Named entity identification method and device, equipment and medium

Abstract

The invention discloses a named entity recognition method, a named entity recognition device, equipment and a medium. The named entity recognition method comprises the steps of: inputting a target sentence into a current named entity recognition model, and acquiring an annotation type and a corresponding probability of each character; calculating a corresponding loss value based on a target loss function according to the corresponding probability of the annotation type of each character, calculating a total loss value according to the current loss values, and improving the attention of the named entity recognition model to the characters with high annotation type uncertainty by using the target loss function; judging whether the current named entity recognition model meets a convergence condition or not; if not, selecting the model with the optimal current effect as a target model to perform entity identification on the target sentence; and if so, updating the named entity recognitionmodel according to the total loss value to obtain an updated named entity recognition model, circulating the previous steps until the judging that the named entity recognition model meets a convergence condition, thereby being capable of increasing the accuracy rate of named entity recognition.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a named entity recognition method, device, equipment and medium. Background technique [0002] Named entity recognition refers to the identification of entities with specific meaning in text, mainly including names of people, places, institutions, proper nouns, etc. In order to obtain more accurate results on the task of named body recognition, that is, to identify possible named entities (such as person names, place names, organization names, etc.) in a sentence, the traditional method maps each word in the vocabulary to a Word vectors, and encode the word vectors in the entire sentence through some encoders, and finally predict the label to which each word belongs. [0003] However, in a sentence, the characters occupied by entities only account for a small number of all characters, most of the characters need to be marked as "negative samples" by the model, and ther...

AI Technical Summary

Problems solved by technology

[0003] However, in a sentence, the characters occupied by entities only account for a small number of all characters, most of the characters need to be marked as "negative samples" by the model, and there are many easy samples in the negative samples, so the learning of the model will be Most of them are easily controlled by negative samples, resulting in poor entity recognition

Images