pytorch_sentiment_rnn icon indicating copy to clipboard operation
pytorch_sentiment_rnn copied to clipboard

Published 20 hours ago • verified publisher icon vanzytay

Task Definition

Open sidoki opened this issue 8 years ago • 5 comments

Hi @vanzytay,

just want to make sure. So given a text review, it should return the aspect with its term, right? Thank you!

sidoki avatar May 10 '17 10:05 sidoki

hi @dyngts

Given the review and term (a sequence of words existing in the review), the model will return the sentiment. (is this what you mean by aspect?).

vanzytay avatar May 10 '17 12:05 vanzytay

hi @vanzytay,

Do you mean the model will return overall sentiment polarity for specific review? i thought at first, it will return the aspects (what i mean by aspect is like food, service, ambience in restaurant aspect) with its polarity instead only overall sentiment.

What do you think about that?

Thank you

sidoki avatar May 12 '17 02:05 sidoki

No, we only return sentiments and assume the aspect is given.

This is in concert with works like https://aclweb.org/anthology/D16-1058 which doesn't consider the extraction of aspects and assume they are given.

vanzytay avatar May 13 '17 01:05 vanzytay

Hi @vanzytay,

Thanks for the clarification. My concern is the model cannot be used for production when there's always new reviews that don't have any pre-defined annotate review.

sidoki avatar May 15 '17 02:05 sidoki

You can always use an off-the-shelf entity extractor for the aspect terms. :)

vanzytay avatar May 15 '17 02:05 vanzytay