基于CNN-LFM模型的个性化推荐

评分数据的稀疏性和新物品的冷启动问题一直是阻碍推荐系统发展的难题。针对这些问题,利用物品的图像数据作为辅助信息以提高评分预测的准确性,提出一种基于卷积神经网络与隐语义模型的推荐模型(CNN-LFM)。CNN-LFM模型利用隐语义模型挖掘评分数据,获得用户和物品的潜在特征,其中物品的潜在特征会在卷积神经网络提取的图像特征的约束下不断完善。在真实数据集下进行实验,对结果的定量和定性分析表明CNN-LFM模型不存在新物品的冷启动问题,即使当评分数据十分稀疏时,其性能也远远优于其它推荐模型。

Personalized Recommendation Based on CNN-LFM Model

The sparseness of rating data and the cold start of new items have been problems that hamper the development of recommendation systems.To solve these problems,the image data of the item were applied as auxiliary information to improve the accuracy of the rating prediction in this article,and a recommendation model based on conv?olutional neural network and latent factor model(CNN-LFM)was proposed.The CNN-LFM model used the latent factor model to mine rating data to obtain potential features of users and items,and the latent factors of the items can be continuously improved under the constraint of image features extracted from the convolutional neural network.Experiments were conducted based on real data sets.Quantitative and qualitative analyses of the results show that there is no cold start problem for new items in this method.Even if the rating data is highly sparse,the performance of CNN-LFM is much better than other recommendation models.