Sentiment Classification Method of Music Lyrics Based on Att-BiLSTM-XGBoost Hybrid Model

Sentiment classification is a hot topic in the field of natural language processing, music is a way to express people's emotions. At present, in the research of music emotion classification, scholars mainly study audio classification, but it is difficult to extract semantic-related features from audio, music lyrics are an important carrier for conveying emotions, so this paper proposes an Att-BiLSTM-XGBoost hybrid model based on the two-way long short-term memory network (BiLSTM) based on the attention mechanism and the XGBoost algorithm, and classifies the emotions of Chinese music lyrics based on the model. The model extracts deep semantic features from the experimental dataset of music lyrics through the Att-BiLSTM feature extraction module, and then input s them into the XGBoost classifier for sentiment preference classification. Experimental verification is carried out on the crawled Chinese lyrics dataset, and the verification results show that the mixed model proposed in this paper can achieve a good classification effect in emotion classification. By observing the performance evaluation indexes (P, R, F1) of each category (happy, sad, healing, calm), we can find that a good classification effect is achieved, which proves the validity of the model in this paper.