berlino / sa-sent Goto Github PK

Thanks for sharing the code.
I'm reproducing the results.
The data files and Jar file is placed in the corresponding place.
Then an error is encountered in running tokenizer.

INFO:CoreNLP_JavaServer: INPUT: 8000 documents, 397869 characters, 81940 tokens, 49.7 char/doc, 10.2 tok/doc RATES: 1047.806 doc/sec, 10732.2 tok/sec

> /Users/hongzhizhang/wp/sa_sentiment/SA-Sent/reader.py(125)to_index()
-> pdb.set_trace()
(Pdb) c
> /Users/hongzhizhang/wp/sa_sentiment/SA-Sent/reader.py(126)to_index()
-> if target_start < 0 or target_end < 0:  pdb.set_trace()
(Pdb) c
Traceback (most recent call last):
  File "reader.py", line 222, in <module>
    train, test = reader.read()
  File "reader.py", line 146, in read
    self.to_index(self.train_data)
  File "reader.py", line 126, in to_index
    if target_start < 0 or target_end < 0:  pdb.set_trace()
IndexError: list assignment index out of range
WARNING:CoreNLP_PyWrapper:Killing subprocess 42798
INFO:CoreNLP_PyWrapper:Subprocess seems to be stopped, exit code -9

Hello, I found this code due to a paper. So I try to run it on my machine. But the question is I can't get the best accuracy. It‘s even 20% lower than the accuracy written on the paper on python 3.6 and 10% lower on python 2.7. Could I adjust some configs or python condition to get better accuracy? Looking forward to your reply.

您好，最近读了您的论文，有一些不明白的地方，比如说若一句话中含有两个目标，我怎么对它进行标注呢？标注目标、标注表达意见的词以及表达意见的词的情感类别？这样会不会太麻烦了呢？还有就是最后的结果是判断每个目标分别的情感对吗？这个结果如何得出呢？因为之前都是一句话的情感，最后生成一个每类情感的概率，您这个是生成每个目标分别每类情感概率吗？您举得模型是一个目标及它的意见表达，所以有点不明白。可以交流下吗？请问您邮箱是什么呢？

hi，my python environment is 3.5，so i use stanford_corenlp_pywrapper has error，i found that reader.py uses tokenizer.tokenize ,Can I replace tokenizer.tokenize (sent_str) with word_tokenize (sent_str)?

def tokenize(self, sent_str):
sent_str = " ".join(sent_str.split("-"))
sent_str = " ".join(sent_str.split("/"))
sent_str = " ".join(sent_str.split("!"))
return word_tokenize(sent_str)
#return tokenizer.tokenize(sent_str)

I need your help ,thanks