Chinese Semantic Role Labeling Integrating Global Information and Attention Mechanism

Abstract

With the rapid development of artificial intelligence and Chinese information processing technology, natural language processing related research has gradually deepened to the level of semantic understanding, and Chinese semantic role labeling is the core technology in the field of semantic understanding. In the field of Chinese information processing where statistical machine learning is still the mainstream, traditional labeling methods rely heavily on the parsing degree of sentence syntax and semantics, so the labeling accuracy is limited and cannot meet current needs. In response to the above problems, this paper proposes a CNN-BiLSTM-Attention-CRF fusion model for Chinese semantic role tagging, and at the same time the model performance optimization research. In the model training stage, convolution kernels of different sizes are used to capture the local features of the sentence, and then the different local features are spliced into a new feature vector through the average pooling technology, which is the global feature that integrates the semantic information of the entire sentence sequence, the same part of speech, sentence phrase Multi-level linguistic feature groups such as structure are fed into the model together. Through multiple sets of experimental demonstrations, the feature set integrated into the sequence global information can significantly improve the various indicators of the model.