A Bert Model with Deep Learning Approach in Natural Language Processing (NLP)

Sentiment analysis is a branch of natural language processing (NLP). It entails determining the polarity of a corpus, typically a corpus of writing. The most widely used text categorization technology is sentiment analysis, which evaluates an incoming message to determine whether the overall sentiment is positive, negative, or neutral. Sentiment analysis may assist in learning more about how people perceive a product in general. The current experiment is the first step in the investigation into the performance of three models: CNN, LSTM, and Bert along with two feature extraction techniques such as bigrams and word embeddings. The study starts with an experiment on the performance of the Bert mul tilingual pretrained model. First, a preprocess work is done on the data by removing any artifacts (words, symbols, numbers, etc.) that could interfere with model training. Then an extraction of features using bigram and word embeddings is done, the latter using Glove. Finally, the model was trained based on its configurations. On the Amazon Polarity Review dataset, the model performed the best, with an accuracy of 90.25%. This current study provides an overview of Bert model performance on sentiment analysis. In the following experiment, a comparative study will be conducted of all three of the most commonly used sentiment analysis models in order to refresh the literature and broaden our understanding of their performance in natural language processing sentiment analysis.