Extraction method for test question corpus emotion and reason sentence pair

Abstract

The invention discloses a method for extracting sentences of emotions and reasons in a test question text based on a graph convolutional network, and the method comprises the following steps: obtaining test question text data, and carrying out the sentence segmentation preprocessing of the data; converting the sentences into vector representation by using a BERT pre-training language model; using a full connection layer network to generate corresponding pairing combination representations from the sentence representations; embedding a vector by using a relative position to enhance a pairing combination representation vector; performing average pooling on the sentence vector representation to obtain document representation; carrying out coding interaction on four expressions of an emotion sentence, a reason sentence, a pairing combination and a document by using a graph convolutional network; and finally, predicting whether a pairing combination is established by using a multi-layer perceptron model. According to the method, the causal relationship between the two sentences in the pairing combination is fully considered when the pairing combination is modeled, and the condition that the wrong pairing combination is predicted to be true can be effectively avoided, so that the prediction accuracy is improved.

Description

technical field [0001] The invention relates to the field of sentence pairing relationship analysis between emotion and reason of educational evaluation test text in natural language processing, and in particular to a sentence pairing extraction method for emotion and reason in test text based on a graph convolutional network. Background technique [0002] Letting natural language processing technology participate in the field of educational examination has always had important research, practice and application value. For example, using the method of text emotion and reason sentence extraction, it is possible to automatically extract sentences expressing emotions from characters in the text content from English reading comprehension test texts, and extract the sentences corresponding to the emotions expressing reasons, thereby helping teachers in face-to-face When asking reading comprehension questions for a large number of reading materials, quickly find the question point...

AI Technical Summary

Problems solved by technology

However, in the process of modeling paired combinations, due to the lack of interaction between sentences and paired combinations, there are only interactions between various paired combinations, and the context information between paired combinations contains multiple sentences, and the information contained is very Messy, it is difficult to provide causal relationship information for two sentences, so that each pairing combination cannot learn the most important causal relationship information

The most intuitive consequence of the lack of this causal relationship information is the decline in prediction performance, resulting in more incorrect pairing combinations being predicted

[0004] At present, the methods at home and abroad do not model the causal relationship between the emotional sentence and the reason sentence very well, and can only rely on the final fully connected neural network to simply learn

Images