Product evaluations hold significant importance for prospective consumers in guiding their purchasing choices. A number of opinion mining techniques have been proposed in this field, and one of the main issues is determining the sentiment direction of the overview phrase. Recently, deep learning has surfaced as a highly effective approach for addressing sentiment classification issues. An effective representation is automatically learned by a neural network without the need for people to participate. Nonetheless, the effectiveness of deep learning is largely contingent upon the presence of extensive training datasets. We provide a novel deep learning system for sentiment classification of product reviews based on easily accessible ratings acting as weak supervision signals. The structure consists of two stages: first, using rating data, obtain a basic model (an embedding space) that captures the overall sentiment pattern of phrases; second, add a classification layer on top of the embedding layer and use labeled sentences for supervised fine-tuning. For simulating review phrases, we look into two different minimal networking frameworks: long short-term memory networks and multi-layer pattern extractors. We assembled a dataset from Amazon that included over a million poorly tagged comment statements and over 12,000 tagged assessment statements in order to evaluate the effectiveness of the suggested framework. The results of the experiments show that the suggested framework is both preferable to baseline models and effective.