KPC-CF: Korean Aspect-Based Sentiment Analysis via NLI-Based Pseudo-Classifier with Corpus Filtering

Anonymous

TL;DR: Korean aspect-based-sentiment analysis, NLI-based pseudo classifier, Translation, LaBSE-based filtering, Cross-lingual PLM

Abstract:

Previous research on Aspect-Based Sentiment Analysis (ABSA) for Korean reviews in the restaurant domain not has been conducted.
Nowadays, most state-of-the-art results for a wide array of NLP tasks are achieved by utilizing pre-trained language representation. This paper seeks to develop a LM-based pseudo classifier that generates the best prediction labels by integrating translated data and unlabeled actual Korean data. We utilized the common ML concept of semi-supervised learning, along with LaBSE-based filtering, on the basis of transformation to the sentence-pair task and fine-tuned the crosslingual model. This achieved state-of-the-art results in Korean ABSA with low resources, showing approximately a 3\% difference in F1 scores and accuracy compared to English ABSA results.
We show the model and data for Korean ABSA, publicly available at https://huggingface.co/KorABSA.

Paper Type: long

Research Area: Sentiment Analysis, Stylistic Analysis, and Argument Mining

Contribution Types: Publicly available software and/or pre-trained models, Data resources

Languages Studied: Korean