Span Detection for Aspect-Based Sentiment Analysis in Vietnamese

Span Detection for Aspect-Based Sentiment Analysis in Vietnamese Kim Thi-Thanh Nguyen, Sieu Khai Huynh, Phuc Huynh Pham, Luong Luc Phan, Duc-Vu Nguyen, Kiet Van Nguyen University of Information Technology, Ho Chi Minh City, Vietnam Vietnam National University, Ho Chi Minh City, Vietnam {18520963,18520348,18521260,18521073}@gm.uit.edu.vn {vund, kietnv}@uit.edu.vn Abstract Aspect-based sentiment analysis is a chal- lenging task that plays an essential role in natural language processing (NLP) and arti- ficial intelligence. However, recent works only focused on aspect detection and senti- ment classification but ignored the sub-task of detecting user opinion, which has im- mense potential in practical applications. In this paper, we present a new Vietnamese dataset (UIT-ViSD4SA) consisting of 35,396 human-annotated spans on 11,122 feedback comments for evaluating span detection for aspect-based sentiment analysis. Besides, we also propose a novel system using Bidirec- tional Long Short-Term Memory (BiLSTM) with a Conditional Random Field (CRF) layer (BiLSTM-CRF) for the span detection task in Vietnamese. The best result is a 62.76% F1macro for span detection using BiLSTM- CRF with embedding fusion of syllable em- bedding, character embedding, and contex- tual embedding from XLM-RoBERTa. In fu- ture work, span detection will be extended in many NLP tasks such as opinion min- ing, emotion recognition, complaint analy- sis, and constructive detection. Our dataset is freely available at https://github. com/kimkim00/UIT-ViSD4SA for re- search purposes. 1 Introduction Typically, before buying an item or deciding to use a service, people tend to seek advice from their prede- cessors who purchased the item or used the service. With the explosive growth of e-commerce, more and more people find advice from websites, e-commerce sites, forums, or product review channels. There- fore, the number of reviews is increasing and becom- ing a valuable resource for customers and business. For customers, this data source provides information about products and helpful advice to help them avoid buying products or signing up for services that are not suitable for their personal needs. On the other hand, user reviews are also valuable information for businesses, and if used correctly and effectively, this data can help businesses improve product quality, accurately identify the target customers for each seg- ment. Aspect-based sentiment analysis (ABSA) (Hu and Liu, 2004) on user feedback is a challenging natural language processing task that attracts interest from both research and business (Jo and Oh, 2011; Kir- itchenko et al., 2014; Chen et al., 2017). Given spe- cific feedback about a product or service, the main task of ABSA is to detect what aspect is being dis- cussed, then give sentiment analysis to the explored aspect. The ABSA problem can be divided into three basic tasks as follows: aspect detection, opinion tar- get expression (OTE), sentiment polarity. In this paper, we focus on detecting the opinions of users based on aspects and their sentiment, which we call span detection for ABSA. Specifically, when a re- view is given ”Although staffs are nice, the phone is terrible!”, the span detection for ABSA task aims to get two opinions ”staffs are nice” and ”the phone …

Content truncated to 500 words