Font Size: a A A

Research On Chinese Coreference Resolution Based On Pre-trained Language Model

Posted on:2022-07-17Degree:MasterType:Thesis
Country:ChinaCandidate:W M HuangFull Text:PDF
GTID:2518306569481694Subject:Software engineering
Abstract/Summary:PDF Full Text Request
Coreference resolution aims to identify different mentions of the same entity in the text,which plays an important role in many high-level natural language tasks.With the rapid de-velopment of natural language processing,more and more scholars apply deep learning to the research of Chinese coreference resolution.However,the current model based on deep learning often uses static word vectors and recurrent neural networks to encode the text,which can not well model the context semantics of the text.In addition,these models only focus on the lo-cal features of mention,ignoring the importance of entity information features for coreference resolution tasks.To solve the above problems,this paper proposes an end-to-end neural network Chinese coreference resolution model based on a pre-trained language model and entity information enhancement.It adopts the idea of span ranking and takes the spans in the text as potential mentions for subsequent resolution operations.In this model,the Chinese pre-trained language model BERT-wwm-ext is used as the text coding layer to fully learn the deep semantic features of the text and obtain the text vector representation with richer semantic knowledge.In addition,in order to fully consider the global feature,we use multiple iterations to optimize the vector representation of mention through the gating mechanism,so as to strengthen the influence of entity information in the process of resolution.The experimental results show that the proposed model achieves an average F1 score of68.88% on the Onto Notes 5.0 dataset,which is better than the current mainstream methods and can effectively improve the effect of the Chinese coreference resolution task.In addition,through the design of multiple groups of comparative experiments,we prove that the perfor-mance of the model can be effectively improved by integrating the pre-trained language model and global entity information features.Finally,based on the model proposed in this paper,we implement a Chinese coreference resolution system,introduce the overall structure,module de-sign,and result display of the system in detail,and discuss the feasibility of the proposed model in practical application.
Keywords/Search Tags:Pre-trained language model, Coreference resolution, Neural network
PDF Full Text Request
Related items