Enhancing Implicit Sentiment Learning via the Incorporation of Part-of-Speech for Aspect-based Sentiment Analysis

Enhancing Implicit Sentiment Learning via the Incorporation of Part-of-Speech for Aspect-based Sentiment Analysis Junlang Wang, Xia Li*, Junyi He, Yongqiang Zheng and Junteng Ma School of Information Science and Technology, Guangdong University of Foreign Studies, Guangzhou, China {junlangwang,xiali}@gdufs.edu.cn Abstract Implicit sentiment modeling in aspect-based sentiment analysis is a challenging problem due to complex expressions and the lack of opinion words in sentences. Recent efforts focusing on implicit sentiment in ABSA mostly leverage the dependency between aspects and pretrain on extra annotated corpora. We argue that linguistic knowledge can be incorporated into the model to better learn implicit sentiment knowledge. In this paper, we propose a PLM-based, linguis- tically enhanced framework by incorporating Part-of-Speech (POS) for aspect-based sentiment analysis. Specifically, we design an input template for PLMs that focuses on both aspect-related contextualized features and POS-based linguistic features. By aligning with the representations of the tokens and their POS sequences, the introduced knowledge is expected to guide the model in learning implicit sentiment by capturing sentiment-related information. Moreover, we also design an aspect-specific self-supervised contrastive learning strategy to optimize aspect-based contextualized representation construction and assist PLMs in concentrating on target aspects. Experimental results on public benchmarks show that our model can achieve competitive and state-of-the-art performance without introducing extra annotated corpora. 1 Introduction Aspect-based Sentiment Analysis (ABSA) aims to identify the sentiment polarities towards specific as- pects in sentences. For example, in the sentence “The dessert is incredible but the service is terrible,” the sentiment polarities towards the aspects “dessert” and “service” are positive and negative respectively. Previous work on aspect-based sentiment analysis has focused on explicit sentiment expression for specific aspect terms. It means that the sentiment polarities towards the aspects can be explicitly revealed by opinion words. e.g., the sentence “The dessert is incredible” contains the opinion word “incredible” which carries the positive sentiment towards the corresponding aspect “dessert”. Many studies have been proposed and achieved promising results towards this task, such as attention mechanism-based methods (Ma et al., 2017; Huang et al., 2018; Zhang et al., 2019a; Wu et al., 2022), graph neural network-based methods (Zhang et al., 2019b; Wang et al., 2020; Liang et al., 2022; Zheng et al., 2023), and pre-trained language model-based methods (Song et al., 2019; Phan and Ogunbona, 2020; Dai et al., 2021; Cao et al., 2022). However, due to the diversity and flexibility of natural language, sentences containing implicit sen- timent expressions are common in human speech. For implicit sentiment, we refer to the recognition of subjective textual units where no polarity markers, opinion words or obvious descriptions are present but people are still able to state whether the text portion under analysis expresses the sentiment (Russo et al., 2015). As shown in Table 1, the four sentences can clearly express the sentiment without any opinion words. Taking the second sentence as an example, no opinion words can be found to determine the sentiment polarities towards the aspects “food”, but people can still recognize that its polarity is neg- ative. Additionally, we …

Content truncated to 500 words