Convolutional Neural Network based sentiment analysis with TF-IDF based vectorization

Abstract

Recent years have seen an increase in interest in sentiment analysis(SA), which is the process of determining if a textual item communicates a favorable or negative opinion about a certain entity such as a business, people, or government. It plays a crucial part in NLP. The growth of user-generated material in recent years, such as traveler reviews, has resulted in a significant volume of unstructured data that is challenging to extract usable information from. Predicting the precise sentiment polarity of the customer reviews, user ratings, recommendations etc. is still a difficult problem, particularly for fine-grained sentiment categorization, because of the variations in length of the sequence, texts ordering, and complex logics. In this research, first propose sentiment analysis using deep learning, a unique approach compared with other existing techniques which makes the input data sample of a constant sizing and enhances the percentage of sentiments data in each review. Research present CNN-TDIDF family models, which combine CNN and TF-IDF in parallel and are based on deep learning and are inspired by the most current research on neural networks. Experiments on a number of difficult datasets, show that the suggested strategy performs better than many standard approaches. This system has been shown to operate better than conventional machine learning methods and reach 87 percent accuracy rate. Comparing this work to past efforts, we are also able to attain a very high accuracy rate.

URN:NBN:sciencein.jist.2023.v11.503