Research On Self-Supervised Sequential Recommendation Algorithm Based On User’s Long And Short-Term Interests

In recent years,deep neural networks have been widely used in recommender systems because of their effectiveness in capturing and modeling user preferences.At the same time,hypergraph and self-supervised learning are also gradually into people’s vision and has achieved good results in the field of recommended systems.For the next-item recommendation task,we summarize the problems existing in the current work into the following three points:(1)In the past,people used complementary relationships between the user’s long-term interests and short-term interests to recommend,but this paper found that they have the same recommendation objectives,and therefore there are constraints.(2)The sequential behavior of the user in the current session will contribute to the recommendation.(3)The user’s short-term interest plays an important role in the prediction and recommendation of the next item.This paper will make the following improvements for the above issues:(1)To give full play to the complementary and restrictive relationship between users’ long and short-term interests,this article innovatively applies self-supervised learning to the use of users’ long and short-term interests,to maximize the mutual information on users’ long and short-term interests to model a more accurate representation of user’s interest.(2)This paper applys hypergraph convolution to user short-term interest modeling by considering the recent achievements of hypergraph convolution in user short-term interest modeling.This paper is the first to consider the sequential relationship in the interaction sequence when constructing the hypergraph.This paper proposes a Self-supervised Recommendation Network with users’ Long and Short-term interests(SRNLS)to solve the above problems.SRNLS consists of three main components.Specifically:(1)SRNLS captures the evolution of the user’s long-term sequence of interests through the GRU to model the longterm interest of the user.(2)SRNLS uses hypergraph convolution to model the user’s shortterm interests and to make more accurate recommendations,taking into account the sequential information of the items in the current session when composing hypergraph.(3)A selfsupervised learning component(SSL)with a user’s long and short-term interests is proposed,which can better play the constraint relationship between the user’s long-term interests and the user’s short-term interests.A large number of experiments were conducted on the common movie datasets MovieLens-100 K and MovieLens-1M,and the results showed that SRNLS was significantly better than the advanced baseline in capturing user preferences.