Sentiment analysis based on deep features through textual tweets classification using optimal hierarchical CLSTM network model
摘要
Opinion mining has gradually become tough to handle as the amount of user-generated content on social media platforms increases exponentially. As a matter of fact, Twitter is the most common platform to collect voices in terms of products, advancements, and policies. Sentiment Analysis (SA) deals with people's thoughts, feelings, and opinions about various subjects. Through the examination of tweets, one can gauge public views on news, rules, the community, and even celebrities. Unfortunately, currently existing SA mechanisms often have limited prediction capabilities and are still a long way from being able to function in real, time commercial applications. The main causes of inaccuracies are lack of data and difficulties in model configuration in deep learning (DL). This research introduces a classification learning-based Optimal Tiered blocks of Convolutional Neural Long Short, Term Memory (OTCNLSTM) for emotion recognition. Four Local Features Training Blocks (LFTBs), which are capable of hierarchically extracting spatiotemporal local emotional cues, make up the OTCNLSTM model. On top of that, the Boosted Killer Whale Predation Optimization (BKWOP) technique is introduced to pinpoint the perfect hyperparameters and solution sets, thus forming a stable neural network model. The newly designed system efficiently categorizes the sentiments generated from the Twitter users' comments into four categories, namely, positive, negative, neutral, and irrelevant. The Kaggle Twitter dataset powered a thorough experimental study, which showed the model's performance like a rock. The OTCNLSTM model reached an overall accuracy.