Time and Space Aggregation Recommendation Model Based on Synthetic Negative Samples

Abstract

Currently, the recommendation model of the neural network is usually based on space aggregation neighborhood embedded, focusing on gathering information from the perspective of spatial structure information to learn the characteristics of user projects. The negative sampling method is difficult to balance positive and negative samples. In response to the above issues, this article proposes a time-space aggregation recommendation model based on synthetic negative samples. The model uses the multi-header attention mechanism to capture the chronological order of the neighborhood through the multi-header attention mechanism through interactive sequence diagram and sample mixed negative sampling strategies. A mixed sampling of different layers of pooling, thus synthesizing high-quality negative samples so that the model can better learn the boundary between positive and negative instances. Experiments show that the model fully captures users' dynamic interests, enhances the extraction effect of timing characteristics, and alleviates the problem of imbalance of positive and negative samples. And this model can be naturally inserted into the recommendation model of the neural network, which is universal.