For many business applications, we often seek to analyze sentiments\nassociated with any arbitrary aspects of commercial products, despite having a\nvery limited amount of labels or even without any labels at all. However,\nexisting aspect target sentiment classification (ATSC) models are not trainable\nif annotated datasets are not available. Even with labeled data, they fall\nshort of reaching satisfactory performance. To address this, we propose simple\napproaches that better solve ATSC with natural language prompts, enabling the\ntask under zero-shot cases and enhancing supervised settings, especially for\nfew-shot cases. Under the few-shot setting for SemEval 2014 Task 4 laptop\ndomain, our method of reformulating ATSC as an NLI task outperforms supervised\nSOTA approaches by up to 24.13 accuracy points and 33.14 macro F1 points.\nMoreover, we demonstrate that our prompts could handle implicitly stated\naspects as well: our models reach about 77% accuracy on detecting sentiments\nfor aspect categories (e.g., food), which do not necessarily appear within the\ntext, even though we trained the models only with explicitly mentioned aspect\nterms (e.g., fajitas) from just 16 reviews - while the accuracy of the\nno-prompt baseline is only around 65%.\n