Feature Extraction Techniques in Sentiment Analysis Using Deep Learning: A Review

Text mining and information retrieval are fundamentally dependent on the selection of text feature items. Features must be created manually for use in traditional feature extraction methods. While it may take a lot of effort to manually build an effective feature, deep learning makes it possible to automatically learn new representations of successful features from training data in the context of new applications. Text mining has made a lot of progress thanks to deep learning, which is a new way to find features. Deep learning is different from more traditional methods because it automatically learns features from large datasets, while the latter adopts handcrafted features, which rely heavily on the prior knowledge of designers and thus make it nearly impossible to take advantage of large datasets. Feature representation in deep learning may be automatically taught using big data, which typically consists of millions of parameters. This thesis starts with a brief overview of the most common ways to extract text features. It then talks about the widespread use of deep learning methods in text feature extraction and their applications. Finally, it makes some predictions about the future of deep learning's role in feature extraction.