Reinforced Transformer with Cross-Lingual Distillation for Cross-Lingual Aspect Sentiment Classification

Though great progress has been made in the Aspect-Based Sentiment Analysis(ABSA) task through research, most of previous work focuses on English-based ABSA problems, and there are few efforts other languages mainly due to lack training data. In this paper, we propose an approach for performing a Cross-Lingual Aspect Classification (CLASC) which leverages rich resources one language (source language) aspect sentiment classification under-resourced (target language). Specifically, first build bilingual lexicon domain-specific data translate category annotated source-language corpus then sentences from source target via Machine Translation (MT) tools. However, MT systems general-purpose, it non-avoidably introduces translation ambiguities would degrade performance CLASC. context, novel called Reinforced Transformer with Distillation (RTCLD) combined target-sensitive adversarial learning minimize undesirable effects sentence translation. We conduct experiments different combinations, treating English as Chinese, Russian, Spanish languages. The experimental results show that our proposed outperforms state-of-the-art methods