Implicit discourse relation analyzing method based on recurrent neural network

Abstract

The invention provides an implicit discourse relation analyzing method based on a recurrent neural network and belongs to the technical field of natural language processing application. The method comprises the following steps that firstly, word vectors of a training corpus are initialized based on a certain regulation; then, the word vectors serve as inputs of a Bi-LSTM model, two hidden-layer vectors of the Bi-LSTM model are obtained, and the two hidden-layer vectors are stitched and then serve as inputs of the recurrent neural network, wherein a syntactic parsing tree of the network structure is obtained through annotation of a PDTB corpus, and a composite function is synthesized by using neural tensors; finally, the vector representation of each argument is obtained, the two argument vectors are stitched and then input an MLP for classification, parameters in the model are updated by using a stochastic gradient descent method to be convergent, and the analysis of the implicit discourse relation is completed by using the parameters with optimal performance.

Description

technical field [0001] The invention relates to an implicit discourse relationship analysis method, in particular to a recursive neural network-based implicit discourse relationship analysis method, which belongs to the technical field of natural language processing applications. Background technique [0002] As an important task in the field of natural language processing application technology, textual relationship analysis, especially implicit textual relationship analysis, has been unremittingly studied by scholars, and has played an important role in statistical machine translation, information extraction, sentiment analysis and other fields. important role. As the semantic analysis of natural language has gradually become the mainstream of academics and applications, more and more researchers have paid attention to how to efficiently and correctly understand the structure and content of an article. Now coincides with the era of big data, massive and unstructured infor...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to solve the following problems in the past in the implicit discourse relationship analysis: 1) the method for feature engineering cannot effectively use deep semantic information and the problem of data sparseness; 2) the method for ordinary neural networks does not consider The problem of the original coherence of the sentence; 3) The problem of the original syntactic structure information of the sentence is not considered in the ordinary deep learning method; 4) The problem of polysemy in the word itself

The invention proposes to use the distributed representation of words to solve the sparsity problem, use the two-way LSTM network to solve the word polysemy problem, and use the recurrent neural network to fuse the syntactic structure information, so as to perform implicit discourse analysis on the basis of deep semantic understanding

Images