基于卷积神经网络与随机森林算法的专利文本分类模型

为解决专利文档的自动化分类,根据机械领域专利文本的特点,提出了一种基于卷积神经网络与随机森林的机械专利文本分类模型;该模型应用卷积神经网络作为有监督的文本特征提取器,结合随机森林作为分类器,面向机械领域专利文本进行专利文本分类。该模型被应用在包含96类的107 302份英文机械专利文档的数据集上。实验结果表明,该模型相比k近邻、Na6ve Bayes、随机森林等经典机器学习算法在准确率、召回率以及查全率方面均有显著提高。

A Patent Classification Model Based on Convolutional Neural Networks and Rand Forest

This paper proposed an English mechanical patent classification model based on convolutional neural networks and random forest to address automatically patent classification problem. The convolutional neural networks work as the supervised feature extractor and the random forest algorithm serves as the classifier. A series of experiments have conducted in a dataset which consists of 107,302 English mechanical patent documents distributed in 96 categories at subclass level. The experiment results show that our model achieved a significant improvement when comparing to classical machine learning methods such as the k-nearest neighbor, Na?ve Bayes, and random forest, in precision, recall, and F1 aspects respectively.