Multi-task joint learning-based discussion mining system and working method thereof

Abstract

The invention relates to a multi-task joint learning-based discipline mining system and a working method thereof. The system comprises a data preprocessing module which carries out the data preprocessing; the text embedding module is used for extracting feature representation from the word level and the character level by a CNN (Convolutional Neural Network), and combining task specific features to serve as vector input of the next module; the joint learning module adopts a hard sharing mechanism of parameters in multi-task learning, a plurality of tasks share a hidden layer of one model, training learning of the model is carried out in parallel, and in addition, a stack type bidirectional long-short-term memory (LSTM) neural network is used for learning context information of a text to complete a sequence labeling task; and the discussion mining label prediction output module is used for completing discussion mining discussion point type prediction work and outputting the discussion point type in the text. According to the method, high-quality text features can be learned from the discussion text data, and finally the type of discussion points contained in the text is detected.

Description

technical field [0001] The invention relates to the field of natural language processing, in particular to an argument mining system based on multi-task joint learning and its working method. Background technique [0002] Currently, there are many methods that can be applied to the argument mining task. In the early work, most of the research only started from a single subtask of argumentation mining, and modeled and solved the task for a single subtask, while ignoring the correlation information between the three subtasks, resulting in low system performance. [0003] In addition, some works use the pipeline model to jointly model the three subtasks to join the connection of the three subtasks. The pipeline method has the problem of error propagation because the error in the identification of the argument type will affect the extraction error of the argument relationship. In addition, this method pairs the identified arguments in pairs, and then classifies the argument re...

AI Technical Summary

Problems solved by technology

However, this method fails to fully consider the diversity of label samples in the debate mining corpus. It can only solve some datasets with sufficient data and complete labels, and cannot solve the problem of data sparsity, which leads to poor generalization performance of the model and is difficult to solve. applied to a new dataset

Images