Simple Approach for Aspect Sentiment Triplet Extraction Using Span-Based Segment Tagging and Dual Extractors

Aspect sentiment triplet extraction (ASTE) is a task which extracts aspect terms, opinion terms, and sentiment polarities as triplets from review sentences. Existing approaches have developed bidirectional structures for term interaction. Sentiment polarities are subsequently extracted from aspect-opinion pairs. These solutions suffer from: 1) high dependency on custom bidirectional structures, 2) inadequate representation of the information through existing tagging schemes, and 3) insufficient usage of all available sentiment data. To address the above issues, we propose a simple span-based solution named SimSTAR with Segment Tagging And dual extRactors. SimSTAR does not introduce any additional bidirectional mechanism. The segment tagging scheme is capable to indicate all possible cases of spans and reveals more information through negative labels. Dual extractors are employed to make the sentiment extraction independent of the term extraction. We evaluate our model on four ASTE datasets. The experimental results show that our simple method achieves state-of-the-art performance.