A deep neural network of multi-form alliances for personalized recommendations

Abstract

The collaborative filtering adopted by traditional recommendation system has data sparsity problem, and the matrix decomposition method simply decomposes users and items into linear models for potential factors. These limitations have led to limited effectiveness for traditional recommendation algorithms. In this case, the recommendation system based on deep learning has emerged. Most of the current deep learning recommendations use deep neural networks to model some basic information, and in the modeling process, according to the input data categories, multiple mapping paths are used to map the original data to the potential vector space. However, these recommendations ignore that the alliance between different categories may have a potential impact on the recommendation effect. Aiming at this problem, this paper proposes a feedforward deep neural network of multi-form category features combination for recommendation, which is deep alliance neural network. According to the different alliance modes, it can be divided into deep series network (DSN), deep parallel network (DPN) and deep random network (DRN), which are used to solve the recommendation problem of implicit feedback. At the same time, a fusion model SMLP based on deep alliance neural network and traditional Multi-Layer Perceptron (MLP) is proposed to try to explore the performance of the fusion model. Finally, experiments on public datasets show that our proposed method significantly improves the existing methods. Empirical evidence indicates that deep series network and deep parallel network can provide better recommendation performance, while the recommended performance of deep random network and fusion model SMLP is not ideal. This indicates that the deep alliance network needs to pay special attention to the order of the category features in the process of category features association.