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

Aspect Sentiment Triplet Extraction (ASTE) deals with extracting opinion\ntriplets, consisting of an opinion target or aspect, its associated sentiment,\nand the corresponding opinion term/span explaining the rationale behind the\nsentiment. Existing research efforts are majorly tagging-based. Among the\nmethods taking a sequence tagging approach, some fail to capture the strong\ninterdependence between the three opinion factors, whereas others fall short of\nidentifying triplets with overlapping aspect/opinion spans. A recent grid\ntagging approach on the other hand fails to capture the span-level semantics\nwhile predicting the sentiment between an aspect-opinion pair. Different from\nthese, we present a tagging-free solution for the task, while addressing the\nlimitations of the existing works. We adapt an encoder-decoder architecture\nwith a Pointer Network-based decoding framework that generates an entire\nopinion triplet at each time step thereby making our solution end-to-end.\nInteractions between the aspects and opinions are effectively captured by the\ndecoder by considering their entire detected spans while predicting their\nconnecting sentiment. Extensive experiments on several benchmark datasets\nestablish the better efficacy of our proposed approach, especially in the\nrecall, and in predicting multiple and aspect/opinion-overlapped triplets from\nthe same review sentence. We report our results both with and without BERT and\nalso demonstrate the utility of domain-specific BERT post-training for the\ntask.\n