bhkai: canlı sürüm (bhkai)
Browse files- README.md +44 -0
- agirlik.npz +3 -0
- bilgi.json +86 -0
- sozluk.json +0 -0
README.md
ADDED
|
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
license_name: proprietary
|
| 4 |
+
language:
|
| 5 |
+
- tr
|
| 6 |
+
tags:
|
| 7 |
+
- text-classification
|
| 8 |
+
- turkish
|
| 9 |
+
- news
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# BHKAI: Turkish news category classifier
|
| 13 |
+
|
| 14 |
+
> **Private repository, for the BültenHub team only.**
|
| 15 |
+
|
| 16 |
+
BHKAI assigns a category to each incoming story. When it is confident it answers on its own; otherwise the story goes to the AI model.
|
| 17 |
+
|
| 18 |
+
## Model
|
| 19 |
+
|
| 20 |
+
**Method:** TF-IDF (300,000 features, min_df 3) with an SGD classifier (alpha 2e-6).
|
| 21 |
+
|
| 22 |
+
**Version:** `bhkai-20261030`, trained nightly on the home PC and promoted automatically after its exam.
|
| 23 |
+
|
| 24 |
+
**Training data:** 437,769 stories labelled by the teacher model (`kategori_kaynagi='ai'`). Data cut-off: 2026-10-30 03:24.
|
| 25 |
+
|
| 26 |
+
## Exam
|
| 27 |
+
|
| 28 |
+
The exam uses the last 2 days: 749 stories, half of them, split by crc32.
|
| 29 |
+
|
| 30 |
+
| Threshold | Coverage | Agreement |
|
| 31 |
+
|---|---|---|
|
| 32 |
+
| 0.8 (live) | 0.512 | 0.997 |
|
| 33 |
+
| 0.7 | 0.659 | 0.973 |
|
| 34 |
+
| 0.5 | 0.855 | 0.902 |
|
| 35 |
+
|
| 36 |
+
The previous live version scored coverage 0.488 and agreement 1.000 at threshold 0.8.
|
| 37 |
+
|
| 38 |
+
## Files
|
| 39 |
+
|
| 40 |
+
- `agirlik.npz`: weights
|
| 41 |
+
- `sozluk.json`: vocabulary
|
| 42 |
+
- `bilgi.json`: version, exam and comparison details
|
| 43 |
+
|
| 44 |
+
They are loaded by BültenHub's `pipeline/bhkai_*` code.
|
agirlik.npz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ec3300fbba558396616738c888ae4bf7052df637adfff3a5f30f64c2f78f6b00
|
| 3 |
+
size 22586051
|
bilgi.json
ADDED
|
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"surum": "bhkai-20261030",
|
| 3 |
+
"tarih": "2026-10-30T23:02:54",
|
| 4 |
+
"ogretmen": "mini (kategori_kaynagi='ai')",
|
| 5 |
+
"egitim_adet": 437769,
|
| 6 |
+
"veri_kesimi": "2026-10-30T03:24:00",
|
| 7 |
+
"sinav": {
|
| 8 |
+
"gun": 2,
|
| 9 |
+
"adet": 749,
|
| 10 |
+
"uyum": 0.6996,
|
| 11 |
+
"esikler": {
|
| 12 |
+
"0.5": {
|
| 13 |
+
"kapsam": 0.8552,
|
| 14 |
+
"uyum": 0.9019
|
| 15 |
+
},
|
| 16 |
+
"0.6": {
|
| 17 |
+
"kapsam": 0.7677,
|
| 18 |
+
"uyum": 0.9402
|
| 19 |
+
},
|
| 20 |
+
"0.7": {
|
| 21 |
+
"kapsam": 0.6588,
|
| 22 |
+
"uyum": 0.9728
|
| 23 |
+
},
|
| 24 |
+
"0.75": {
|
| 25 |
+
"kapsam": 0.5986,
|
| 26 |
+
"uyum": 0.988
|
| 27 |
+
},
|
| 28 |
+
"0.8": {
|
| 29 |
+
"kapsam": 0.5118,
|
| 30 |
+
"uyum": 0.9972
|
| 31 |
+
},
|
| 32 |
+
"0.85": {
|
| 33 |
+
"kapsam": 0.3814,
|
| 34 |
+
"uyum": 1.0
|
| 35 |
+
},
|
| 36 |
+
"0.9": {
|
| 37 |
+
"kapsam": 0.172,
|
| 38 |
+
"uyum": 1.0
|
| 39 |
+
}
|
| 40 |
+
}
|
| 41 |
+
},
|
| 42 |
+
"ayar": {
|
| 43 |
+
"ozellik": 300000,
|
| 44 |
+
"min_df": 3,
|
| 45 |
+
"alpha": 2e-06
|
| 46 |
+
},
|
| 47 |
+
"karsilastirma": {
|
| 48 |
+
"canli": "bhkai-20261029",
|
| 49 |
+
"kesim": "2026-10-29T20:17:58",
|
| 50 |
+
"sinav_adet": 749,
|
| 51 |
+
"denetim_ornegi": 34,
|
| 52 |
+
"esik": 0.8,
|
| 53 |
+
"canli_esikler": {
|
| 54 |
+
"0.5": {
|
| 55 |
+
"kapsam": 0.8473,
|
| 56 |
+
"uyum": 0.8942
|
| 57 |
+
},
|
| 58 |
+
"0.6": {
|
| 59 |
+
"kapsam": 0.7434,
|
| 60 |
+
"uyum": 0.9373
|
| 61 |
+
},
|
| 62 |
+
"0.7": {
|
| 63 |
+
"kapsam": 0.6358,
|
| 64 |
+
"uyum": 0.9673
|
| 65 |
+
},
|
| 66 |
+
"0.75": {
|
| 67 |
+
"kapsam": 0.567,
|
| 68 |
+
"uyum": 0.9823
|
| 69 |
+
},
|
| 70 |
+
"0.8": {
|
| 71 |
+
"kapsam": 0.4875,
|
| 72 |
+
"uyum": 1.0
|
| 73 |
+
},
|
| 74 |
+
"0.85": {
|
| 75 |
+
"kapsam": 0.3584,
|
| 76 |
+
"uyum": 1.0
|
| 77 |
+
},
|
| 78 |
+
"0.9": {
|
| 79 |
+
"kapsam": 0.1434,
|
| 80 |
+
"uyum": 1.0
|
| 81 |
+
}
|
| 82 |
+
},
|
| 83 |
+
"karar": "yayina",
|
| 84 |
+
"neden": "eşik 0.8: kapsam %48.8→%51.2, uyum %100.0→%99.7 (749 sınav haberi)"
|
| 85 |
+
}
|
| 86 |
+
}
|
sozluk.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|