Identifying the underlying sentiments, attitudes, and emotions in product reviews, and comments is a challenging but vital aspect of market research, brand monitoring, social media monitoring, and customer service. Here we have performed a comparative study of long short-term memory (LSTM), convolutional neural network (CNN), and combinations of CNN with LSTM (CNN-LSTM) and Gated Recurrent Unit (GRU), to recommend the most suitable deep learning architecture for the sentiment analysis of product reviews thereby revealing the emotional undertone of a group of words. Our study employed the TF-IDF (Term Frequency-Inverse Document Frequency) technique to compute a weight for each word, indicating its importance in the document and corpus. The results indicate that the CNN-LSTM cascaded approach significantly boosts the accuracy of sentiment analysis on the latest dataset with a promising accuracy of 98.3. This study can assist researchers in selecting an appropriate sentiment analysis model for their use case and in applying effective pre-processing techniques to enhance the accuracy of their model.