A system for disease-specific knowledge integration

Summary

Since the dawn of the Information Age, man struggled to bring order to the ever increasing amount of Data. Focusing on the Biomedical domain, the unyielding accumulation of literature in combination with the abundance of structured resources make the integration and navigation through existing knowledge a very challenging task. Developing and expanding on previous research, we propose an end-to-end system for integrating Disease-specific knowledge - from literature and structured resources - on a semantic Knowledge Graph. The proposed system provides an automated solution for knowledge integration accompanied by a user-friendly graphical interface for management and exploration of the produced Knowledge Graph. Furthermore, to enhance the user experience, we train and evaluate multiple ranking Machine Learning models on data from our graph, intending to select a model that will rank the entities on our graph based on their usefulness. The highest scoring model exhibits good performance in a task that has proved challenging even for human annotators. Closing our work, we discuss our findings and contemplate on future enhancements to the proposed system.Since the dawn of the Information Age, man struggled to bring order to the ever increasing amount of Data. Focusing on the Biomedical domain, the unyielding accumulation of literature in combination with the abundance of structured resources make the integration and navigation through existing knowledge a very challenging task. Developing and expanding on previous research, we propose an end-to-end system for integrating Disease-specific knowledge - from literature and structured resources - on a semantic Knowledge Graph. The proposed system provides an automated solution for knowledge integration accompanied by a user-friendly graphical interface for management and exploration of the produced Knowledge Graph. Furthermore, to enhance the user experience, we train and evaluate multiple ranking Machine Learning models on data from our graph, intending to select a model that will rank the entities on our graph based on their usefulness. The highest scoring model exhibits good performance in a task that has proved challenging even for human annotators. Closing our work, we discuss our findings and contemplate on future enhancements to the proposed system.