Personalized Hashtag Recommendation

Data

Weibo

  • Basic Statistics
Num Embed Train Test
Weibo TODO TODO TODO
User TODO TODO TODO
Hashtag TODO TODO TODO
  • Distribution

Twitter

  • Basic Statistics
Num Embed Train Test
Tweet TODO TODO TODO
User TODO TODO TODO
Hashtag TODO TODO TODO
  • Distribution

Model

  • bert+lstm+mlp
  • bert+mlp
  • bert+gbdt
  • Lda+gbdt
  • tfidf+gbdt

ranklib:

train: java -jar RankLib-2.14.jar -train trainBert.dat -ranker 0 -save modelBertGbdt.dat

test: java -jar RankLib-2.14.jar -load modelBertGbdt.dat -test testBert.dat -metric2T MAP@5 -idv BertGbdt.MAP@5.txt

** Description ** :

  1. train.dat & test.dat example #query user_index label(0 or 1, 1 positive) qid:user_index features... ...
  2. BertGbdt.MAP@5.txt : MAP of each user (qid)