A method for obtaining rich data from PubMed using SVM

Junbum Cha, Jeongwoo Kim, Yunku Yeu, Sanghyun Park

Abstract: As text mining advances rapidly in the biomedical field, the importance of text data is increasing. Most text data is obtained through a Medical Subjects Headings (MeSH) term search; in this process, a large amount of valuable data is missed because the data is not indexed yet with MeSH terms. In this paper, we propose a method for obtaining additional text data in addition to that obtained using a conventional MeSH term search. In order to obtain additional data, we used the Support Vector Machine (SVM) as the data mining method for classifying documents to related or unrelated. We evaluated the results using a frequency-based text mining approach measuring the quality of data in study of lung cancer. This was confirmed that the data extracted using our method provided as much valuable information as searching using MeSH terms. Further, we found that the amount of information found was increased by 40% using additional extracted data.