Dispute focus category and similarity judgment method, system and device and recommendation method

Abstract

The invention discloses a dispute focus category and similarity judgment method, system and device and a recommendation method, and belongs to the field of natural language processing. The dispute focus category and similarity judgment method comprises data set construction, model construction and training. According to the invention, a recall model and a refined arrangement model are used in the whole architecture; the recall model directly classifies dispute focuses, the accuracy is limited, but the speed is high, so the overall efficiency is guaranteed; the refined model makes full use of a Self-attention mechanism of bert to perform semantic interaction on dispute focuses, and although the speed is slightly slow, the accuracy is high; therefore, the speed and the accuracy are effectively balanced by the recall-refined dual model.

Description

technical field [0001] The present invention relates to the field of natural language processing, in particular, to a method, system and device for judging the category of the focus of dispute and similarity, and a recommendation method. Background technique [0002] At present, the main methods of judging the similarity of sentences that are the focus of disputes in the judicial field are as follows: [0003] 1) Transform the sentence similarity problem, which is the focus of controversy, into a multi-classification task. This method faces the problem of small-sample learning. Usually, the categories marked by each case are greater than 100, and the number of samples in each category is less than 10, which will easily lead to the accuracy of the classification results. Low. [0004] 2) Using semantic similarity matching technology, in order to better judge whether the semantic similarity of two controversial sentences is usually an interactive model, the accuracy rate of t...

AI Technical Summary

Problems solved by technology

[0003] 1) Transform the sentence similarity problem, which is the focus of controversy, into a multi-classification task. This method faces the problem of small-sample learning. Usually, the categories marked by each case are greater than 100, and the number of samples in each category is less than 10, which will easily lead to the accuracy of the classification results. Low

[0004] 2) Using semantic similarity matching technology, in order to better judge whether the semantic similarity of two controversial sentences is usually an interactive model, the accuracy rate of this method is generally higher than that of the first method , but the main problem of this method is that the query focus sentence needs to interact with each marked dispute focus sentence during inference, and then determine the classification or extract similar dispute focus sentences, which will be less efficient

Images