NIGAN: Neighbor Interaction Based Graph Attention Network for Click - Through Rate Prediction

CTR (Click-Through Rate refers) to the probability of users clicking on advertisements or products in an online rec-ommendation system. CTR prediction is a crucial task in online recommendation systems. In recent years, the application of graph neural networks in CTR prediction has been a highly active re-search area. Graph neural networks have the ability to capture higher-order relationships between entities and their neighbors, forming representations of entities by aggregating neighbor infor-mation. However, these networks often do not pay sufficient attention to the interaction among different neighbor features of enti-ties. In this paper, we propose a click-through rate prediction model called NIGAN, which stands for Neighbor Interaction Based Graph Attention Network. In this paper, we utilize the in-teraction of feature information among different neighbors of items as a method for information aggregation in the knowledge graph attention network. This allows us to learn importance weights that reflect the significance of different feature interactions, thereby improving the model's robustness to noisy neigh-bors. We harness the advantages of graph structure in handling high-dimensional sparse data and consider the impact of the inter-action of feature information among item neighbors on item rep-resentation. Our model obtains superior results on two datasets, MovieLens and Amazon-book.