Automated Identification of Potential Conflict-of-Interest in Online Biomedical Articles Using Hybrid Deep Learning Neural Network
Brown Bag Lecture by Dr. Incheol Kim | 6/6/2017 11AM – 12PM | 7th Floor Conference Room, Bldg 38A
Abstract: Financial conflict-of-interest (FCOI) in biomedical research may cause a number of potential ethical risks, including an increased possibility of pro-pharmaceutical industry conclusions, restrictions on the behavior of the investigators, and the use of biased study designs. In order to ensure the impartiality and objectivity of research, many journal publishers require authors to fully disclose any financial supports and also provide a COI statement within the body text of their articles at the time of peer-review and publication, thereby letting reviewers and readers easily know the integrity of research. However, author’s self-reported COI disclosure often does not explicitly appear in their article, and may not be very accurate or reliable due to the lack of understanding of relatedness between a certain financial gain they received and their current research. In this study, we designed and developed a two-stage machine learning scheme using a hybrid deep learning neural network (HDNN) that tightly combines a multi-channel convolutional neural network (CNN) and a feed forward neural network (FNN) , to identify a potential COI in online biomedical articles. It first distinguishes “support” sentences from the last page of the body text of a given biomedical article (stage 1), and then classifies “support” sentences into two classes according to their funding sources: “for-profit” or “non-profit” (stage 2). In a series of classification experiments, the proposed HDNN is evaluated by comparing its performance with that of other types of classifiers such as single/multi-channel CNNs, support vector machine (SVM), and a voting scheme.
Bio: Dr. Incheol Kim is a Senior System Analyst at CEB since 2004. He has a Ph.D. degree in Information Processing Engineering from the Kyungpook National University, South Korea. His previous experience includes two years as a postdoctoral researcher at Concordia University and five years as a senior system engineer in an industrial research lab. His research interests are Web-based document analysis and processing, pattern recognition and classification, text data mining, neural networks, and statistical learning methods.