Gauging the quality of a video or post by inspecting the user comments is a challenging task, as it requires reading through the comments. Over the years, researchers have employed various models to automatically identify the comments' polarity automatically. This paper explores how deep-learning models such as convolutional neural networks (CNNs) and long-short term memory (LSTM) models can perform polarity analysis on a collection of comments and how the results can be used to assign polarity to user comments automatically. To accomplish this goal, we utilize three labeled datasets: the Apple Twitter Sentiment Text dataset, the Airline Sentiment dataset, and a general Twitter dataset that contains positive, negative, and neutral sentiments. We develop a hybrid CNN-BiLSTM-based prediction model for polarity prediction and assess its performance using accuracy, precision, recall, and Fl-score measures. Additionally, we employ three feature extraction techniques (TF-IDF, bag-of-words, and Word2vec) and five machine learning models (K-nearest neighbors, support vector machine, naive Bayes, random forest, and XGBoost) to compare the performance of the CNN-BiLSTM model with these models. The results showed that the developed models achieved the best performance with the accuracy, precision, recall, and Fl-score values of 0.931, 0.939, 0.925, and 0.931, respectively. Among the used ML models, Random Forest with TF-IDF produced the best performance with the accuracy, precision, recall, and Fl-score values of 0.820, 0.830, 0.831, and 0.829, respectively. Other models performed moderately.