About this resource
<p>This repository contains the implementation of a <strong>Graph Convolutional Network (GCN)</strong> and <strong>Deep Reinforcement Learning (DRL)</strong> based recommender system designed to optimize long-term user engagement in video recommendation systems. The system leverages <strong>multimodal data</strong> (visual, textual, and acoustic features) and a <strong>multi-head attention mechanism</strong> to capture complex user-item interactions.</p> <p>The model has been evaluated on the <strong>MicroLens-100k</strong> and <strong>MicroLens-1M</strong> datasets, demonstrating superior performance compared to traditional recommendation methods like <strong>Matrix Factorization</strong> (MF), <strong>Collaborative Filtering</strong>, and <strong>GraphSAGE</strong>. Key features of the repository include:</p> <ul> <li><strong>Graph-based model</strong> for capturing user-item interactions using GCN.</li> <li><strong>Reinforcement learning</strong> for optimizing long-term user engagement using <strong>Q-learning</strong>.</li> <li><strong>Multimodal data integration</strong> for richer user and item representations.</li> <li><strong>Extensive evaluation</strong> using Precision@K, Recall@K, and NDCG@K metrics.</li> <li><strong>Ablation study</strong> to evaluate the impact of different model components and configurations.</li> <li><strong>Scalability tests</strong> to ensure efficient model performance on larger datasets.</li> <li><strong>Hyperparameter tuning</strong> to optimize model performance.</li> </ul> <p>This model can be applied to any recommendation task where long-term user engagement is a priority, such as video, music, or e-commerce product recommendations. The repository includes code for preprocessing data, training the model, and evaluating results.</p> <p><strong>Keywords</strong>: Recommender system, GCN, Graph Convolutional Network, DRL, Deep Reinforcement Learning, Precision@K, Recall@K, NDCG@K, long-term engagement, multimodal features, attention mechanism, scalability, ablation study.</p>
Files and access
Resources are read online through DARE. Direct file links are not published; access is checked against DSpace when a document is opened. Authorised librarian workflows are unaffected.
This record contains bibliographic metadata, but no files are currently attached.
