Research On The Method Of Reading Comprehension By Extracting The Facts Of Judgment Documents

The legal system is an important guarantee for the stable and hea lthy development of society,and the promotion of judicial modernization is an inevitable requirement for adapting to social development.Smart justice is one of the important aspects of judicial modernization.The promotion of judicial intelligence can ef fectively promote the fast and efficient operation of judicial work and serve the society.Natural language processing technology is an important technical means of intelligence.The application of deep learning algorithms can learn a large amount of legal document text information,and then intelligently assist judicial work.Case reading comprehension is a kind of natural language processing technology.It can intelligently imitate human beings to read legal documents,and then automatically analyze and q uery relevant information from the current content.Its significance is to measure the ability of algorithms to understand judgment documents,and its value lies in its implementation.It can be applied to question answering system intelligently.Reading j udgment documents through computers and answering related questions are one of the important applications of judicial intelligence.The specific research content is as follows:(1)Proposing a method of case reading comprehension,which combines syntactic guidance and character attention mechanism.Through syntactic analysis,it is possible to analyze the important information expressed in sentences and the potential semantic information of Chinese characters.Considering that the important information in th e question is the key to extracting answers in the process of reading comprehension,this paper uses syntactic analysis to mask the interference information in the question during the self-attention of the question,so as to increase the weight of the impo rtant information.Secondly,on the basis of word embedding,word embedding is carried out on the text,and a new word embedding is formed by the word embedding,and the problem-to-text attention on the coding end is carried out to increase the text coding ability.Experiments show that on the 2019 Law Research Cup reading comprehension data set,the case reading comprehension method that combines syntax guidance and character attention mechanism proposed in this paper has a certain improvement on the baseline model.(2)A method of reading and comprehension of supporting sentences is proposed,which realizes the method of reading and comprehension of supporting sentences while giving answers.Aiming at the problem of poor interpretability extracted from the answers to some questions in the case reading comprehension and the problem of some query question information that needs to be inferred between multiple information in the text,this article adopts the method of multi-task joint learning and combines the answer types And the answer extraction task gives the answer fragments and the supporting sentence of the answer at the same time.First,the three tasks jointly use a Ro Berta encoder for word embedding coding and semantic understanding,and then design different sub-networks for the three tasks to perform their respective target task calculations,and the last three losses are assigned different weights according to the impact size for joint learning.Experiments show that the proposed method has a good improvement on the basis of the baseline model.(3)Build an extractive reading comprehension prototype system for case judgment documents.Integrating the extractive case reading and comprehension method proposed in this article and the case reading and u nderstanding method for identifying supporting sentences,using computers and other smart devices,according to the user's questioning needs of the relevant case,quickly analyze and understand the data,and then calculate and think to make a reasonable Th e answer to the question is then displayed to the user in an interface-based form.