Design of a Biomedical Natural Language Processing System for Mining Scientific Articles

Abstract

Natural language processing has an important role in empirical life science research as it involves extracting information from various domains encompassing several levels of abstractions. In neuroscience, information at different translational levels relating to atomic, molecular, cellular, circuitry and behavior are pertinent to understand brain structure and function. Neuroinformatics based approaches employing biomedical natural language processing (BioNLP) and related data mining methods have enabled seamless integration of sub-molecular, cellular to physiology and behavior data in neuroscience, to explore non-trivial structure-function relationship in the central nervous system. In the recent years, there has been an exponential increase in the size of scientific literature repositories, which are sources for relevant information, leading newlineto increased complexity in information retrieval. This thesis proposes and involves the design and development of the ABioNLP platform, to reduce this complexity and identify relationships among the research articles in PubMed based on their relevance, correlating multiple levels of abstraction within underlying scientific domains. newlineInformation retrieval methods for PubMed do not necessarily provide enhanced insight into data inherent relations based on content similarity. Using scientific document datasets, evaluation of classification and clustering algorithms was performed. Clustering algorithms showed similar accuracy as in the case of classification, albeit a prior information was not needed during clustering. In this context, document clustering enabled finding relevant documents that were grouped under unique cluster labels. newlineABioNLP employs querying, document clustering, cluster label validation, and newlinevisualization methods packaged as a platform with the objective to generate newlineinterconnections amongst similar documents within the queried literature...