Resume Information Extraction with Cascaded Hybrid Model

Proceedings of the 43rd Annual Meeting of the ACL, pages 499–506, Ann Arbor, June 2005. c⃝2005 Association for Computational Linguistics Resume Information Extraction with Cascaded Hybrid Model Kun Yu Gang Guan Ming Zhou Department of Computer Science and Technology Department of Electronic Engineering Microsoft Research Asia University of Science and Technology of China Tsinghua University 5F Sigma Center, No.49 Zhichun Road, Haidian Hefei, Anhui, China, 230027 Bejing, China, 100084 Bejing, China, 100080 yukun@mail.ustc.edu.cn guangang@tsinghua.org.cn mingzhou@microsoft.com Abstract This paper presents an effective approach for resume information extraction to support automatic resume management and routing. A cascaded information extraction (IE) framework is designed. In the first pass, a resume is segmented into a consecutive blocks attached with labels indicating the information types. Then in the second pass, the detailed information, such as Name and Address, are identified in certain blocks (e.g. blocks labelled with Personal Information), instead of searching globally in the entire resume. The most appropriate model is selected through experiments for each IE task in different passes. The experimental results show that this cascaded hybrid model achieves better F-score than flat models that do not apply the hierarchical structure of resumes. It also shows that applying different IE models in different passes according to the contextual structure is effective. 1 Introduction Big enterprises and head-hunters receive hundreds of resumes from job applicants every day. Automatically extracting structured information from resumes of different styles and formats is needed to support the automatic construction of database, searching and resume routing. The definition of resume information fields varies in different applications. Normally, resume information is described as a hierarchical structure The research was carried out in Microsoft Research Asia. with two layers. The first layer is composed of consecutive general information blocks such as Personal Information, Education etc. Then within each general information block, detailed information pieces can be found, e.g., in Personal Information block, detailed information such as Name, Address, Email etc. can be further extracted. Info Hierarchy Info Type (Label) General Info Personal Information(G1); Education(G2); Research Experience(G3); Award(G4); Activity(G5); Interests(G6); Skill(G7) Personal Detailed Info (Personal Information) Name(P1); Gender(P2); Birthday(P3); Address(P4); Zip code(P5); Phone(P6); Mobile(P7); Email(P8); Registered Residence(P9); Marriage(P10); Residence(P11); Graduation School(P12); Degree(P13); Major(P14) Detailed Info Educational Detailed Info (Education) Graduation School(D1); Degree(D2); Major(D3); Department(D4) Table 1. Predefined information types. Based on the requirements of an ongoing recruitment management system which incorporates database construction with IE technologies and resume recommendation (routing), as shown in Table 1, 7 general information fields are defined. Then, for Personal Information, 14 detailed information fields are designed; for Education, 4 detailed information fields are designed. The IE task, as exemplified in Figure 1, includes segmenting a resume into consecutive blocks labelled with general information types, and further extracting the detailed information such as Name and Address from certain blocks. Extracting information from resumes with high precision and recall is not an easy task. In spite of 499 Figure 1. Example of a resume and the extracted information. constituting a restricted domain, resumes can be written in multitude …

Content truncated to 500 words