LEGO-ABSA: A Prompt-based Task Assemblable Unified Generative Framework for Multi-task Aspect-based Sentiment Analysis

Proceedings of the 29th International Conference on Computational Linguistics, pages 7002–7012 October 12–17, 2022. 7002 LEGO-ABSA: A Prompt-based Task Assemblable Unified Generative Framework for Multi-task Aspect-based Sentiment Analysis Tianhao Gao1,∗, Jun Fang2,∗, HanyuLiu2∗, Zhiyuan Liu2 Chao Liu2†, Pengzhang Liu2, Yongjun Bao2, Weipeng Yan2 1 School of Software and Microelectronics, Peking University 2 JD, Retail, Beijing, China gaotianhao@pku.edu.cn {junfang8, liuhanyu11, liuzhiyuan8, liuchao397, liupengzhang, baoyongjun, Paul.yan}@jd.com Abstract Aspect-based sentiment analysis (ABSA) has received increasing attention recently. ABSA can be divided into multiple tasks according to the different extracted elements. Existing generative methods usually treat the output as a whole string rather than the combination of different elements and only focus on a single task at once. This paper proposes a unified generative multi-task framework that can solve multiple ABSA tasks by controlling the type of task prompts consisting of multiple element prompts. Further, the proposed approach can train on simple tasks and transfer to difficult tasks by assembling task prompts, like assem- bling Lego bricks. We conduct experiments on six ABSA tasks across multiple benchmarks. Our proposed multi-task approach achieves new state-of-the-art results in almost all tasks and competitive results in task transfer scenar- ios. 1 Introduction ABSA is a fine-grained sentiment analysis task that has attracted increasing attention in recent years(Schouten and Frasincar, 2016; Nazir et al., 2020). ABSA aims to extract different elements in- cluding: 1) the aspect term(a); 2) opinion term(o); 3) the aspect category(c) corresponding to the as- pect term; 4) the sentiment polarity(s) for a specific aspect term. For example, in the sentence “Pizza is delicious”, “Pizza” is an aspect term belonging to the food category, and the corresponding opinion term is “delicious”, which expresses positive sen- timent. As shown in Table 1, based on the combi- nation of different elements to be extracted, ABSA can be divided into multiple tasks. This paper explores tasks containing two or more elements. In general, most ABSA tasks are trans- ferred to classification tasks. Previous works often *These authors have equal contribution. Work done during gaotianhao’s internship at JD, Retail, Beijing, China †Corresponding authors. designed a new architecture carefully and trained with the corresponding dataset for a specific sub- task. We review them as follows: Pair Extraction The pair extraction task in- cludes AOPE(Aspect-Opinion Pair Extraction), ACSA(Aspect-Category Sentiment Analysis) and E2E-ABSA(End-to-End Aspect-based Sentiment Analysis) in our method. ACSA is usually treated as a multi-task classification task (Hu et al., 2018; Dai et al., 2020; Ma et al., 2018) Some works convert the AOPE and E2E-ABSA tasks into se- quence tagging problems(Wu et al., 2020b; Gao et al., 2021; Chen et al., 2020; He et al., 2019), specifically using the BIO tagging strategies (Wang and Pan, 2018; Wu et al., 2020a; Li et al., 2019a,b) The pair extraction is also named the basic task in this paper. Triplets Extraction The triples extraction tasks contain ASTE(Aspect Sentiment Triplet Extrac- tion) and TASD(Target Aspect Sentiment Detec- tion) in our paper. Most previous works still treat them as a sequence tagging task (Xu et …

Content truncated to 500 words