Research And Implementation Of Multi-field And Multi-range Entity Recognition Based On Bi-LSTM

Named entity recognition is the basis for the application of natural language processing tasks.For the natural language processing tasks in the fields of network security,law,and medicine,such as information extraction in the field of network offense and defense,judge judgment,medical diagnosis and diagnosis,etc.,entity recognition is particularly accurate important.Entity recognition errors in these fields are costly.Entity recognition is a critical and basic task in natural language processing in these specific fields.The purpose of entity recognition is to determine the semantic category of the entities in the text in a specific text.In the entity recognition task,there are complex entities with less contextual effective information and longer entity word length,such as "cross-site scripting attack",a network attack technology entity,and the existing deep learning combined with machine learning methods may only recognize "scripts Attack ”without being able to identify the entity completely.Faced with the problems of misidentification and missing recognition in the recognition of complex entities in these special fields,this paper proposes a method model based on bidirectional long-short-term memory network combined with conditional random field and complex entity library matching(BiLSTM-CRF-CEL)improves the accuracy of complex entity recognition.In this model,Bi-LSTM first extracts the semantic vector features of a text sentence,and then classifies and classifies the entities according to the context semantic information of the entities in the text.Then access the CRF after Bi-LSTM to label the labels based on the dependencies between the labels.At the same time,for the recognition process,a long-term word is difficult to identify,there is ambiguity and other complex entities to establish a complex entity library,and a complex labeling recognition model for complex entities.The model is based on deep learning and has good portability.The model is trained based on corpora in different fields,making the model applicable in multiple fields and in many areas.In order to verify the effectiveness of the BiLSTM-CRF-CEL model,on the one hand,an experimental comparison is made by controlling the number of complex entities in themodel complex entity library and the size of the word vector dimension;on the other hand,Bi-LSTM and CRF are used to fuse the complex entity matching model with other Experiment comparison of commonly used entity recognition models.The experimental results show that the model proposed in this paper helps to improve the entity recognition rate.