Feature-based preprocessing and recommendation method

The invention relates to a feature-based preprocessing and recommending method. The feature-based preprocessing method comprises the steps of dividing feature data to be preprocessed into a pluralityof sub-buckets according to distribution of objects; obtaining a plurality of mean values of to-be-preprocessed feature data in the plurality of sub-buckets; obtaining a plurality of logarithms of theplurality of mean values; and performing normalization processing on the plurality of logarithms. According to the bucket mean logarithm standardization method, the model can be more stable, the reliability is higher, and the negative influence of abnormal distribution in the data is reduced. 本发明涉及一种一种基于特征的预处理及推荐方法。基于特征的预处理方法，包括：根据对象的分布将待预处理的特征数据划分成多个分桶；获取多个所述分桶内待预处理的特征数据的多个均值；获取多个所述均值的多个对数；以及对多个所述对数进行归一化处理。本发明的分桶均值对数标准化方法，能够使得模型更稳定，可靠性更高，降低数据中异常分布的负面影响。