关键词: 深度学习, 语义文本, 重叠区域检索, 深度置信网络, 特征学习

Abstract:

As a hot topic in natural language processing， overlapping region recognition needs to be further explored and studied. Aiming at the problem of poor accuracy and recall in traditional text overlapping region retrieval methods， a large-scale semantic text overlapping region retrieval method based on deep learning is proposed. Combined with sparse automatic encoder and depth confidence network， a hybrid model is constructed. According to the hybrid model， a text classifier is designed and constructed. The main components of the classifier are text preprocessing， feature learning and classification retrieval. In this paper， a series of preprocessing， such as de-noising， word segmentation and stop word removal， are carried out. Finally， softmax regression is used to realize text classification， and the learned text features are used as the input of the classifier to get the classification and retrieval results of the overlapping regions. The experimental results show that the accuracy and recall of the method are both high， showing reliability and robustness.

Key words: deep learning, semantic text, overlapping region retrieval, deep confidence network, feature learning