Research On Dynamic Context-aware Sequential Recommendation

Recommender system plays an important role in the modern Internet platform.Its appearance solves the problem of "information overload" to a great extent.It is a very effective method of information filtering.It can learn the user's interest preferences according to the user's relevant information or historical purchase records,so as to recommend the needed information or items for users.Sequence based recommendation is a new topic in the field of recommender system,and has attracted great attention from academia and industry in recent years.However,the current model still has the following problems:(1)the traditional sequential recommendation algorithm ignores the information such as item neighbors,which will affect the user's purchase behavior.(2)most of the existing algorithms use the short-term memory of the last click to consider the user's current information,so as to obtain the user's short-term information,However,the last click can not fully represent the short-term interests of users,and the existing data sets are large in scale,and there will be problems such as memory explosion when inputting more data at one time.This paper proposes a new algorithm based on sequence recommendation for(1).The algorithm calculates the item with the most similar characteristics to the user's last item by similarity calculation,and puts it into the memory network as the neighbor information of the item.Then,the recurrent neural network is used to learn the user's historical interest from the context information of the user's purchase.Finally,the user's historical interest is integrated with the neighbor information of the item to complete the sequence recommendation.This paper proposes a new method to capture short-term preferences for(2).Sliding window is used to capture users' short-term interests dynamically.The features learned from each sliding window are regarded as users' short-term interest preferences,and users' short-term preferences also change dynamically during the sliding process.This dynamic context based method can not only avoid the memory explosion caused by learning too many features at one time,but also obtain the short-term preferences of users in a more fine-grained way,so as to achieve the purpose of better recommendation.Experimental results show that,compared with several benchmark algorithms and current frontier algorithms,the results of the proposed model on data sets reach a better level.Thus,the effectiveness of the dynamic context recommendation algorithm is verified.