In this paper. we explore the approaches to the problem of cross-domain few-shot classification of sentiment aspects. By cross-domain few-shot. we mean a setting where the model is trained on large data in one domain (for example. https://www.markbroyard.com/hot-save-turbo-air-m3f24-1-n-29-wide-m3-series-reach-in-freezer-featuring-one-solid-door-energy-star-on-sale-top-super/