Unsupervised Aspect Term Extraction for Sentiment Analysis through Automatic Labeling

Danny Vargas, Lucas Pessutto, Viviane Moreira

2022

Abstract

In sentiment analysis, there has been growing interest in performing finer granularity analysis focusing on entities and their aspects. This is the goal of Aspect-based Sentiment Analysis which commonly involves the following tasks: Opinion Target Extraction (OTE), Aspect term extraction (ATE), and polarity Classification (PC). This work focuses on the second task, which is the more challenging and least explored in the unsupervised context. The difficulty arises mainly due to the nature of the data (user-generated contents or product reviews) and the inconsistent annotation of the evaluation datasets. Existing approaches for ATE and OTE either depend on annotated data or are limited by the availability of domain- or language-specific resources. To overcome these limitations, we propose UNsupervised Aspect Term Extractor (UNATE), an end-to-end unsupervised ATE solution. Our solution relies on a combination of topic models, word embeddings, and a BERT-based classifier to extract aspects even in the absence of annotated data. Experimental results on datasets from different domains have shown that UNATE achieves precision and F-measure scores comparable to the semi-supervised and unsupervised state-of-the-art ATE solutions.

Download


Paper Citation


in Harvard Style

Vargas D., Pessutto L. and Moreira V. (2022). Unsupervised Aspect Term Extraction for Sentiment Analysis through Automatic Labeling. In Proceedings of the 18th International Conference on Web Information Systems and Technologies - Volume 1: WEBIST, ISBN 978-989-758-613-2, pages 344-354. DOI: 10.5220/0011549100003318


in Bibtex Style

@conference{webist22,
author={Danny Vargas and Lucas Pessutto and Viviane Moreira},
title={Unsupervised Aspect Term Extraction for Sentiment Analysis through Automatic Labeling},
booktitle={Proceedings of the 18th International Conference on Web Information Systems and Technologies - Volume 1: WEBIST,},
year={2022},
pages={344-354},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0011549100003318},
isbn={978-989-758-613-2},
}


in EndNote Style

TY - CONF

JO - Proceedings of the 18th International Conference on Web Information Systems and Technologies - Volume 1: WEBIST,
TI - Unsupervised Aspect Term Extraction for Sentiment Analysis through Automatic Labeling
SN - 978-989-758-613-2
AU - Vargas D.
AU - Pessutto L.
AU - Moreira V.
PY - 2022
SP - 344
EP - 354
DO - 10.5220/0011549100003318