A Negative Sampling-Based Service Recommendation Method

As the number of services has increased dramatically in recent years, it has become a challenge to provide users with high quality services. At present, the collaborative filtering technology based on graph neural network has achieved good results in service recommendation. However, these methods ignore the influence of negative sample neighbor nodes. Therefore, we propose a service recommendation method based on negative sampling, which uses the information of neighbor nodes of non-interacted services to generate negative samples to improve the recommendation accuracy. In this method, firstly, we construct the dataset for training based on the interaction records between users and services; second, we use the interpolation mixing technique to pollute the negative samples, and generate synthetic hard negative samples by fusing the negative samples selected by the hard negative selection strategy through the pooling operation; then, the samples are used in GNN-based recommender systems to obtain more accurate user and service embedding representations; next, calculate the inner product score embedded by the user and the service, and recommend high inner product score services to user. Finally, experiments were conducted on three real datasets, Shopify, Programmable, and 360Apps, and the experimental results demonstrate that the proposed method can achieve better recommendation results such as average relative increase of 30% for LightGCN and 8% for NGCF in terms of NDCG@20.