Literature mining; A REAL-time WEB-based text mining application

Objective: The aim of this study is to develop a web based literature mining system which retrieves Pubmed abstracts to provide tools for information search and evaluation needs of healthcare professionals and researchers in their research and clinical routines. Method: System development process includes retrieving abstracts from Pubmed literature database, text preprocessing by using text mining techniques, annotating and extracting medical entities, aim sentences and statistical methods of studies, and presenting the results through the web interfaces. In order to retrieve abstracts from Pubmed, a library called BioPython has been used. Becas annotator has been prefered to annotate the medical entities like disease, gene and protein, drug etc. A new algorithm based on dictionary-based method was developed to extract aim sentence of studies. Frequency distribution has been calculated to discover relationship between the tagged entities. Results: The system tags entities in different color in accordance with their classes and presents the results in a similar interface with Pubmed. It automatically extracts aim of a study and statistical terms used in a study and it demonstrates the results in a different interface with tabular format along with several features of article and the tagged medical entities. Based on the selected entity class by user, co-occurrence frequency of entities are calculated and presented in a table format and visualized with a bar chart. The aim extraction module achieved 83.5% recall, 95% precision and 90% f-measure and statistical term extraction module achieved 95.4% precision, 88.3% recall ve 91.7% f-measure in partial evaluation, 94.1% precision, 67.8% recall and 78.8% f-measure in exact evaluation. Conclusion: The system provides a web-based platform for mining medical information from Pubmed and it is unique in that it (i) extracts a wide range of entity classes; (ii) allows users to rapid review the results with different interfaces; and (iii) extracts not only binary relation but also relation between more than two entity types with multiple selection choices.