一种基于 Inception 结构的神经网络协同过滤推荐算法

在商品推荐领域，商品评论信息往往难以得到有效利用。为了充分利用商品评论信息，提高商品推荐系统精度，对 NCF 神经网络协同过滤模型进行改进，将 NCF 模型与 Inception 结构的卷积神经网络相结合，提出基于 Inception 结构的神经网络协同过滤方法（NCF-i 模型），将商品评论信息融入模型进行预测和推荐。首先基于 Inception 结构的卷积神经网络对商品评论信息进行分析并提取多元特征模型，然后将多元特征模型添加到 NCF 模型中，通过多层全连接层获取用户、商品及商品评论之间的非线性关系，最后基于此非线性关系对商品进行预测和推荐。通过基于真实数据集的实验证明，应用 NCF-i 模型的推荐算法，推荐系统的预测精度和稳定性均优于当前常用的推荐模型。

An Neural Collaborative Filtering Recommendation Algorithm Based on Inception Structure

In the field of product recommendation，product review information is often difficult to be used effectively. In order to make full use of commodity review information and improve the accuracy of commodity recommendation system，this paper improves the neural network collaborative filtering model of neural collaborative filtering（NCF），combines the NCF model with the convolutional neural network of Inception structure，and proposes a neural network based on Inception structure. The collaborative filtering method（NCF-i model）integrates product review information into the model for prediction and recommendation. The method firstly analyzes the commodity review information based on the convolutional neural network of Inception structure and extracts its multi-feature model. Then，the multi-feature model is added to the NCF model to obtain users，commodities and commodity reviews through multiple layers of all-connected layers. The nonlinear correlation between the two is based on this nonlinear relationship to predict and recommend commodities. Through the experiment based on the real data set，the recommended algorithm of the NCF-i model is applied，and the pre? diction accuracy and stability of the recommendation system are better than the currently recommended models.