PASTE: A Tagging-Free Decoding Framework Using Pointer Networks for Aspect Sentiment Triplet Extraction

Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 9279–9291 November 7–11, 2021. c⃝2021 Association for Computational Linguistics 9279 PASTE: A Tagging-Free Decoding Framework Using Pointer Networks for Aspect Sentiment Triplet Extraction Rajdeep Mukherjee∗, Tapas Nayak∗, Yash Butala, Sourangshu Bhattacharya and Pawan Goyal Department of Computer Science and Engineering, IIT Kharagpur, India {rajdeep1989, yashbutala}@iitkgp.ac.in, tnk02.05@gmail.com, {sourangshu, pawang}@cse.iitkgp.ac.in Abstract Aspect Sentiment Triplet Extraction (ASTE) deals with extracting opinion triplets, consist- ing of an opinion target or aspect, its associ- ated sentiment, and the corresponding opinion term/span explaining the rationale behind the sentiment. Existing research efforts are ma- jorly tagging-based. Among the methods tak- ing a sequence tagging approach, some fail to capture the strong interdependence between the three opinion factors, whereas others fall short of identifying triplets with overlapping aspect/opinion spans. A recent grid tagging approach on the other hand fails to capture the span-level semantics while predicting the sen- timent between an aspect-opinion pair. Differ- ent from these, we present a tagging-free so- lution for the task, while addressing the lim- itations of the existing works. We adapt an encoder-decoder architecture with a Pointer Network-based decoding framework that gen- erates an entire opinion triplet at each time step thereby making our solution end-to-end. Interactions between the aspects and opinions are effectively captured by the decoder by con- sidering their entire detected spans while pre- dicting their connecting sentiment. Extensive experiments on several benchmark datasets es- tablish the better efficacy of our proposed ap- proach, especially in recall, and in predict- ing multiple and aspect/opinion-overlapped triplets from the same review sentence. We re- port our results both with and without BERT and also demonstrate the utility of domain- specific BERT post-training for the task. 1 Introduction Aspect-based Sentiment Analysis (ABSA) is a broad umbrella of several fine-grained sentiment analysis tasks, and has been extensively studied since its humble beginning in SemEval 2014 (Pon- tiki et al., 2014a). Overall, the task revolves around ∗Equal contribution Sent 1: The film was good , but could have been better . Triplets [Aspect ; Opinion ; Sentiment] (1) film ; good ; positive (2) film ; could have been better ; negative Sent 2: The weather was gloomy , but the food was tasty . Triplets (1) weather ; gloomy ; negative (2) food ; tasty ; positive Table 1: Examples of Aspect-Opinion-Sentiment triplets (opinion triplets) present in review sentences. automatically extracting the opinion targets or as- pects being discussed in review sentences, along with the sentiments expressed towards them. Early efforts on Aspect-level Sentiment Classification (Tay et al., 2018; Li et al., 2018a; Xue and Li, 2018) focus on predicting the sentiment polarities for given aspects. However, in a real-world sce- nario, aspects may not be known a-priori. Works on End-to-End ABSA (Li et al., 2019; He et al., 2019; Chen and Qian, 2020) thus focus on extracting the aspects as well as the corresponding sentiments si- multaneously. These methods do not however cap- ture the reasons behind the expressed …

Content truncated to 500 words