利用函数拟合对DeepFM算法的改进研究

Research on improvement of DeepFM algorithm by function fitting

DeepFM模型是基于FM模型与Wide&Deep模型的改进,该推荐算法主要基于深度学习通过已知特征来预测用户点击某一按钮的概率。但随着电子商务的发展,不仅需要通过横向特征预测用户点击某一按钮的概率,还要纵向考虑该按钮在不同时间段的点击概率变化。文中对DeepFM进行了改进,引用了拟合函数的方法,通过各个拟合的函数计算出点击概率变化的函数图像,延展该图像得知该按钮被点击概率随着第三维坐标时间值的变化,从而实现了各种用户在不同时间对于不同商品需求的预测。该算法运用了相对平滑的函数曲线来拟合模型计算的结果,提高了模型的精确度。

DeepFM model is based on the improvement of FM and Wide&Deep.The recommendation algorithm is mainly based on deep learning to predict the probability of users clicking a button through known features.However,with the development of e-commerce,we may not only need to predict the probability of users clicking a button through horizontal features(taking time as an example),but also consider the change of click probability of the button in different time periods vertically.Therefore,DeepFM is improved by quoting the method of fitting function to calculate the function image of the change of click probability through each fitted function,and extend the image to know that the click probability of the button changes with the third dimensional coordinate(here is time).Thus,the demand prediction of various users for different commodities at different times is realized.At the same time,the algorithm uses a relatively smooth function curve to fit the calculation results of the model,which improves the accuracy of the model.