Exploring Deep Belief Nets to Detect and Categorize Chinese Entities

Abstract

This paper adapts a novel model, deep belief nets (DBN), to extract entity mentions in Chinese documents. Our experiments were designed to develop entity detection system and entity categorization system applying DBN, to complete entity extraction. Our results exhibit how the depth of architecture and quantity of unit in hidden layer of DBN influence the performance. In DBN Systems, token labels are produced independently and DBN does not concerned what the labels of surrounding token are. Viterbi algorithm is a good solution to overcome this issue. It can find the most likely probability label path to make DBN to be more suitable for entity detection. Furthermore, this paper demonstrates DBN is proper model for our tasks and its results are better than Support Vector Machine (SVM), Artificial Neural Network (ANN) and Conditional Random Field (CRF).