An Effective Sentiment Analysis System using Hybrid Approach of CNN and Machine Learning

This research introduces a hybrid approach to sentiment analysis with the integration of Convolutional Neural Networks (CNNs) with traditional machine learning techniques. We aim to take advantage of CNNs' deep learning capabilities for extracting features and use traditional algorithms for classification. Our integrated model identifies sentiment polarity across different types of textual data. The model demonstrates increased performance by incorporating pre-trained word embeddings and adopting a multi-channel architecture. We aim to emphasize the efficiency of CNNs in text classification and the crucial role of pre-trained word vectors in deep learning methodologies for Natural Language Processing (NLP). Our results revealed that SVC was the highest-performing classifier compared to logistic regression and Random Forest, achieving an accuracy of 75.45% when used with the Multichannel CNN, indicating opportunities for further refinement and exploration of ensemble methods to boost classification accuracy.