Aurelius84 / ugc-sentiment

sentinews

Machine built to monitor public opinion. Models based on bidirectial LSTM with attention was trained by online ugc-comment from 大众点评.

Requirements

• Keras >= 2.0
• tensorflow >= 1.1.0
• h5py
• matplotlib
• jieba
• PyYAML
• python2.7

How to use

• Install Package

$ pip install -U ugc_sentiment

• sentiment

# test.py
# coding: utf-8
from ugc_sentiment import UgcSentiment
conts = [
  u'服务态度太差了。团购去吃的，中午到的时候不到12点吧，手机在12点3分收到了已使用团购券。到了12点35的时候只上了一个菜，我实在是受不了了，就催服务员，服务员说让我稍等一下，饭菜20分钟左右就能好，我都等了40分钟了，才上一个菜，旁边的桌比我们晚来，菜上的都比我们多，你TMD能不开团购啊，能就开不能就别TM凑热闹 。。。。',
  u'店里的食物不算太全但食物味道还算可以，店员服务态度很好，价格比起同类的有些小贵，店面还是比较整洁的'
]
if __name__ == '__main__':
    # it will load model automatically
    senti = UgcSentiment()
    # receive params of unicode string
    result = senti.sentiment(conts[0])

• result：

    {
    "labels": "2",
    "probs_json": {
        "1": "0.43460175,,",
        "2": "0.44080707,",
        "3": "0.0980910137296",
        "4": "0.007562303,",
        "5": "0.0075623029843",
        },
    "probs": "0.43460175395 0.440807074308 0.0980910137296 0.0189378540963 0.0075623029843"
    }

• Support list of unicode string

    # also receive list of unicode string
    results = senti.sentiment(conts)

• result:

    [{
    "labels": "2",
    "probs_json": {
        "1": "0.43460175,,",
        "2": "0.44080707,",
        "3": "0.0980910137296",
        "4": "0.007562303,",
        "5": "0.0075623029843",
        },
    "probs": "0.43460175395 0.440807074308 0.0980910137296 0.0189378540963 0.0075623029843"
  },
    {
    "labels": "3",
    "probs_json": {
        "1": "0.0150475967675,,",
        "2": "0.118314310908,",
        "3": "0.647881686687",
        "4": "0.167027801275,",
        "5": "0.0517285950482",
        },
    "probs": "0.0150475967675 0.118314310908 0.647881686687 0.167027801275 0.0517285950482"
    }]

License

GNU (for details, please refer to LICENSE)

Copyright (c) 2017 LiujieZhang