Feature selection method for multi-label data, terminal equipment and storage medium

The invention relates to a feature selection method for multi-label data, terminal equipment and a storage medium. The method comprises the following steps: constructing a target function for feature selection by adopting a maximum entropy model; solving the target function to obtain the size of a mapping parameter corresponding to each feature in the feature space; and carrying out feature selection based on the mapping parameters. According to the method, a shared generic feature relationship is constructed in an output space by using linear correlation among the marks, and feature weights obtained by mark distribution mutual information analysis are used for feature weighting. In addition, a sparse normal form is used for selecting common features with high classification capability in the whole marking space, and the robustness of the algorithm is improved. 本发明涉及一种多标记数据的特征选择方法、终端设备及存储介质，该方法中包括：采用最大熵模型构建用于特征选择的目标函数；通过对目标函数进行求解，得到特征空间中各特征对应的映射参数的大小；基于映射参数的大小进行特征选择。本发明利用标记间的线性相关性在输出空间上构建它们所共享的类属特征关系，并使用标记分布互