Research On Automatic Annotation Of Semantic Roles Of Professional Documents ——Taking Statistical Journal Articles As An Example

In recent years,people have conducted a lot of research on semantic analysis.Semantic role labeling is an implementation method of shallow semantic analysis in semantic analysis.Shallow semantic analysis simplifies the form of semantic analysis.It is one of the important components in natural language processing technology.Semantic role labeling is used as an implementation in shallow semantic analysis.At present,it is widely used in many fields such as question answering systems and information extraction.Semantic role tags analyze the arguments of predicates in sentences and the relationship between predicates and arguments.At present,the mainstream of semantic role labeling based on machine learning is based on phrase structure syntax analysis and dependency analysis.Although some achievements have been made,the performance of this research method depends on feature engineering,requiring domain knowledge and a lot of feature extraction Work,no features can express long-distance dependencies and can not introduce heterogeneous resources to solve the problem of insufficient data.Therefore,semantic role labeling based on deep learning has been continuously explored in recent years.The development of deep learning in recent years is very rapid,and it is widely used in many fields,such as natural language processing,speech recognition,image recognition and other computer fields,and has achieved good research results.In the current deep learning model,the long-term and short-term memory network is an improved model of the recurrent neural network.Because it can effectively rely on long-distance information in sequence data,it is considered to be very suitable for processing text sequence data.This paper proposes an end-to-end model of semantic role labeling based on a two-way long-term and short-term memory neural network.This method replaces the traditional feature extraction in semantic role labeling.The result is F1 value,this article mainly studies from the following aspects:(1)This article makes a detailed explanation of semantic role labeling,gives the concept of the relationship between predicate and argument in semantic role labeling,and explains the data set of semantic role labeling,(2)Based on the existing common reference analysis model,a semantic role labeling based on a two-way long-term and short-term memory neural network is proposed,and an end-to-end method is used to predict a given text without a given predicate.The predicate in and the argument parameter span corresponding to the predicate.First,the input data is processed and converted into a word vector.The word vector is used as the input of the two-way LSTM.After the word vector passes the two-way LSTM,the context representation and parameter span of the word are obtained.Factor score to get the most likely combination.(3)The model is trained with CoNLL-2005 and CoNLL-2012 as the training set,the already trained glove vector as the word vector,and the Elmo word vector as the basis to further improve the effect.The test set is from the JASA statistical journal 1054 statistical literature data,make statistics of high-frequency words of statistical literature keywords and draw a word cloud map,select keywords of nearly three hundred documents in 2016-2018 as co-occurrence matrix and co-occurrence map,check the correlation before the keyword.The abstract content of all documents is used as experimental data and processed into json format.As can be seen from the experimental results,the accuracy of semantic role labeling based on deep learning is very high.The follow-up work arrangement is given at the end of this article.