Emotional Analysis Based On LSTM-CNN Hybrid Neural Network Model

In the realm of natural language processing (NLP), text classification is an important task. Current techniques for text classification tasks still need development, however, because of the complex abstraction of text semantic information and the considerable impact of context. This research synthesizes the strengths of the Bidirectional Long Short-Term Memory (Bi-LSTM) and the Convolutional Neural Network (CNN) models of neural networks. The LSTM-CNN hybrid model is a combination of the two models: the LSTM-generated text feature vector is then extracted using the CNN structure. The Bi-LSTM can retain long text sequences while taking into account the importance of the whole text, and it can subsequently use the organizational principles of the CNN to extract local text features. The research's experiment compares the effectiveness of the LSTM-CNN, the Bi-LSTM, and the Random Forest (RF) in a sentiment analysis project based on the ternary classification (Neutral, Positive, and Negative) of the metaverse, which can be described as a virtual reality space where individuals can interact with a computer-generated environment and other users in real-time, related topics. From what can be gathered from this research experiments, the LSTM-CNN outperforms other models when it comes to boosting text classification accuracy and achieving lower loss.