Research On Convolutional Sequence Recommendation Model Based On Inception

Sequential recommendation is a hot spot in the field of recommendation systems in recent years.Unlike traditional recommendation systems that make use of the long-term interests of users to make recommendations,sequential recommendation considers that the user's interests will change with time,and relies on the sequential information generated by the user's interaction with the items to dynamically build the user's interests.It can more accurately complete the recommended tasks for users.At present,there are many ways to extract user's sequential information.The most novel one is to embed the user's behavior sequence information into an "image" in time and latent spaces,and use the convolutional filters to extract the local features.However,this method has certain problems,existing convolutional networks either cannot extract sequence information well,or are prone to overfitting.This paper draws on the design idea of Inception network to construct a convolutional sequence recommendation model based on Inception.This recommendation model is established by setting up two different convolutional network layers: dynamic convolutional layer and static convolutional layer.It can more fully extract the short-term interest of users.And embed the user embedding as the user's long-term interest in the output of the convolutional layer to build a complete user's interest.Based on this,the user obtains the recommendation result.Through experiments on three public data sets: Movie Lens 1M,Gowalla,Steam,and compared with other benchmark models.It verifies that the performance of Inception-based convolutional sequence recommendation model is better than the latest sequential recommendation model.Among the three evaluation indicators of Top-N series(Precision@N,Recall@N,MAP),the average increase is about 10%,and the maximum increase on a single index is 14%.