With the rapid growth of the Internet advertising market, accurate advertisement recommendations have become crucial. Effectively capturing the interaction between user and advertisement features is key to improving the prediction of click-through rate (CTR) and conversion rate (CVR). However, existing CTR and CVR prediction models suffer from feature dependency bias and insufficient semantic information extraction from ads. To address these issues, this study proposes a knowledge graph-enhanced advertisement recommendation algorithm (KGEARA). The algorithm constructs a knowledge graph by converting structured data into triplets, effectively integrating advertisement feature information and capturing relationships between ads. Through knowledge graph representation learning, these features are transformed into embeddings that merge the semantic characteristics of ads and capture interaction details. Further, by combining advertisement feature embeddings with other feature embeddings, CTR and CVR are predicted through expert networks, gated networks, and task-specific towers. In addition, inverse propensity scoring (IPS) is introduced to address click-bias issues and correct prediction biases. Extensive experiments on real-world advertisement datasets demonstrate the effectiveness of the proposed model in improving CTR and CVR prediction accuracy.