search

teapot123 / JASen

Code and Data for our EMNLP-2020 paper Weakly-Supervised Aspect-Based Sentiment Analysis via Joint Aspect-Sentiment Topic Embedding.
49 stars 5 forks source link

The evaluation of sentiment #3

Closed GeneZC closed 3 years ago

GeneZC commented 3 years ago

Accordig to my understanding, the evaluation of sentiment prediction is conducted at sentence-level instead of aspect-level, right?

teapot123 commented 3 years ago

Thanks for your question! For each sentence, we output a tuple of <aspect, sentiment> so both the aspect and sentiment of the sentence are captured. Our current method can only output one tuple for each sentence. Do you mean you want to capture the sentiment for each aspect in a sentence?

GeneZC commented 3 years ago

Yes, exactly.

teapot123 commented 3 years ago

Currently, our method is doing single-class classification for each sentence thus can only output one tuple for each sentence. Extracting sentiment at an aspect level can be a future work, such as extending it into a multi-class classification problem. Then a possible solution can be changing the single-class classification objective into a multi-class objective one in the CNN training objective.

GeneZC commented 3 years ago

Thanks a lot!