Serial cascaded hybrid adaptive deep networks-based lyrics text classification using optimization approach.

Area of Science:

• Artificial Intelligence
• Music Information Retrieval
• Natural Language Processing

Background:

• The digital music shift increased song variety, necessitating content analysis for discovery and age-appropriateness.
• Emotional content analysis is a key method for music discovery, but classifying lyrics remains challenging.
• Deep learning shows promise in Natural Language Processing (NLP), yet its application to filtering inappropriate music lyrics is limited.

Purpose of the Study:

• To propose a deep learning-based lyrics text classification process for identifying and filtering inappropriate music.
• To develop and evaluate a novel deep learning model for classifying song lyrics based on their content.

Main Methods:

• Text data was pre-processed and fed into the Serial Cascaded Hybrid Adaptive Deep Networks (SCHADNet) model.
• SCHADNet integrates Transformer-based Bidirectional Long Short-Term Memory (Trans-BiLSTM) with Gated Recurrent Unit (GRU).
• Model parameters were optimized using the Improved Marine Predators Algorithm (IMPA).

Main Results:

• The SCHADNet model achieved high classification performance.
• Accuracy rate reached 93.4%, with a recall of 93.47% and Negative Predictive Value (NPV) of 99.2%.
• Numerical analysis demonstrated the model's superiority over classical text classification techniques.

Conclusions:

• The proposed deep learning model significantly enhances lyrics text classification.
• SCHADNet offers an effective solution for filtering inappropriate music content.
• The model's high performance indicates its potential for improving music recommendation and content moderation systems.