Research On Sequential Bundle Recommendation Model Based On Neural Attention Mechanism

With the advent of the Internet era,hundreds of millions of information is updated on the network every day.When the user needs are not clear,how to quickly find the content that the user needs in the complicated information becomes a challenge.The recommendation system can not only recommend other similar items that users have bought,but also can increase the users’ purchase volume by recommending the product attachments.The bundle recommendation is more in line with the users’ consumption habits and purchase preferences.In order to capture the sequential relationship among the items which in bundles,this paper proposes a sequential bundle recommendation model based on neural attention mechanism by comprehensively considering static bundle and dynamic bundle.This model not only considers the sequential relationship among bundles,but also the sequential order of items which in bundles,so as to capture the users’ global preference and the users’ local preference at the same time.The model firstly uses the neural attention mechanism to merge multiple items of users,and adds the time latent representation of each item to ensure the sequence of items which in the bundle,learning the weight distribution of each item in the bundle.Then adding the location information of each bundle to ensure the sequence of bundles.Secondly,the convolutional neural network is used to capture the feature information of each item in the bundle and the sequence pattern among the bundles.Then concatenate it with the users’ global preference,so as to more accurately capture the users’ overall preference and improve the accuracy of the recommendation.In addition,due to the large amount of semantic information contained in the review text,this paper divides the review information which typewrite by users into user review and item review,aiming to alleviate the cold start problem existing in the bundle recommendation by merging user review and item review with the users’ and the items’ potential presentation,respectively.In this paper,the experimental results on three real data sets of Amazon show the effectiveness of the model.Compared with other four models by setting different parameters,the results show that the model in this paper still performs well under different parameter settings,and the proposed model has some improvement over the existing methods in the bundle recommendation.