使用LabelAwareAttention计算user embedding的一个疑问

shenweichen / DeepMatch

A deep matching model library for recommendations & advertising. It's easy to train models and to export representation vectors which can be used for ANN search.

https://deepmatch.readthedocs.io/en/latest/

Apache License 2.0

2.22k stars 530 forks source link

使用LabelAwareAttention计算user embedding的一个疑问 #56

Closed bubbleocean closed 3 years ago

bubbleocean commented 3 years ago

调用mind的时候翻看代码，有个关于LabelAwareAttention有个小疑问，如果要使用这个attention机制，user embedding的返回应该是user_embedding_final吧？（deepmatch.models.mind中150行的user embedding）即： model.setattr("user_embedding", user_embeddings) 改为：model.setattr("user_embedding", user_embedding_final)