An Intelligent Approach Toward Lyrics Text Classification Using Multilevel Cross Attention‐Based Adaptive <scp>BiLSTM</scp> With Relevant Feature Extraction

ABSTRACT In recent years, much literature has determined the classification of music in order to classify lyrics that are inappropriate for young people. Many studies utilize different types of classifiers for analyzing the metadata and music lyrics that have failed to achieve the lyrics classification. In the music information retrieval (MIR) community, lyrics‐based classification for music genre is still insufficiently examined. The earlier techniques have considered the lyrics only in the English language. Hence, it is important to replace the standard text classification approach with other advanced methods to classify the different language lyrics effectively. To solve these issues, an efficient deep learning‐based lyrics text classification is developed in this work to classify the lyrics text. Initially, the required text data are collected from the benchmark dataset. Subsequently, the collected text data are fed into the preprocessing performance to enhance the text data quality. After that, the preprocessed text data are fed into the features extraction process, where the features are retrieved using GloVe embedding, term frequency inverse document frequency (TF‐IDF), and bidirectional encoder representations from transformers (BERT). The extracted features are passed to the multilevel cross attention‐based adaptive bidirectional long short‐term memory (MCABi‐LSTM) for recognizing the lyrics text. The text classification framework of the MCABi‐LSTM is enhanced through fine‐tuning the attributes using the modified ring toss game optimization (MRTGO). The classified outcome from the MCABi‐LSTM is used for examining the musical patterns. Finally, the experimental validation is conducted on the developed lyric text classification model to prove its effectiveness.