A Semantic-Context Embedding Enhanced Attention Fusion BiLSTM: Unraveling Multilingual Sentiments in Product Reviews with Advanced Deep Learning

Abstract

For natural language processing (NLP), sentiment analysis (SA) is crucial since it helps to understand users' feelings and opinions in a variety of contexts. Even though Deep Learning (DL) techniques are becoming more popular in SA, effective model optimization can be difficult because they frequently call for a great deal of hyperparameter tuning. However, because current models can't sufficiently capture the variety of review contexts, it introduces bias and inaccuracies, especially in product reviews.  For Multilingual Sentiment Analysis (MSA) in product reviews, this research proposed a Semantic-Context Embedding Enhanced Attention Fusion BiLSTM (SCEEAF-BiLSTM). The proposed model combines Continuous Bag-of-Word (CBoW) and Skipgram techniques to extract semantic context after the preprocessing stages of tokenization, stop word removal, and case normalization. A novel Convolutional BiLSTM with Enhanced Attention (CoBLEA) architecture is introduced for multilingual sentiment prediction to extract comprehensive context representations. The model ultimately shows efficacy in dividing multilingual sentiments into positive, neutral, and negative states, providing a viable method for complex SA in several circumstances. The outcome signifies that the proposed approach obtains a high accuracy attained 0.987, precision attained 0.985, recall attained 0.978 and F1-Score attained 0.986 when compared with prior works. With practical applications in sentiment-driven platforms operating in multiple languages, the research presents a method for complex SA in e-commerce, social media, and customer feedback systems. It also emphasizes the significance of comprehending multilingual opinions for enhancing marketing strategies, driving business decisions, and improving customer satisfaction.