Research on Personalized Product Recommendation Algorithm Based on Deep Learning

In view of the insufficiency of feature information utilization and overfitting in the recommendation task of Transformer network, this paper proposes a model combining random shared embeddings and multi-layer multi-head attention blocks to solve the problems still existing in the above deep learning-based recommendation algorithm. Aiming at the problem of multi-feature information utilization, this paper studies the use of multi-layer multi-head attention block to fuse multi-information. CARCA-SSE model is designed in this paper. Different from traditional location embedding, this paper uses the absolute date of user interaction as time embedding to make full use of existing information. To solve the overfitting problem, random shared Embedding is added to the recommended model based on CARCA, and the Embedding layer is randomly replaced by another Embedding with a certain probability. Using the replaced embedding can reduce the risk of overfitting and further improve the generalization ability, thus increasing the robustness of the model.