Research On The Method And Application Of Null Instantiation Recognition And Filling

Null Instantiation(NI)refers to the implicit semantic elements in a sentence.The correct understanding of these semantic elements will affect the semantic integrity of the text.For the machine,it is a difficult task to recognize and implement the content fil ing of implicit semantic components.It is necessary to accurately understand the text with the help of the corresponding context to effectively solve this problem.Chinese FrameNet(CFN)describes the context of a sentence from the perspective of frame semantics,and contains semantic information describing the specific context of the sentence.This paper recognizes and fils in the NI based on the CFN,and applies it to the task of frame relation discrimination.The main work of this paper is as follows:(1)Null Instantiation recognition.According to the different semantic understanding of the missing semantic elements,this paper identifies the NI,which is based on the traditional machine learning and deep learning methods.In the method based on deep learning,the input layer first contains word embedding and frame representation,then BiLSTM is used to obtain context information,and finally the NI category is predicted through the full connection layer.Three frame representation methods based on WASBIE,Word2 vec algorithm and frame relation are explored.In the traditional machine learning algorithm,the optimal selection was constructed,and experiments were carried out on the two algorithms of decision tree and random forest respectively to obtain the NI category.The results show that the NI model built in this article is 2-9% better than the baseline results.(2)Definite Null Instantiation filling.In this paper,we search for the content of the DNI in the context of the text.Aiming at the problem of non-equilibrium of candidate set data,this paper proposes an Optimized smote(synthetic minority oversampling technique)algorithm to extend the data to provide a balanced data set for DNI fil ing.In addition,incorporating semantic features in the process of building a classification models,and the mapping relationship between frame elements is used to improve the effect.The results show that the balanced processing of the integrated data and the semantic information of the frame are 12% higher than the baseline result.(3)Application of Null Instantiation recognition and filling in frame relation discrimination.The task of frame relation discrimination is the difficulty of the task of frame semantic analysis.In this paper,NI recognition and fil ing is applied to this task,and the influence of NI on the task is explored.First of all,for the NI involved in the sentence,the aforementioned method is used to find out the filling content in the context,and then the relevant features are selected to build a classification model to determine whether there is a connection between the frames.The results show that the addition of the NI increases the F1 value of the frame relation discrimination task by 1.53%,indicating that the NI has a positive effect on the frame relation discrimination.