神经因子分解机推荐模型改进研究

因子分解机对特征各维度之间的一阶线性关系和二阶线性关系建模，在推荐系统中已有较多应用。神经因子分解机模型（NFM） 是因子分解机与神经网络的结合模型，它能捕获特征之间的高阶交互信息，使得模型预测效果更佳。但由于神经因子分解机模型一般都是采用全连接的前馈神经网络，使得整个推荐网络过于复杂，存在过拟合风险。为了降低神经因子分解机模型的整体复杂度，提高推荐模型的泛化性能，提出一种基于交叉网络的因子分解机模型（CFM），降低模型复杂度，提高模型泛化性能。实验表明，该模型在数据集上的预测准确度为77%左右，相比NFM预测准确度提高了约2%，整体模型泛化性能也有所提高。

Research on Improvement of Recommendation Model Based on Neural Factorization Machines

The factorization machine models the first-order linear relationship and the second-order linear relationship between the various dimensions of the feature， and it has been widely used in the recommendation system. The neuron factorization machine model （NFM） is a combination model of factorization machine and neural network. It can capture high-order interaction information between features， which makes the prediction effect better. However， since the neural factorization machine model generally uses a fully connected feedforward neural network， the entire recommendation network is too complicated and there is a risk of overfitting. In order to reduce the overall complexity of the neural factorization machine model and improve the generalization performance of the proposed model， this paper proposes a cross-network-based factorization machine model （CFM） to reduce the complexity of the model and improve the generalization performance of the model. Experiments show that the prediction accuracy of the model on the data set is about 77%， which is about 2% higher than the NFM prediction accuracy， and the generalization performance of the overall model is improved.