Minor error of aspect term extraction

[DorisFangWork]

Hi Author,

I used fast_lcfs_atepc model to train my own dataset about "Immigration", the result is pretty good:

• Aspect sentiment classification accuracy: 80.54%
• Aspect sentiment classification F1: 80.76%
• Aspect term extraction F1: 97.84%

However, I noticed minor errors. Original sentence:"and gov keeps importing more wp and ep , and they stay like 5 person in 1 room to survive and depress everyone else ' s wages" Results from model inference and inference JSON file not match, position should be [6] and [8], and "keeps" is an entire word.

[yangheng95]

Hi, Thanks for your report. Would you share your dataset for debugging?