Visualizing Food-Disease-Gene Networks
Oral Defence Date:
Professors Hui Yang, Ilmi Yoon & Dragutin Petkovic
We have previously proposed an information extraction system to extract and analyze relationships between food, diseases and genes to construct food-centric networks. Modules for extracting relationships and analyzing relationship strength/polarity have already been implemented. Hierarchical relationships (e.g. is a, such_as) are required to integrate entities and to help in creating the visualization and making it more readable. We use the Stanford parser to identify dependencies in a sentence which leads to patterns to identify hierarchical relationships. This method has an F-measure of 0.705 and is able to identify five types of hierarchical relationships, (1) is-a (2) such-as (3) component-of (4) containing (5) including from the abstracts. We next utilize the hierarchical relationships to create a hierarchical tree structure, which allows the user to explore the resulting food-centric networks at different levels of information granularity. We use the prefuse toolkit to create a Force Directed Layout (FDL) to visualize the network. The entities and relationships are color coded to make it easier for the user to differentiate between entities and relationships. Our user study compares this system with two other similar systems using NASA TLX and scored high in interpretability and performance. It also scored the lowest in mental and temporal demand required.
Hierarchical relationships, typed dependencies, biomedical text mining, stanford typed dependency parser, prefuse toolkit