Identification Of Chinese Framenet Semantic Role Based On Neural Networds Model

Given a target word in a sentence,the task of semantic role labeling(Semantic Role Labeling)is to identify all the semantic roles dominated by target word label the role tape automatically.In natural language processing,the semantic role labeling is an important part of shallow semantic analysis(Shallow Semantic Parsing),which has been widely used in the field of machine translation,information retrieval,automatic abstract,etc.Semantic role labeling usually can be divided into two sub-tasks i.e.,role identification and role classification.This paper only focus on the role identification.The paper based on Chinese Frame Net(CFN)of Shanxi University.First,we formalize semantic role identification as a sequence labeling and set the Chinese word as the basic annotation unit,then use neural network model to research.We use block 3x2 fold cross validation to experiment.Based on the depth neural network model,reference[1]research the recognition of the semantic role of Chinese characters.(1)In our paper,for over-fitting of neural network model in[1],we compare the experimental results of Dropout’s neural network model with which not using the Dropout.The result Showed that,compared with[1],the neural network model which using the Dropout significantly increase the accuracy of the model role identification by 10%,recall rate by 1%,F-value by 6%,which illustrate that using the Dropout can relieve the over-fitting of neural network model.It significantly improve the performance of the model.(2)In the neural network model,for experiment,we select word embedding-Word2vec,NLM,RNNLM as initial input.the performance is slightly improved than random vector input.The F-value increase by about 1%when using the word embedding of NLM.In addition,this paper also studied the influence of the dimension word,part of speech,location and the target word characteristics on performance of neural network model,finding the model is optimal when these characteristics dimension are 100,30,30,40.Besides,the result illustrate the neural network model is optional when learning rate is 0.001.The main innovation of this paper is that,it is the first time the Dropout punishment mechanism be introduced into neural network model for CFN semantic role recognition,and combine with word embedding to compare the test.The experiment proved that,it is remarkable to Chinese framework semantic role identification when use the Dropout punishment mechanism and the word embedding in the neural network model.