Paper is this.
This code use livedoor_news_corpus instead of youtube movie courpus, because we could not get it.
download learned word2vec model from (https://github.com/Kyubyong/wordvectors). and put in project dir, for example
$ mv ja.bin ~/../youtube_recommendation/.
get mecab
$ brew install mecab mecab-ipadic
$ git clone --depth 1 https://github.com/neologd/mecab-ipadic-neologd.git
$ cd mecab-ipadic-neologd
$ ./bin/install-mecab-ipadic-neologd -n
get liverdoor news corpus
$ pwd
~/../youtube_recommendation
$ wget https://www.rondhuit.com/download/ldcc-20140209.tar.gz
$ tar xvzf ldcc-20140209.tar.gz
train
$ python3 main.py -target_model ['c' or 'r'] # c is trainning candidate model, r is trainnig ranking model