Title Trigger - All titles you want

Google ML Camp-燃烧我的GPU-项目

Progress

• Tag Classification

• (Pipeline) Training code
• (Pipeline) Test code
• (Pipeline) Inference API
• (Model) PoolLSTM
• (Model) RCNN+
• (Model) Bert

• Keywords Extraction

• (Pipeline) Test code
• (Pipeline) Inference API

• Abstractive Summarization

• (Pipeline) Training code
• (Pipeline) Test code
• (Pipeline) Inference API
• (Model) Seq2seq+Attn
• (Model) Effective Pointer generator

• Text Style Transfering

• (Pipeline) Training code
• (Pipeline) Test code

• Demo

• Basic Layout
• Attention Visualization

Usage

1. Download the code and install

$ git@github.com:OwenMasculinity/ML-Camp-BurnMyGpu.git
$ python setup.py install

2. API

2.1 Keywords Extraction

from titletrigger.api import extract_keywords

content = "国务院总理李克强21日下午在中南海紫光阁会见中印边界问题印方特别代表"
extract_keywords(content, K=3)

2.2 Extractive Summarization

from titletrigger.api import ext_summarize
content = "国务院总理李克强21日下午在中南海紫光阁会见中印边界问题印方特别代表xxxx"

ext_headline = " ".join(ext_summarize([content])[0])

2.3 Tag classification

import sys
from titletrigger.api import tag_classify
sys.path.append('titletrigger/textclf')
clf_model_path = "YOU_MODEL_PATH"
clf_model_file = load_model(clf_model_path)
content = "国务院总理李克强21日下午在中南海紫光阁会见中印边界问题印方特别代表xxxx"
tag = tag_classify([content], clf_model_file)

pre-trained models can be downloaded in this link

2.4 Abstractive classification

import sys
from titletrigger.api import abs_summarize
sys.path.append('titletrigger/textsum')
clf_model_path = "YOU_MODEL_PATH"
content = "国务院总理李克强21日下午在中南海紫光阁会见中印边界问题印方特别代表xxxx"
result_dict = abs_summarize([content_list], sum_model_file)
tag = tag_classify(content_list, clf_model_file)

pre-trained models can be downloaded in this link

2.5 Text Style Transfer

from titletrigger.api import style_transfer
content = "国务院总理李克强21日下午在中南海紫光阁会见中印边界问题印方特别代表xxxx"
transferred_text = style_transfer(content)

3. Training

3.1 Tag Classification

Downloading data from this link split it into train.csv, valid.csv, test.csv move them into titletrigger/textclf/data/chinese_news

$ cd titletrigger/textclf
$ ls data/chinese_news
train.csv
valid.csv
test.csv
$ python train.py -config conifg/config_rcnn.json

3.2 Abstractive Summarization

move data splits into titletrigger/textclf/data/chinese_news

$ cd titletrigger/textclf
$ ls data/chinese_news
train.csv
valid.csv
test.csv
$ python train.py -config conifg/config_copynet.json

4. Demo

First open a httpserver

$ cd website
$ python -m http.server 8000

Open http://localhost:8000/demo.html in your browser