Knowledge-Enhanced Contextual Auction Design Based on Deep Neural Networks

The optimal auction represents a fundamental challenge in auction design, primarily focused on maximizing the seller’s revenue. Recently, context-aware auction design has gained traction as it enhances the model’s understanding of the auction environment by incorporating contextual information about buyers and items. However, limited contextual information can hinder the model’s ability to capture potential correlations. Therefore, we propose a Knowledge-Enhanced Contextual Auction Design Model (KECA) based on neural networks. KECA aims to improve the comprehensive understanding of the auction environment by integrating external knowledge. Specifically, we first construct a structured knowledge graph of triples using Wikipedia. This knowledge graph is utilized to enhance the contextual feature representation of buyers and items, thereby providing a more comprehensive description of the information in the auction environment. Additionally, we employ the K-means clustering algorithm to enhance feature fusion between auction information and the knowledge graph within the locally sensitive hashing self-attention mechanism, thereby improving the model’s feature representation capabilities. Finally, extensive experimental results indicate that KECA outperforms current mainstream auction models under various settings, verifying the effectiveness of this approach.