WKE: Word-Level Knowledge Enrichment for Aspect Term Extraction

We tackle Aspect Term Extraction (ATE), a task of automatically extracting aspect terms. Supervised learning for ATE suffers from scarce annotated data. This causes the difficulty in recognizing unseen or less-labeled samples during testing. To address the problem, we propose a Word-level Knowledge Enhancement (WKE) approach. WKE detects long-tail words which occur in the training dataset with a lower frequency. On this basis, it acquires definitions of long-tail words from a knowledge base Wiktionary, and uses the definitions as knowledge to expand the representations of long-tail words in the encoding channel. This further enables the decoding under a knowledge-aware condition. We experiment on the benchmark corpora of SemEval semantic evaluation, including R14-16 and L14 towards the domains of Restaurant and Laptop. The test results show that our approach yields substantial improvements compared to the baselines BERT \(_{base}\) and BERT-PT, and achieves comparable performance to the state-of-the-art models which use data augmentation.