cross-domain aspect extraction
http://www.zmonster.me/2016/06/08/use-stanford-nlp-package-in-nltk.html
https://blog.csdn.net/weixin_38246633/article/details/80637192
res get 0.73 with only gen_latent
[*] loading parameters success!!!
epoch:0, batch:165, loss:0.0110373273491859446
acc1: 0.90263 acc2: 0.962184 f1_score: 0.71351380254
epoch:1, batch:165, loss:0.0125725995749235156
acc1: 0.913277 acc2: 0.968922 f1_score: 0.732774935103
epoch:2, batch:165, loss:0.0085210474207997326
acc1: 0.910676 acc2: 0.965071 f1_score: 0.724399417143