基于小批量梯度下降法的个性化推荐模型

Personalized Recommendation Model Based on Mini-Batch Gradient Descent

摘要: 隐语义模型(LFM)是推荐系统中应用比较广泛的模型之一。其核心思想是通过隐含特征联系用户兴趣和物品，得到用户对物品偏好关系的目标函数，然后通过随机梯度下降法求得最优解，从而个性化地对用户进行物品的推荐。但是计算过程中随机梯度下降法会造成目标函数值震荡的比较剧烈，准确度欠缺，所以提出对目标函数优化选择用小批量梯度下降法，建立基于小批量梯度下降法的个性化推荐模型，减少目标函数最优解的随机性，提高准确度，减少运行时间，从而达到提高个性化推荐质量的目的。实验数据采用Movielens数据集，Python作为工具，均方根误差(RMSE)、平均绝对误差(MAE)作为标准，将改进前后的算法结果做对比，验证基于小批量梯度下降法的个性化推荐模型能够得到更好的推荐效果。

Abstract: Latent Factor Model (LFM) is one of the most widely used models in the recommendation system. The core idea is to connect users’ interests and items through implicit features, obtain the objective function of users’ interests relationship for items, and then obtain the optimal solution by Stochastic Gradient Descent, so as to make personalized recommendation for users. However, in the process of calculation, the stochastic gradient descent method will cause the value of the objective function to oscillate violently and the accuracy is insufficient, so we choose Mini-Batch Gradient Descent method to optimize the objective function, establish a model of personalized recommendation based on Mini-Batch Gradient Descent, reduce the randomness of the optimal solution of the objective function, and improve the accuracy and reduce the running time, so as to improve the quality of personalized recommendation. The experimental data adopted Movielens data set, Python as the tool, RMSE (root mean square error) and MAE (mean absolute error) as the standard, and compared the experimental results of several algorithms, verifying that the personalized recommendation model based on Mini-Batch Gradient Descent can get better recommendation effect.