berryhn / textgrocery_shortad

微信广告正负样本分类 textGrocery实现

TextGrocery_shortAd_Classification

微信广告正负样本分类

参考https://github.com/2shou/TextGrocery.git

Dependencies

Only test under Unix-based System

$ pip install tgrocery

正样本：

• xx,xxx,xxxxxxxxx

负样本：

• 微信公众号 AppSo，回复「钱包」看看微信钱包这 6 个秘密使用技巧

• 微信号：wszs1981

• 长按二维码关注

其中，训练集正负样本各1W+，测试集样本各5K+

训练数据集：data/train/tgrocery_train.txt

测试数据集：data/test/tgrocery_test.txt

输入数据为未分词的带标签数据，数据预处理:

python data/train/pre_processing.py
python data/test/pre_progressing.py

训练模型：

python train_model.py

Building prefix dict from the default dictionary ...
Loading model from cache /tmp/jieba.cache
Loading model cost 0.192 seconds.
Prefix dict has been built succesfully.
***.**
optimization finished, #iter = 18
Objective value = -554.355654
nSV = 9239
True

测试模型：

python test_model.py

result

# 0 : ad     1 : not_ad 

{u'1': 4973, u'0': 4115}
{u'1': 5021, u'0': 4160}
{'1': 5018, '0': 4163}
1:	 precision:0.991032	 recall:0.990440	 f:0.990736
0:	 precision:0.988470	 recall:0.989183	 f:0.988826