Research On Resume Event Extraction On Deep Learning

The electronic recruitment resumes received by major companies and the electronic announcements issued by financial institutions contain a large amount of personnel resume information,which is presented in a semi-structured or unstructured form.Resume event extraction can extract the event information contained in the resume text,and store it in a structured manner,and then explore the complex relationships between personnel,build a character relationship graph,for the construction of talent pools,financial institutions' investment,decision-making and Development is of great significance.In view of the special problems in the resume text,the resume event extraction is divided into three stages: event type recognition,event element extraction and event separation.A BERT-TC(BERT based Token Classification)model based on token classification is proposed to identify event types,classify the event types as tags,identify event types in the text,and obtain the location information of the tokens corresponding to each event type.The BERT-TPTC(BERT based Two-Phased Token Classification)model and the MRC-MAN(Machine Reading Comprehension with Max Argument Number)model are proposed to extract elements.The BERT-TPTC model regards event element extraction as a two-stage token classification task Integrate the location information of the tokens corresponding to each event type to improve the performance of the model;the MRC-MAN model solves the uncertain number of event roles in the resume text by predicting the maximum number of each event role in the text.Combining the structural characteristics and heuristic rules of the resume text,an event separation algorithm is designed to separate different events of the same type to obtain a structured event list.The design experiment was tested and analyzed on the manually annotated resume data set.In the event type recognition experiment,the F1 value reached 0.9861,which was an increase of 7% compared with the sentence classification method.It can accurately identify the events in the resume.Type;the F1 value of the event element extraction experiment reaches 0.9571,which is 0.5% higher than the BERT method;the results of the ablation experiment show that the proposed MRC-MAN model effectively improves the performance of the MRC model in event element extraction,and can effectively solve the resume There are multiple problems with the same event in the text,which verifies that the proposed model and method can effectively solve the problem of resume event extraction.