Add model card
Browse files
README.md
ADDED
|
@@ -0,0 +1,62 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
license: apache-2.0
|
| 5 |
+
base_model: bert-base-uncased
|
| 6 |
+
pipeline_tag: text-classification
|
| 7 |
+
tags:
|
| 8 |
+
- text-classification
|
| 9 |
+
- bert
|
| 10 |
+
- disaster-tweets
|
| 11 |
+
- kaggle
|
| 12 |
+
datasets:
|
| 13 |
+
- startificial/nlp-getting-started
|
| 14 |
+
metrics:
|
| 15 |
+
- accuracy
|
| 16 |
+
- f1
|
| 17 |
+
---
|
| 18 |
+
|
| 19 |
+
# Disaster Tweet Detection(灾难推文检测)
|
| 20 |
+
|
| 21 |
+
基于 `bert-base-uncased` 微调的英文推文二分类模型,判断一条推文是否**关于真实灾难**
|
| 22 |
+
(Kaggle 竞赛 *Real or Not? NLP with Disaster Tweets*,`nlp-getting-started`)。
|
| 23 |
+
|
| 24 |
+
## 训练数据
|
| 25 |
+
|
| 26 |
+
- 来源:Kaggle 灾难推文数据集(7613 条训练样本)
|
| 27 |
+
- 标签:0 = 非灾难推文,1 = 灾难推文
|
| 28 |
+
- 训练:2 个 epoch,lr=2e-5,batch=16,max_len=128
|
| 29 |
+
|
| 30 |
+
## 验证集指标
|
| 31 |
+
|
| 32 |
+
| 指标 | 数值 |
|
| 33 |
+
| --- | --- |
|
| 34 |
+
| Accuracy | **0.8451** |
|
| 35 |
+
| F1 (binary) | **0.8168** |
|
| 36 |
+
| Precision | 0.8297 |
|
| 37 |
+
| Recall | 0.8043 |
|
| 38 |
+
|
| 39 |
+
## 使用方式
|
| 40 |
+
|
| 41 |
+
```python
|
| 42 |
+
from transformers import AutoTokenizer, AutoModelForSequenceClassification
|
| 43 |
+
import torch
|
| 44 |
+
|
| 45 |
+
model_id = "Operaer/disaster-tweet-bert"
|
| 46 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
| 47 |
+
model = AutoModelForSequenceClassification.from_pretrained(model_id)
|
| 48 |
+
model.eval()
|
| 49 |
+
|
| 50 |
+
text = "Forest fire near La Ronge Sask. Canada"
|
| 51 |
+
enc = tokenizer(text, truncation=True, padding=True,
|
| 52 |
+
max_length=128, return_tensors="pt")
|
| 53 |
+
with torch.no_grad():
|
| 54 |
+
logits = model(**enc).logits
|
| 55 |
+
pred = int(torch.argmax(logits, dim=1))
|
| 56 |
+
print("灾难推文" if pred == 1 else "非灾难推文")
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
## 说明
|
| 60 |
+
|
| 61 |
+
- 本模型为《自然语言处理》课程作业产物;
|
| 62 |
+
- 微调代码:`train.py`;交互式 Demo:`app.py` / Gradio Space。
|