Aspect-Based Sentiment Analysis for Virtual Reality Roaming System

Through sentiment analysis technology, NPCs (Non-Player-Controlled Character) in virtual reality roaming system are able to recognize the user's emotional state and react accordingly. In aspect-based sentiment analysis, NLP techniques often use manually constructed features combined with machine learning models for classification. However, these methods are complex in feature engineering and lack generalisation capabilities. At the same time, traditional models often ignore the importance of local context for the correct classification of the sentiment polarity of aspect term, and process all input data indiscriminately. In order to solve the above problems and improve the processing efficiency of aspect-based sentiment analysis, an attention weight decay mechanism is proposed. On this basis, a multi-module model is constructed by fusing BERT, BiLSTM and TextCNN, and the performance of the model is verified experimentally. This model can be better applied to virtual reality roaming systems.