Sentiment Analysis Using a Hybrid Model Using Machine Learning

In the contemporary landscape, user feedback on social media platforms has become ubiquitous, driving the exploration of sentiment mining and analysis in the realm of machine learning research. Recent studies have shown a profound reliance on unimodal and multimodal approaches within machine learning algorithms, particularly in the context of sentiment analysis. Moreover, the evolution of deep learning models, particularly in image and text processing, has significantly advanced the field of sentiment analysis and classification. This paper introduces a hybrid deep learning framework comprising Convolutional Neural Network (CNN), K-Nearest Neighbours (KNN), stacked over Long Short-Term Memory (LSTM). Additionally, it proposes a hybrid CNN + LSTM model utilizing pre-trained vectors for enhanced performance. This study explores different hyperparameters and optimization strategies to address overfitting in the model, aiming to attain peak performance while mitigating the impact of overfitting it has been validated on Stanford Sentiment Treebank2 Dataset (SST2).