
+
.\.venv\Scripts\python.exe app.py
|
|
206
|
+
```
|
|
207
|
+
Kemudian buka peramban web di:
|
|
208
|
+
👉 **`http://127.0.0.1:5000`**
|
|
209
|
+
|
|
210
|
+
#### ✨ Rangkuman Fitur Web UI:
|
|
211
|
+
- **Deteksi Bahasa Otomatis**: Mendeteksi kecocokan teks bahasa Bali secara instan dengan indikator persentase.
|
|
212
|
+
- **Visualisasi Teks Beranotasi**: Penandaan entitas interaktif dengan badge warna BIO dan *tooltip popover* detail (*Skor, Kategori, Rentang Span*).
|
|
213
|
+
- **Preset Contoh Satua Bali**: 5 kutipan sastra rakyat klasik siap uji dalam 1-klik.
|
|
214
|
+
- **Tabel Rincian Entitas & Bilah Keyakinan**: Tabel data lengkap dengan bilah progres keyakinan dan tombol salin per entitas.
|
|
215
|
+
- **Evaluasi & Penilaian Kualitas**: Fitur umpan balik (👍 Akurat / 👎 Perlu Koreksi) untuk mencatat hasil penilaian linguis.
|
|
216
|
+
- **Riwayat Analisis & Pemulihan Instan**: Histori deteksi tersimpan lokal di browser dengan kemampuan *restore* ke layar utama.
|
|
217
|
+
- **Mode Gelap & Terang**: Tema *professional minimalist* elegan dengan persistensi preferensi.
|
|
218
|
+
- **REST API Terbuka**: Endpoint `/api/predict`, `/api/info`, `/api/examples`, `/api/detect-language`, dan `/api/feedback`.
|
|
219
|
+
|
|
220
|
+
---
|
|
221
|
+
|
|
222
|
+
### Opsi 2: Menjalankan via Jupyter Notebook (Interaktif)
|
|
223
|
+
Anda dapat menjalankan notebook baseline utama:
|
|
224
|
+
👉 **[BaliNER.ipynb](file:///d:/Project/BaliNER/BaliNER.ipynb)** *(IndoBERT Base)*
|
|
225
|
+
|
|
226
|
+
Atau memilih notebook spesifik per model pada direktori **[`notebooks/`](file:///d:/Project/BaliNER/notebooks)** untuk eksperimen komparatif:
|
|
227
|
+
- 🇮🇩 **[BaliNER_IndoBERT_base.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_IndoBERT_base.ipynb)** (`indobenchmark/indobert-base-p1`)
|
|
228
|
+
- 🐘 **[BaliNER_IndoBERT_large.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_IndoBERT_large.ipynb)** (`indobenchmark/indobert-large-p1`)
|
|
229
|
+
- 🏝️ **[BaliNER_NusaBERT.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_NusaBERT.ipynb)** (`LazarusNLP/NusaBERT-base`)
|
|
230
|
+
- 🇮🇩 **[BaliNER_IndoRoBERTa.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_IndoRoBERTa.ipynb)** (`cahya/roberta-base-indonesian-1.5G`)
|
|
231
|
+
- 🤖 **[BaliNER_RoBERTa.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_RoBERTa.ipynb)** (`roberta-base`)
|
|
232
|
+
- 🌐 **[BaliNER_XLM_RoBERTa.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_XLM_RoBERTa.ipynb)** (`xlm-roberta-base`)
|
|
233
|
+
- ⚡ **[BaliNER_mDeBERTa_v3.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_mDeBERTa_v3.ipynb)** (`microsoft/mdeberta-v3-base`)
|
|
234
|
+
- 🌍 **[BaliNER_mBERT.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_mBERT.ipynb)** (`bert-base-multilingual-cased`)
|
|
235
|
+
|
|
236
|
+
### Opsi 3: Pelatihan Cepat via Terminal (CLI)
|
|
237
|
+
```bash
|
|
238
|
+
python train_ner.py --epochs 5 --batch_size 16 --learning_rate 3e-5
|
|
239
|
+
```
|
|
240
|
+
|
|
241
|
+
### Opsi 4: Menjalankan Inferensi pada Kalimat Baru via CLI
|
|
242
|
+
```bash
|
|
243
|
+
python infer_ner.py --text "I Pekak Dukuh lantas ngaukin Sang Lutung ane sedeng ngamah biu di duur punyane."
|
|
244
|
+
```
|
|
245
|
+
|
|
246
|
+
### Opsi 5: Menjalankan Pengujian Otomatis (Unit & Integration Tests)
|
|
247
|
+
```bash
|
|
248
|
+
python tests/test_web_ui.py
|
|
249
|
+
# Atau menggunakan discovery:
|
|
250
|
+
python -m unittest discover tests
|
|
251
|
+
```
|
|
252
|
+
|
|
253
|
+
### Opsi 6: Melakukan Komparasi Performa Seluruh Model Terlatih (Benchmark CLI)
|
|
254
|
+
Untuk mengevaluasi seluruh model yang telah dilatih dan menghasilkan laporan perbandingan lengkap (tabel leaderboard terminal, berkas CSV, JSON, Markdown, serta visualisasi grafik PNG):
|
|
255
|
+
```bash
|
|
256
|
+
python compare_models.py
|
|
257
|
+
```
|
|
258
|
+
*Argumen opsional:*
|
|
259
|
+
- `--test_file`: Lokasi berkas data uji (default: `data/test.xlsx`)
|
|
260
|
+
- `--batch_size`: Ukuran batch evaluasi (default: `32`)
|
|
261
|
+
- `--output_dir`: Direktori penyimpanan laporan (default: `reports/`)
|
|
262
|
+
- `--no_plots`: Lewati pembuatan grafik gambar PNG
|
|
@@ -0,0 +1,234 @@
|
|
|
1
|
+
# 🦁 BaliNER: Named Entity Recognition untuk Teks Bahasa Bali Berbasis Transformer
|
|
2
|
+
|
|
3
|
+
Sistem ekstraksi entitas bernama (*Named Entity Recognition*) mutakhir untuk mendeteksi karakter dan tokoh pada sastra cerita rakyat Bali (*Satua Bali*) berbasis fine-tuning model Transformer (IndoBERT / Multilingual). Dilengkapi antarmuka web interaktif profesional minimalis (**BaliNER Studio**) serta REST API siap produksi.
|
|
4
|
+
|
|
5
|
+
---
|
|
6
|
+
|
|
7
|
+
## 📊 Hasil Evaluasi & Akurasi Model (IndoBERT Base)
|
|
8
|
+
|
|
9
|
+
Pengujian performa model dievaluasi secara ketat pada data uji buta (*blind test set*) dari berkas `test.xlsx` yang sepenuhnya terisolasi dari proses pelatihan dan validasi (*mencegah data leakage*). Evaluasi menggunakan standar akademik **chunk-level CoNLL** melalui pustaka `seqeval` dengan strategi *first-subword alignment*.
|
|
10
|
+
|
|
11
|
+
### 1. Laporan Klasifikasi Rinci per Kelas Entitas (Data Uji `test.xlsx`)
|
|
12
|
+
|
|
13
|
+
| Kategori Entitas | Deskripsi Karakter | Precision | Recall | F1-Score | Support (Sampel) |
|
|
14
|
+
| :--- | :--- | :---: | :---: | :---: | :---: |
|
|
15
|
+
| **`ADJ`** | Penanda Sifat Karakter Tokoh | **0.88** | **0.93** | **0.91** | 132 |
|
|
16
|
+
| **`PNAME`** | Tokoh Manusia (*Person Name*) | **0.81** | **0.85** | **0.83** | 845 |
|
|
17
|
+
| **`ANM`** | Tokoh Binatang / Fabel (*Animal*) | **0.79** | **0.82** | **0.81** | 872 |
|
|
18
|
+
| **`GODS`** | Dewa & Bhatara (*Deities & Spiritual*) | **0.63** | **0.70** | **0.66** | 162 |
|
|
19
|
+
| **`OBJ`** | Benda Pusaka / Keramat (*Sacred Object*) | **0.60** | **0.60** | **0.60** | 577 |
|
|
20
|
+
| **Micro Average** | Rata-rata Agregat Global | **0.75** | **0.78** | **0.77** | **2.588** |
|
|
21
|
+
| **Macro Average** | Rata-rata Tak Berbobot per Kelas | **0.74** | **0.78** | **0.76** | **2.588** |
|
|
22
|
+
| **Weighted Average** | Rata-rata Berbobot Proporsi Kelas | **0.75** | **0.78** | **0.77** | **2.588** |
|
|
23
|
+
|
|
24
|
+
> **Catatan Analisis**:
|
|
25
|
+
> - Kategori **ADJ** memperoleh skor F1 tertinggi (**0.91**) karena pola leksikal sifat tokoh yang sangat konsisten (*ririh*, *sugih*, *lacur*).
|
|
26
|
+
> - Kategori **PNAME** (**0.83**) dan **ANM** (**0.81**) memiliki rekognisi yang sangat kuat berkat pemanfaatan partikel budaya (*I*, *Men*, *Pan*, *Sang*, *Ni*) dan normalisasi batas kata (*boundary snapping*).
|
|
27
|
+
> - Kategori **OBJ** (**0.60**) dan **GODS** (**0.66**) memiliki tantangan ambiguitas konteks yang berhasil ditingkatkan melalui penyelarasan fitur POS tag.
|
|
28
|
+
|
|
29
|
+
---
|
|
30
|
+
|
|
31
|
+
### 2. Riwayat Pelatihan & Validasi per Epoch
|
|
32
|
+
|
|
33
|
+
Fine-tuning dilakukan selama 5 epoch pada model backbone serumpun Austronesia `indobenchmark/indobert-base-p1` dengan learning rate `3e-5`, batch size `16`, dan optimizer AdamW:
|
|
34
|
+
|
|
35
|
+
| Epoch | Training Loss | Validation Loss | Precision | Recall | F1-Score |
|
|
36
|
+
| :---: | :---: | :---: | :---: | :---: | :---: |
|
|
37
|
+
| **1** | 0.2262 | 0.2221 | 0.7169 | 0.7096 | 0.7133 |
|
|
38
|
+
| **2** | 0.1437 | 0.1908 | 0.7305 | 0.7873 | 0.7579 |
|
|
39
|
+
| **3** | 0.0731 | 0.1836 | 0.7174 | 0.7996 | 0.7563 |
|
|
40
|
+
| **4** | 0.0380 | 0.2147 | 0.7281 | 0.8160 | 0.7695 |
|
|
41
|
+
| **5** | **0.0191** | 0.2262 | **0.7383** | **0.8078** | **0.7715** |
|
|
42
|
+
|
|
43
|
+
Model terbaik pada checkpoint optimal secara otomatis diekspor ke direktori `./best_balinese_ner_model_base`.
|
|
44
|
+
|
|
45
|
+
---
|
|
46
|
+
|
|
47
|
+
## 📑 Daftar Diagram Arsitektur & Rekayasa Perangkat Lunak UML
|
|
48
|
+
|
|
49
|
+
Dokumentasi lengkap analisis sistem, kebutuhan fungsional (11 use cases), kebutuhan non-fungsional, dan visualisasi arsitektur UML tersimpan dalam berkas-berkas markdown terpisah:
|
|
50
|
+
|
|
51
|
+
- 📐 **[RINGKASAN_DIAGRAM.md](file:///d:/Project/BaliNER/docs/RINGKASAN_DIAGRAM.md)**: **Versi Sederhana / Ringkasan Seluruh Diagram** (Use Case, Activity, Class, Flowchart, dan Sequence) untuk pemahaman cepat.
|
|
52
|
+
1. 🎯 **[USE_CASE_DIAGRAM.md](file:///d:/Project/BaliNER/docs/USE_CASE_DIAGRAM.md)**: Diagram kasus penggunaan UML untuk 11 fungsi sistem serta matriks kepatuhan kebutuhan operasional, keamanan, informasi, dan kinerja.
|
|
53
|
+
2. 🔄 **[ACTIVITY_DIAGRAM.md](file:///d:/Project/BaliNER/docs/ACTIVITY_DIAGRAM.md)**: Diagram aktivitas UML dengan jalur tanggung jawab (*swimlanes*) antarmodul dari input teks hingga evaluasi.
|
|
54
|
+
3. 🏛️ **[CLASS_DIAGRAM.md](file:///d:/Project/BaliNER/docs/CLASS_DIAGRAM.md)**: Diagram kelas berorientasi objek memodelkan modularitas pipeline inferensi, model manager, dan peladen web API.
|
|
55
|
+
4. 📊 **[FLOWCHART.md](file:///d:/Project/BaliNER/docs/FLOWCHART.md)**: Diagram alur menyeluruh (*end-to-end*) dari verifikasi bahasa, ekstraksi entitas, hingga pencatatan riwayat.
|
|
56
|
+
5. ⏱️ **[SEQUENCE_DIAGRAM.md](file:///d:/Project/BaliNER/docs/SEQUENCE_DIAGRAM.md)**: Diagram urutan waktu interaksi kronologis antarkomponen saat inferensi dan pencatatan umpan balik.
|
|
57
|
+
6. 🎨 **[UI_UX_DOCUMENTATION.md](file:///d:/Project/BaliNER/docs/UI_UX_DOCUMENTATION.md)**: Dokumentasi komprehensif desain antarmuka pengguna (UI/UX), dilengkapi 19 tangkapan layar asli, pemetaan 11 kebutuhan fungsional, dan evaluasi heuristik.
|
|
58
|
+
|
|
59
|
+
---
|
|
60
|
+
|
|
61
|
+
## 🏷️ Taksonomi Label Entitas Karakter (11 Kelas BIO)
|
|
62
|
+
|
|
63
|
+
Sistem mengidentifikasi 5 kategori utama karakter budaya Bali dengan skema anotasi *Beginning-Inside-Outside* (BIO):
|
|
64
|
+
|
|
65
|
+
| Tag BIO | Kategori Karakter | Penjelasan & Contoh |
|
|
66
|
+
| :--- | :--- | :--- |
|
|
67
|
+
| `B-PNAME` / `I-PNAME` | **Person Name** | Tokoh manusia (contoh: *I Pekak Dukuh*, *Men Sugih*, *I Tiwas*). |
|
|
68
|
+
| `B-ANM` / `I-ANM` | **Animal** | Tokoh binatang / fabel (contoh: *Sang Lutung*, *I Kekua*, *I Macan*). |
|
|
69
|
+
| `B-OBJ` / `I-OBJ` | **Object** | Benda magis / pusaka bertuah (contoh: *Gada*, *Pecut Sakti*). |
|
|
70
|
+
| `B-GODS` / `I-GODS` | **Gods & Deities** | Dewa, bhatara, atau entitas spiritual (contoh: *Bhatara Guru*). |
|
|
71
|
+
| `B-ADJ` / `I-ADJ` | **Adjective Modifier** | Penanda sifat karakter tokoh (contoh: *anak ririh*, *tualen*). |
|
|
72
|
+
| `O` | **Outside** | Kata umum non-entitas. |
|
|
73
|
+
|
|
74
|
+
---
|
|
75
|
+
|
|
76
|
+
## 📂 Struktur Direktori Proyek
|
|
77
|
+
|
|
78
|
+
Proyek ini telah ditata secara modular dan rapi sesuai standar rekayasa perangkat lunak dan pembelajaran mesin (*Machine Learning*):
|
|
79
|
+
|
|
80
|
+
```text
|
|
81
|
+
BaliNER/
|
|
82
|
+
├── data/ # Dataset anotasi aksara/teks Latin Satua Bali
|
|
83
|
+
│ ├── train.xlsx # Dataset latih (60.627 token, 4.398 kalimat)
|
|
84
|
+
│ ├── test.xlsx # Dataset uji blind test (28.986 token, 2.052 kalimat)
|
|
85
|
+
│ └── all.xlsx # Dataset gabungan lengkap
|
|
86
|
+
├── docs/ # Dokumentasi arsitektur sistem & rekayasa perangkat lunak
|
|
87
|
+
│ ├── images/ # Tangkapan layar antarmuka & aset visual sistem
|
|
88
|
+
│ ├── ACTIVITY_DIAGRAM.md # Activity diagram & alur swimlanes
|
|
89
|
+
│ ├── CLASS_DIAGRAM.md # Class diagram & pemodelan komponen OOP
|
|
90
|
+
│ ├── FLOWCHART.md # Flowchart alur inferensi & verifikasi bahasa
|
|
91
|
+
│ ├── GITHUB_SETUP.md # Panduan unggah & konfigurasi Git LFS
|
|
92
|
+
│ ├── RINGKASAN_DIAGRAM.md # Rangkuman komprehensif diagram sistem
|
|
93
|
+
│ ├── SEQUENCE_DIAGRAM.md # Sequence diagram interaksi inferensi & umpan balik
|
|
94
|
+
│ ├── UI_UX_DOCUMENTATION.md # Dokumentasi UI/UX & evaluasi heuristik
|
|
95
|
+
│ └── USE_CASE_DIAGRAM.md # Use case diagram 11 kebutuhan fungsional
|
|
96
|
+
├── tests/ # Pengujian unit & integrasi otomatis
|
|
97
|
+
│ ├── __init__.py # Inisialisasi paket tes
|
|
98
|
+
│ ├── test_venv_imports.py # Pengujian kompatibilitas environment & dependensi
|
|
99
|
+
│ └── test_web_ui.py # Suite 9 unit test Flask & pipeline inferensi
|
|
100
|
+
├── notebooks/ # Koleksi Jupyter Notebook varian model Transformer
|
|
101
|
+
│ ├── README.md # Panduan lengkap tiap model eksperimen
|
|
102
|
+
│ ├── BaliNER_IndoBERT_base.ipynb # Model IndoBERT Base (IndoBenchmark)
|
|
103
|
+
│ ├── BaliNER_IndoBERT_large.ipynb # Model IndoBERT Large (IndoBenchmark)
|
|
104
|
+
│ ├── BaliNER_NusaBERT.ipynb # Model NusaBERT Base (LazarusNLP)
|
|
105
|
+
│ ├── BaliNER_IndoRoBERTa.ipynb # Model IndoRoBERTa Base (Cahya)
|
|
106
|
+
│ ├── BaliNER_RoBERTa.ipynb # Model RoBERTa Base standar
|
|
107
|
+
│ ├── BaliNER_XLM_RoBERTa.ipynb # Model XLM-RoBERTa Base multilingual
|
|
108
|
+
│ ├── BaliNER_mDeBERTa_v3.ipynb # Model mDeBERTa-v3 Base (Disentangled Attention)
|
|
109
|
+
│ └── BaliNER_mBERT.ipynb # Model mBERT Base (Google Multilingual)
|
|
110
|
+
├── reports/ # Hasil komparasi performa model (CSV, JSON, Markdown, PNG)
|
|
111
|
+
│ ├── MODEL_COMPARISON.md # Laporan analisis komparatif publikasi
|
|
112
|
+
│ ├── model_comparison_summary.csv # Ringkasan matriks leaderboard
|
|
113
|
+
│ ├── model_comparison_per_class.csv # Rincian metrik per kategori entitas
|
|
114
|
+
│ ├── model_comparison_overall_f1.png # Grafik bar chart F1-score keseluruhan
|
|
115
|
+
│ ├── model_comparison_per_class.png # Grafik grouped bar chart 5 kategori entitas
|
|
116
|
+
│ └── model_comparison_latency_vs_f1.png# Scatter plot trade-off akurasi vs latensi
|
|
117
|
+
├── submission/ # Arsip berkas pengumpulan tugas akademik (diabaikan Git)
|
|
118
|
+
│ ├── 2408561071_Kelompok_BaliNER.pdf # Laporan dokumentasi tugas
|
|
119
|
+
│ ├── 2408561071_Kelompok_BaliNER.zip # Arsip penyerahan mandiri
|
|
120
|
+
│ └── 2408561071_Kelompok_BaliNER/ # Folder build penyerahan
|
|
121
|
+
├── best_balinese_ner_model_base/ # Model IndoBERT Base terlatih (konfigurasi & tokenizer)
|
|
122
|
+
├── static/ # Aset antarmuka web (CSS & JS)
|
|
123
|
+
│ ├── css/style.css # Gaya CSS profesional minimalis
|
|
124
|
+
│ └── js/app.js # Logika interaktif antarmuka klien
|
|
125
|
+
├── templates/ # Template Jinja2 Flask
|
|
126
|
+
│ └── index.html # Template HTML BaliNER Studio
|
|
127
|
+
├── app.py # Peladen web Flask & backend REST API
|
|
128
|
+
├── infer_ner.py # Utilitas CLI inferensi teks kustom
|
|
129
|
+
├── train_ner.py # Skrip CLI pelatihan & evaluasi Transformer
|
|
130
|
+
├── compare_models.py # Program evaluasi & komparasi benchmark seluruh model
|
|
131
|
+
├── BaliNER.ipynb # Jupyter Notebook eksperimen & analisis data
|
|
132
|
+
├── run_web_ui.bat # Launcher 1-klik untuk sistem operasi Windows
|
|
133
|
+
├── run_web_ui.sh # Launcher 1-klik untuk Linux / macOS
|
|
134
|
+
├── requirements.txt # Daftar dependensi Python
|
|
135
|
+
├── CONTRIBUTING.md # Panduan kontribusi pengembangan
|
|
136
|
+
├── LICENSE # Lisensi proyek (MIT)
|
|
137
|
+
└── README.md # Dokumentasi utama proyek
|
|
138
|
+
```
|
|
139
|
+
|
|
140
|
+
---
|
|
141
|
+
|
|
142
|
+
## 💻 Panduan Penggunaan Sistem
|
|
143
|
+
|
|
144
|
+
### Opsi 0: Python Package Hybrid (`balinese-ner`) — *Untuk Integrasi Developer & Redundansi Server*
|
|
145
|
+
Paket Python yang dapat di-`pip install` dengan dukungan **cache lokal/VPS**, **Hugging Face Hub (`Dramaniako/BaliBERT`)**, **Hugging Face Inference API**, **backup mirror fallback (GitHub Releases)**, serta **Constrained Viterbi BIO Decoding** (0 pelanggaran BIO):
|
|
146
|
+
|
|
147
|
+
```bash
|
|
148
|
+
pip install git+https://github.com/Dramaniako/BaliNER.git
|
|
149
|
+
```
|
|
150
|
+
|
|
151
|
+
```python
|
|
152
|
+
from balinese_ner import PretrainedModelNER
|
|
153
|
+
|
|
154
|
+
# Mode Hybrid (Default): cek cache lokal/VPS -> unduh HF Hub -> fallback mirror jika HF down
|
|
155
|
+
ner = PretrainedModelNER.load("BaliBERT", mode="hybrid")
|
|
156
|
+
print(ner.predict("I Pekak Dukuh lantas ngaukin Sang Lutung ane sedeng ngamah biu."))
|
|
157
|
+
|
|
158
|
+
# Mode API dengan Fallback Lokal (hemat RAM VPS, otomatis fallback ke lokal jika API HF gangguan)
|
|
159
|
+
ner_api = PretrainedModelNER.load("BaliBERT", mode="api_fallback")
|
|
160
|
+
|
|
161
|
+
# Mode Offline Murni (menggunakan cache/bobot lokal tanpa internet)
|
|
162
|
+
ner_offline = PretrainedModelNER.load("BaliBERT", mode="local")
|
|
163
|
+
```
|
|
164
|
+
|
|
165
|
+
Untuk mengisi cache pada server VPS sebelum demo (*pre-warming*):
|
|
166
|
+
```bash
|
|
167
|
+
balinese-ner --prewarm --model BaliBERT
|
|
168
|
+
```
|
|
169
|
+
|
|
170
|
+
---
|
|
171
|
+
|
|
172
|
+
### Opsi 1: Antarmuka Web Interaktif (BaliNER Studio Web UI) — *Direkomendasikan*
|
|
173
|
+
Aplikasi web modern bergaya *profesional minimalis* yang terhubung langsung ke model `best_balinese_ner_model_base`:
|
|
174
|
+
- **Cara 1 (Windows 1-Click Launcher)**: Cukup klik ganda berkas **[run_web_ui.bat](file:///d:/Project/BaliNER/run_web_ui.bat)**.
|
|
175
|
+
- **Cara 2 (Terminal CLI)**:
|
|
176
|
+
```powershell
|
|
177
|
+
.\.venv\Scripts\python.exe app.py
|
|
178
|
+
```
|
|
179
|
+
Kemudian buka peramban web di:
|
|
180
|
+
👉 **`http://127.0.0.1:5000`**
|
|
181
|
+
|
|
182
|
+
#### ✨ Rangkuman Fitur Web UI:
|
|
183
|
+
- **Deteksi Bahasa Otomatis**: Mendeteksi kecocokan teks bahasa Bali secara instan dengan indikator persentase.
|
|
184
|
+
- **Visualisasi Teks Beranotasi**: Penandaan entitas interaktif dengan badge warna BIO dan *tooltip popover* detail (*Skor, Kategori, Rentang Span*).
|
|
185
|
+
- **Preset Contoh Satua Bali**: 5 kutipan sastra rakyat klasik siap uji dalam 1-klik.
|
|
186
|
+
- **Tabel Rincian Entitas & Bilah Keyakinan**: Tabel data lengkap dengan bilah progres keyakinan dan tombol salin per entitas.
|
|
187
|
+
- **Evaluasi & Penilaian Kualitas**: Fitur umpan balik (👍 Akurat / 👎 Perlu Koreksi) untuk mencatat hasil penilaian linguis.
|
|
188
|
+
- **Riwayat Analisis & Pemulihan Instan**: Histori deteksi tersimpan lokal di browser dengan kemampuan *restore* ke layar utama.
|
|
189
|
+
- **Mode Gelap & Terang**: Tema *professional minimalist* elegan dengan persistensi preferensi.
|
|
190
|
+
- **REST API Terbuka**: Endpoint `/api/predict`, `/api/info`, `/api/examples`, `/api/detect-language`, dan `/api/feedback`.
|
|
191
|
+
|
|
192
|
+
---
|
|
193
|
+
|
|
194
|
+
### Opsi 2: Menjalankan via Jupyter Notebook (Interaktif)
|
|
195
|
+
Anda dapat menjalankan notebook baseline utama:
|
|
196
|
+
👉 **[BaliNER.ipynb](file:///d:/Project/BaliNER/BaliNER.ipynb)** *(IndoBERT Base)*
|
|
197
|
+
|
|
198
|
+
Atau memilih notebook spesifik per model pada direktori **[`notebooks/`](file:///d:/Project/BaliNER/notebooks)** untuk eksperimen komparatif:
|
|
199
|
+
- 🇮🇩 **[BaliNER_IndoBERT_base.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_IndoBERT_base.ipynb)** (`indobenchmark/indobert-base-p1`)
|
|
200
|
+
- 🐘 **[BaliNER_IndoBERT_large.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_IndoBERT_large.ipynb)** (`indobenchmark/indobert-large-p1`)
|
|
201
|
+
- 🏝️ **[BaliNER_NusaBERT.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_NusaBERT.ipynb)** (`LazarusNLP/NusaBERT-base`)
|
|
202
|
+
- 🇮🇩 **[BaliNER_IndoRoBERTa.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_IndoRoBERTa.ipynb)** (`cahya/roberta-base-indonesian-1.5G`)
|
|
203
|
+
- 🤖 **[BaliNER_RoBERTa.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_RoBERTa.ipynb)** (`roberta-base`)
|
|
204
|
+
- 🌐 **[BaliNER_XLM_RoBERTa.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_XLM_RoBERTa.ipynb)** (`xlm-roberta-base`)
|
|
205
|
+
- ⚡ **[BaliNER_mDeBERTa_v3.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_mDeBERTa_v3.ipynb)** (`microsoft/mdeberta-v3-base`)
|
|
206
|
+
- 🌍 **[BaliNER_mBERT.ipynb](file:///d:/Project/BaliNER/notebooks/BaliNER_mBERT.ipynb)** (`bert-base-multilingual-cased`)
|
|
207
|
+
|
|
208
|
+
### Opsi 3: Pelatihan Cepat via Terminal (CLI)
|
|
209
|
+
```bash
|
|
210
|
+
python train_ner.py --epochs 5 --batch_size 16 --learning_rate 3e-5
|
|
211
|
+
```
|
|
212
|
+
|
|
213
|
+
### Opsi 4: Menjalankan Inferensi pada Kalimat Baru via CLI
|
|
214
|
+
```bash
|
|
215
|
+
python infer_ner.py --text "I Pekak Dukuh lantas ngaukin Sang Lutung ane sedeng ngamah biu di duur punyane."
|
|
216
|
+
```
|
|
217
|
+
|
|
218
|
+
### Opsi 5: Menjalankan Pengujian Otomatis (Unit & Integration Tests)
|
|
219
|
+
```bash
|
|
220
|
+
python tests/test_web_ui.py
|
|
221
|
+
# Atau menggunakan discovery:
|
|
222
|
+
python -m unittest discover tests
|
|
223
|
+
```
|
|
224
|
+
|
|
225
|
+
### Opsi 6: Melakukan Komparasi Performa Seluruh Model Terlatih (Benchmark CLI)
|
|
226
|
+
Untuk mengevaluasi seluruh model yang telah dilatih dan menghasilkan laporan perbandingan lengkap (tabel leaderboard terminal, berkas CSV, JSON, Markdown, serta visualisasi grafik PNG):
|
|
227
|
+
```bash
|
|
228
|
+
python compare_models.py
|
|
229
|
+
```
|
|
230
|
+
*Argumen opsional:*
|
|
231
|
+
- `--test_file`: Lokasi berkas data uji (default: `data/test.xlsx`)
|
|
232
|
+
- `--batch_size`: Ukuran batch evaluasi (default: `32`)
|
|
233
|
+
- `--output_dir`: Direktori penyimpanan laporan (default: `reports/`)
|
|
234
|
+
- `--no_plots`: Lewati pembuatan grafik gambar PNG
|
|
@@ -0,0 +1,46 @@
|
|
|
1
|
+
[build-system]
|
|
2
|
+
requires = ["setuptools>=68.0"]
|
|
3
|
+
build-backend = "setuptools.build_meta"
|
|
4
|
+
|
|
5
|
+
[project]
|
|
6
|
+
name = "balinese-ner"
|
|
7
|
+
version = "0.1.0"
|
|
8
|
+
description = "Hybrid Balinese Named Entity Recognition (BaliNER) package with Hugging Face Hub, Inference API, local/VPS caching, mirror fallback, and grammar-constrained Viterbi decoding."
|
|
9
|
+
readme = "README.md"
|
|
10
|
+
requires-python = ">=3.9"
|
|
11
|
+
license = { text = "MIT" }
|
|
12
|
+
authors = [
|
|
13
|
+
{ name = "Dramaniako" }
|
|
14
|
+
]
|
|
15
|
+
keywords = ["balinese", "ner", "nlp", "indobert", "transformers", "satua-bali", "viterbi"]
|
|
16
|
+
classifiers = [
|
|
17
|
+
"Development Status :: 4 - Beta",
|
|
18
|
+
"Intended Audience :: Developers",
|
|
19
|
+
"Intended Audience :: Science/Research",
|
|
20
|
+
"License :: OSI Approved :: MIT License",
|
|
21
|
+
"Programming Language :: Python :: 3",
|
|
22
|
+
"Topic :: Scientific/Engineering :: Artificial Intelligence",
|
|
23
|
+
"Topic :: Text Processing :: Linguistic",
|
|
24
|
+
]
|
|
25
|
+
dependencies = [
|
|
26
|
+
"torch>=2.2.0",
|
|
27
|
+
"transformers>=4.38.0",
|
|
28
|
+
"huggingface_hub>=0.24.0",
|
|
29
|
+
"safetensors>=0.4.0",
|
|
30
|
+
"numpy>=1.24.0",
|
|
31
|
+
"requests>=2.31.0",
|
|
32
|
+
]
|
|
33
|
+
|
|
34
|
+
[project.urls]
|
|
35
|
+
Homepage = "https://github.com/Dramaniako/BaliNER"
|
|
36
|
+
Repository = "https://github.com/Dramaniako/BaliNER"
|
|
37
|
+
ModelHub = "https://huggingface.co/Dramaniako/BaliBERT"
|
|
38
|
+
|
|
39
|
+
[project.scripts]
|
|
40
|
+
balinese-ner = "balinese_ner.cli:main"
|
|
41
|
+
|
|
42
|
+
[tool.setuptools.packages.find]
|
|
43
|
+
where = ["src"]
|
|
44
|
+
|
|
45
|
+
[tool.setuptools.package-data]
|
|
46
|
+
balinese_ner = ["weights/**/*"]
|
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
"""Hybrid Balinese Named Entity Recognition (`balinese_ner`) Python package."""
|
|
2
|
+
|
|
3
|
+
from __future__ import annotations
|
|
4
|
+
|
|
5
|
+
import os
|
|
6
|
+
|
|
7
|
+
if os.name == "nt":
|
|
8
|
+
os.environ.setdefault("KMP_DUPLICATE_LIB_OK", "TRUE")
|
|
9
|
+
|
|
10
|
+
from .model import PretrainedModelNER
|
|
11
|
+
from .registry import ENTITY_METADATA, MODEL_REGISTRY
|
|
12
|
+
|
|
13
|
+
BaliNER = PretrainedModelNER
|
|
14
|
+
|
|
15
|
+
__all__ = [
|
|
16
|
+
"PretrainedModelNER",
|
|
17
|
+
"BaliNER",
|
|
18
|
+
"MODEL_REGISTRY",
|
|
19
|
+
"ENTITY_METADATA",
|
|
20
|
+
"__version__",
|
|
21
|
+
]
|
|
22
|
+
__version__ = "0.1.0"
|