New Approach Based on Word Embedding and Deep Learning Algorithms to Optimize the Sentiment Analysis Performance in Social Business Intelligence

Sentiment analysis is an essential component of Natural Language Processing (NLP) that involves analysing the polarity of sentences in a corpus. This paper aims to present a new approach of sentiments analysis. Unlike traditional feature extraction methods such as TFIDF and bag of words. The proposed method is based on word embeddings which preserve the semantic relationships between words. Additionally, the power of deep learning algorithms is more robustness in handling complex tasks, highlighting the importance of finding the optimal combination of deep learning algorithms and word embedding models that improve the accuracy of sentiment analysis models. In this study, we present various word embedding models, specifically Word2Vec, GloVe, and FastText using with an ensemble of deep learning algorithms including Recurrent Neural Networks (RNN), Gated Recurrent Units (GRU), Long Short-Term Memory networks (LSTM), Bidirectional LSTM (BiLSTM), and Convolutional Neural Networks (CNN). These models are applied on an English dataset of hotel reviews. Moreover, we explore hybrid approaches that combine different architectures, such as CNN with BiLSTM, CNN with LSTM, RNN with LSTM and RNN with BILSTM to determine the optimal configuration of these models. Our approach spans various hyperparameter settings to carefully evaluate the performance of the model configuration. According to the best of the authors' knowledge, the method, using various combinations of word embedding models and deep learning algorithms with different hyperparameters in the context of sentiment analysis, represents a new approach over exiting methodologies.