Automatic bibliographic component extraction using conditional random fields

Abstract

Bibliographic data and publication data are composed of subfields such as "author," "title," "journal," and "year." Citation analysis of articles in scholarly journals is a very effective method for their evaluation. This paper proposes a system for analyzing bibliographic component strings, which is based on the technique of Conditional Random Fields (CRF). The system is composed of two major modules: the Bibliographic Extraction Module (BEM) and the Statistical Evaluation Module (SEM). The objective of the Bibliographic Extraction Module is to extract the bibliographic components based on the machine learning technique, and the objective of the Statistical Evaluation Module is to turn the extracted bibliographic information into a statistical report. In this paper, we apply the CRF technique to build a probability model for dividing sequential data and giving proper tags to the components according to their characteristics. This is the framework for building the BEM to segment and label bibliographic information, identifying the author`s name, journal`s name, date of publication and so on. Then we employ the SEM to filter and match the intermediate representations produced by the BEM. In the end, the SEM will output the final evaluation report. Experimental results show that our system is reliable, with excellent overall efficiency.