Classical Chinese machine reading understanding method based on multi-task joint training

Abstract

The invention provides a classical Chinese machine reading understanding method based on multi-task joint training, which comprises the following steps of: establishing a classical Chinese machine reading understanding model based on multi-task joint training, and then performing multi-task joint training on classical Chinese; the classical Chinese machine reading understanding model comprises a context extraction module, a multi-model coding module, a bidirectional matching module, a sentence segmentation module and an answer classification module, wherein extracting classical Chinese by using a context extraction module, and inputting a result into a multi-model coding module for processing; coding the classical Chinese by using a multi-model coding module to obtain vectorized representation of the classical Chinese; fusing by using a door mechanism; and inputting an obtained result into the bidirectional matching module and the sentence segmentation module for processing, and processing an output result of the bidirectional matching module and the sentence segmentation module by using the answer classification module to obtain a final result. The classical Chinese sentence segmentation method can accurately perform sentence segmentation processing on the classical Chinese, can consider the condition that the ancient Chinese and the modern Chinese exist at the same time, and can more accurately process the classical Chinese.

Description

technical field [0001] The invention specifically relates to a machine reading comprehension method for classical Chinese based on multi-task joint training. Background technique [0002] Machine reading comprehension is an important task in natural language processing. Through models and algorithms, machines can understand human language and answer corresponding questions. The single-choice task is a sub-task in the machine reading comprehension task. Given the context, questions and options, the best answer is selected from multiple options. The common framework for existing machine reading comprehension models on single-choice tasks is to encode the context, questions, and options to obtain their vectorized representations, then interact with the context, questions, and options, and finally pass the interaction results of the context, questions, and options Classification is performed and the option with the highest probability is selected as the result. [0003] Existi...

AI Technical Summary

Problems solved by technology

In addition, when using machine reading applications in classical Chinese single-choice topics, there are often modern texts or options in the form of topics in the reading comprehension of classical Chinese texts; existing technologies that only use classical Chinese text models cannot take both ancient texts and modern texts into consideration. At the same time, the vectorized representation obtained in the text encoding is of poor quality

Images