利索-全球专利检索

  1. 登录
  2. 注册
发明授权
CN106022356B 一种基于梯度下降法的多视图GEPSVM网页分类方法 失效 - 权利终止

一种基于梯度下降法的多视图GEPSVM网页分类方法
摘要：
本发明提出了一种基于梯度下降法的多视图GEPSVM网页分类算法,包括MvGDSVM网页分类模型参数训练步骤和网页数据分类步骤；MvGDSVM网页分类模型参数训练步骤包括：步骤A：输入网页训练样本数据；步骤B：对网页训练样本数据进行预处理；步骤C：训练MvGDSVM网页分类模型参数；网页数据分类步骤包括：步骤a：输入待测网页样本数据；步骤b：对待测网页样本数据进行标准化预处理；步骤c：通过MvGDSVM网页分类模型对待测网页样本数据进行分类。本发明提出的基于梯度下降法的多视图GEPSVM网页分类算法，通过引入一个多视图协同规范化项来最大化不同视图间分类的一致性，从而有效地结合了两个单视图的提高性的广义特征值最接近支持向量机，最后利用共轭梯度下降法来求解生成的优化问题。
摘要（英）：

The invention provides a gradient-descent-method-based multi-view generalized-eigenvalue-proximal-support-vector-machine (GEPSVM) webpage classification algorithm comprising an MvGDSVM webpage classification model parameter training step and a webpage data classification step. The MvGDSVM webpage classification model parameter training step is realized by the following steps: step A, inputting webpage training sample data; step B, carrying out pretreatment on the webpage training sample data; and step C, training an MvGDSVM webpage classification model parameter. The webpage data classification step is realized by the following steps: step a, inputting to-be-measured webpage sample data; step b, carrying out standardized pretreatment on the to-be-measured webpage sample data; and step c, classifying the to-be-measured webpage sample data by using an MvGDSVM webpage classification model. According to the invention, classification consistency of different views is maximized by introducing a multi-view cooperative standardized item, so that the improved GEPSVMs of the two single views are combined effectively; and a generated optimization problem is solved by using a conjugate gradient descent method.

公开/授权文献
0/0