implement Dataset Reader in Paddle Book

[reyoung]

This issue is a part of #1392.

There are eight books in the current Paddle Book. We need to write eight reader creators for each dataset. There are:

• 新手入门
• 识别数字
  • http://scikit-learn.org/stable/datasets/ sklearn.dataset
• 图像分类
  • cifar https://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz
• 词向量
  • By @wangkuiyi
  • It seems we could use nltk.corpus.treebank to fetch the exactly dataset used in Book.
• 情感分析
  • By @wen-bo-yang
  • The IMDB dataset is not included in nltk.corpus, but the Amazon reviews is included. So could we change the dataset used in Book?
• 语义角色标注
  • By @reyoung
  • The book uses CONLL 2005 dataset. The nltk.corpus contains CONLL 2000, CONLL 2002, CONLL 2007 dataset. Could we changed the dataset into CONLL 2007?
• 机器翻译
  • nltk.corpus contains WMT dataset.
• 个性化推荐

[wangkuiyi]

data cache 目录 /usr/local/paddle/dataset

[reyoung]

/usr/local/paddle/dataset

这个目录，用户没有root权限就没法写

[wen-bo-yang]

我在nltk.corpus中找到了影评资源，内容和IMDB的组织形式差不都，我想是否可以使用这个文本数据做情感分类的数据

/nltk_data/corpora/movie_reviews/neg# ls
cv000_29416.txt  cv143_21158.txt .....
/nltk_data/corpora/movie_reviews/pos# ls
cv000_29590.txt  cv143_19666.txt .....