Incorporating Knowledge Graph Embeddings into Graph Neural Networks for Sequential Recommender Systems

Traditional recommendation methods often struggle to capture the evolving nature of user interactions and the rich contextual information available in external data sources. This paper presents an innovative approach to sequential recommendations by integrating knowledge graph embeddings into a graph neural network (GNN) architecture. We utilize the MovieLens dataset as our primary data source and enrich it with additional contextual information from Wikidata, an open-source comprehensive knowledge base by incorporating the translation-based knowledge graph embeddings. Experimental results demonstrate that our approach improves recommendation accuracy, particularly in terms of Hit@K metrics, by leveraging the enriched movie embeddings. This study highlights the potential of utilizing open-source knowledge bases to enhance the predictive capabilities of recommendation systems, providing a more nuanced and comprehensive understanding of user preferences.