Adaptive Content Recommendation Framework Using ChatGPT and Self-Coding Neural Architectures
Kun Peng · 2025
Recommendation systems, especially multi-modal content recommendation, are important application scenarios for machine learning algorithms. Therefore, this study proposes an innovative content recommendation framework based on the ChatGPT and auto-encoder neural network. Overall, the designed framework combines the powerful natural language processing capabilities of ChatGPT. At the same time, it integrates the nonlinear feature extraction capabilities of the auto-encoder neural network. The framework achieves efficient content recommendation through user portrait construction, collaborative filtering and content filtering analysis. The specific methods of this framework include: first, the framework uses the BERT model to embed user comments and extract vector representations of key comments; then, it extracts user and product features through the bidirectional gated recurrent unit (BI-GRU), and weights the features in combination with the attention mechanism; finally, the framework optimizes the matrix decomposition loss function through the auto-encoder neural network to further improve the recommendation accuracy. The experiments are conducted based on a real dataset from the YouTube platform. The results show that the framework performs well in both precision and recall.