A joint defense mining system and method based on a dependency parsing tree

Abstract

The invention relates to a joint defense mining system based on a dependency parsing tree, and the system comprises a data preprocessing module which is used for carrying out the preprocessing of data; the text embedding module is used for extracting words, characters, part-of-speech, dependency relationships among the theses and vector representation of the types of the theses from the input text; the sequence encoding module is used for learning context information of the text by using a bidirectional long-short-term memory neural network to finish a task of thesis point boundary detection and thesis point relation extraction; the dependency parsing tree module is used for searching the shortest path in the two thesis point component entities by constructing a dependency parsing tree; and the disguising mining label output module is used for completing label prediction work of three disguising mining tasks, type labels of disguising points and relation labels of the disguising points. According to the method, high-quality text vector characteristics can be learned from the defibration text data, and finally the defibration structure of the text is detected.

Description

technical field [0001] The invention relates to the field of natural language processing, in particular to a system and method for joint debate mining based on dependency parsing trees. Background technique [0002] Currently, there are many technical methods available for argument mining. Traditional argument mining methods mainly model the subtasks independently, while ignoring the correlation information between the three subtasks, resulting in poor performance. In addition, some works adopt the pipeline model to jointly model the three subtasks, and these models have the problem of error propagation during the training process. [0003] Currently, there are some pipeline-based research methods. The basic idea is to use the pipeline method for the three sub-tasks of debate mining, and solve them in the order of the pipeline. The pipeline method has the problem of error propagation because the error in the identification of the argument type will affect the extraction e...

AI Technical Summary

Problems solved by technology

[0004] However, current debate mining research methods often ignore the correlation information between subtasks, and also ignore the different characteristics of each subtask, and correlation information is of great significance to debate mining. The label prediction result of a task can be used as a prediction Efficient features for other debate mining subtask labels

Images