Enhanced Neural Network Factorization Machine Via Pairwise Ranking And Attention

With the development of information technology,a variety of network applications have accumulated a huge amount of data.Although the massive data provides users with rich information,it leads to the problem of “information overload”.The recommendation systems can greatly alleviate the problem of information overload.They infer users latent preferences by analyzing their past activities and provide them with personalized recommendation services.In the field of recommendation systems,collaborative filtering(CF)algorithms are the most popular methods,which utilize users' behavior information to make recommendations and are independent of the specific application domains.However,the traditional collaborative filtering methods ignore the contextual information related to users and items,resulting in a suboptimal recommendation performance.In reality,the contextual information greatly affects the decisions of users.The factorization machine models attract significant attention nowadays since they improve recommendation performance by incorporating context information into recommendation modeling.However,traditional factorization machine models often adopt the point-wise learning method for model parameter learning,as well as only model the linear interactions between features.They substantially fail to capture the complex interactions among features,which degrades the performance of factorization machine models.This article first introduces the basic factorization machine model principle,on this basis,indepth research and implementation of the neural network factorization machine fusion attention mechanism and pair-wise ranking.The main work of the paper is as follows:(1)In order to capture the high-order and nonlinear interactions among features,a neural pairwise ranking factorization machine(NPRFM)for item recommendation is proposed.The multilayer perceptual neural network is integrated into the pairwise ranking factorization machine model,and the pair-wise ranking model is adopted to learn the relative preferences of users.(2)To learn the importance of different feature interactions and reduce the interference of useless feature interactions on the model,the attention mechanism is introduced on the pair-wise interaction layer,and the attention pairwise ranking factorization machine(APRFM)is proposed.It adopts the attention mechanism to learn the importance of feature interaction,and also uses the pairwise ranking model to learn the relative preferences of users.(3)In order to further improve the recommendation performance,an enhanced Attention NPRFM combining the advantages of NPRFM and APRFM is proposed.The attention mechanism is introduced in the pool operation of the Bi-interaction layer,and a multilayer perceptual neural network is stacked on the attention pooling layer.The empirical results on real-world datasets indicate that the proposed the NPRFM,APRFM and Attention NPRFM models outperform the traditional factorization machine models.