LARA:Attribute-to-feature Adversarial Learning For New-item Recommendation

With the continuous expansion of e-commerce,the number and variety of products are growing rapidly,and customers need to spend a lot of time to find the products they want to buy.This process of browsing a lot of irrelevant information and products will undoubtedly keep consumers drowned in the problem of information overload.The importance of a good recommendation system to users is self-evident.The cold start problem has always been an important issue in the recommendation system.The cold start problem is a classic problem that is widely concerned in collaborative filtering recommendation algorithms.This problem has always affected the healthy development of traditional collaborative filtering recommendation systems,and its existence has seriously affected the recommendation quality of recommendation systems.For e-commerce recommendation system,there are a large number of new users access to the system every day,and there are a number of new projects are added to the system,on the one hand,if the recommendation system can recommend for new users prefer goods,recommendation system will win the trust of more users,increase customers for merchants,improve the loyalty of users on the system for users,can get high quality personalized service;On the other hand,if new products can be recommended in time,the sales volume of the products can be improved,the merchants can win greater economic benefits,and the healthy development of e-commerce can be promoted.The key obstacle to solving the cold start problem is the lack of user interaction with the new product.When a new product is produced,it is impossible to judge how much the user likes the product because the user does not know the product.Therefore,it is impossible to recommend products to users.If we can know how certain users like this product,then we can decide whether to recommend the new product to a certain user or not according to the degree of similarity among users,so as to solve the cold start problem.Therefore,how to establish the relationship between users and this new product has become the focus of work.Fortunately,products always have attributes,and we can always use the attributes of a product to make a general guess as to who will be interested in this new product.For example,there is a newly produced down jacket,which has the attributes of hat,long version,99%down content,etc.From these attributes,we can infer that users who like this down jacket have the following characteristics:they like to wear hats,long version clothes,and prefer clothes with high down content to cotton.Through this speculation,we can establish a connection between the new product and an imagined user,and then we can compare the real user with the imagined user one by one,so as to recommend the down jacket to similar users.Recommending new items in real-world e-commerce portals is a challenging problem as the cold start phenomenon,i.e.,lacks of user-item interactions.To address this problem,in this paper we propose a novel recommendation model,i.e.,adversarial neural network with multiple generators,to generate users from multiple perspectives of items' attributes.Namely,the generated users are represented by attribute-level features.As both users and items are attribute-level representation,we can implicitly obtain user-item attribute-level interaction information.In the light of this,we can recommend new item to users based on attribute-level similarity.Extensive experimental results on two item cold-start scenarios,movie and goods recommendation,verify the effectiveness of our proposed model as compared to state-of-the-art baselines.