search

crownpku / Information-Extraction-Chinese

Chinese Named Entity Recognition with IDCNN/biLSTM+CRF, and Relation Extraction with biGRU+2ATT 中文实体识别与关系提取
2.22k stars 814 forks source link

[RE_BGRU_2ATT] entity location embedding issue #1

Open crownpku opened 7 years ago

crownpku commented 7 years ago

Current entity location embedding is the first character of the entity, the rest of the entity characters are calculated as part of the sentence.

An improvement is to rewrite the original sentence, and make all characters within the entity as one unit only so that only entity location is embedded, and characters within the entity will not be involved into embedding calculation.

longbowking commented 6 years ago

I think @Mariobai provides a good solution #22 , which makes rel_e1_posId of certain key word/char to be independent of the length of entity1. For example:

  1. Mike and John are good friends.
  2. Cristiano Ronaldo Dos Santos Aveiro and Messi are friends?