Judgment Of News Oriented Argument Causation

The Internet age has led to a large amount of redundant and meaningless information that people have received.However,many users use the network to make a clear understanding of the "cause and effect" of a particular event,avoiding browsing unrelated or duplication of information.In addition,users hope to find their causal information quickly and accurately from various web pages.So it is necessary to explore the "cause and effect" related tasks in the text.In this paper,the method of causality judgment in the field of news is explored.Firstly,a method of argument labeling based on sentence components is proposed.On the basis of the argument,the method of causality judgment based on syntactic and semantic information fusion is proposed.Finally,the causality knowledge reservoir can be used in information retrieval and question answering and so on.Firstly,the argument labeling task needs to select the two sentence components that can express the relationship from the input natural language sentences.The meaning of the sentence is a sequence of words which can express a complete semantic meaning.In this paper,the input sentence is represented by the syntax tree,and the candidate argument is obtained by the subtree extraction algorithm,and the model is selected from the machine learning method.In this paper,the argument pairs will be used to judge causality.And the task of causal relationship judgment in this paper is mainly to distinguish whether there is a factual and definite effect between the two arguments of input,which is the two classification task.After analyzing the characteristics of the task,we can see that in addition to the semantic information of the element sequence itself,the effect of the syntactic structure information of the input argument on the result performance of causality judgment.Therefore,this paper tries from two aspects: the one method is a causal judgment based on syntactic and semantic information,the support vector machine model is selected and the complex kernel function that is combination of basic kernel function and tree kernel function is used.The sample features include lexicalization features,word vector semantic features and tree syntactic structural features.The two aspect is the causal relation judgment based on the convolution neural network with syntactic information.In this paper,the syntactic tree is used to represent the input arguments,and each “word” node on the syntactic tree is mapped to the word vector space.Then the convolution operation in the convolution neural network model will be completed by the tree convolution kernel.Then the largest number of elements isselected as the output of the pooling operation from the tree window.In this paper,we extend the application of the causal relationship between arguments and construct a causal knowledge base.The key task of constructing the causality knowledge base is to extract the cause and effect.From the input and output forms,it can be seen that the causality has a great similarity to the extraction task and the sequence annotation task.Based on the idea of sequence tagging task,this paper uses conditional random field model,recurrent neural network model,bi-directional long and short memory neural network model and conditional random field and two-way length and short memory neural network model to explore the experimental method of causality extraction.