Design Of Recommendation System Based On Multi-behavior Interactive Data

With the rapid development of information technology and the popularization of network services,data scale has achieved blowout development,and gradually entered the era of big data.As an important way to solve the problem of information overload,recommender system has developed rapidly,and it also plays an important role in all aspects of human production and life.As a branch of recommender system,multi-behavior recommender system started late,but due to the richness of multi-behavior data and the universality of application scenarios,multibehavior recommender system has been favored by people from all walks of life in recent years.At present,the multi-behavior recommendation system has a great improvement in how to reasonably use the multi-level preference information between behaviors,how to consider the item category information and how to learn from the preference information of similar users.This paper mainly starts from the above problems,and designs a new recommendation system based on the application scenario of multi-behavior,which mainly includes three tasks.The first one is a multi behavior recommendation system based on item category and graph neural network,which uses graph neural network to aggregate heterogeneous information of multi-behavior between user and item,and takes into account the homogeneous information between item and item;Second,based on the first work,a multi-behavior recommendation system based on similar user preferences is proposed,which mainly considers the influence of homogeneous information between users;The third work is the recommendation system based on pairwise ranking algorithm.In the negative sample sampling process of pairwise loss function,positive and negative sample pairs are extracted according to multi-level preferences and item category information to reduce the uncertainty of negative samples and improve the accuracy of the recommendation model.The experimental results on two real-world datasets show that our model has achieved good performance and is superior to the more advanced algorithms in the field of multi-behavior recommendation in terms of recommendation accuracy.