Learning Recommendation Algorithm Based on Improved BP Neural Network in Music Marketing Strategy.

Area of Science:

• Computer Science
• Music Information Retrieval

Background:

• Streaming music platforms have transformed music consumption, necessitating effective recommendation systems.
• Data sparsity in user behavior poses a significant challenge for accurately mining music preferences.

Purpose of the Study:

• To propose a metric ranking learning recommendation algorithm with fused content representation to address data sparsity.
• To enhance the mining of user music preference features in digital music marketing.

Main Methods:

• Constructing relative partial order relations using observed and unobserved behavioral data.
• Employing audio feature extraction submodels to alleviate data sparsity.
• Utilizing metric learning to mine user-song preference relationships.
• Applying convolutional neural networks for high-level song semantic feature extraction.
• Developing a bidirectional recurrent neural network model with an attention mechanism for session-based recommendations.

Main Results:

• The proposed algorithm effectively mines user music preferences despite data sparsity.
• Fused content representation improves the accuracy of music recommendations.
• The attention-based bidirectional recurrent neural network model reduces noise and captures song dependencies.

Conclusions:

• The metric ranking learning algorithm with fused content representation offers a robust solution for music recommendation systems facing data sparsity.
• This approach enhances user experience by providing more accurate and relevant music suggestions.
• The integration of deep learning techniques, including CNNs and RNNs with attention, significantly improves the performance of music preference mining.