Influence Search

There is a vast amount of scientific findings scattered across the literature. Too many insights from research have gone undiscovered because fragments of independent but related knowledge have not been systematically retrieved, brought together, and interpreted. Influence Search provides a way to navigate the sea of research and assemble these puzzle pieces to help Ki understand both the scope of existing discoveries, and identify gaps in research that still need to be addressed.

Intended use

This tool is intended to be used by subject matter experts to explore and construct influence models in the domain of child growth and development. A second intended use is for decision-makers to understand opportunities for future investments to improve children’s health, both generally and in ways particular to different geographic settings.

Temporal timelines and intervention pathways

The tool can be used to search for influence pathways and aggregate them to into influence models that address various aspects of child growth and development. Thirdly, a collaboration between subject matter experts and decision-makers led to this tool being leveraged to construct a conceptual model of risk factors and mechanisms connecting infant growth restriction to later obesity.

The three main benefits of Influence Search

The Challenge

A key challenge for Ki is the acceleration of constructing multidisciplinary influence models, which, in turn, impact the systems understanding of child growth and development, and enable future research in this domain. With the aggregation of vast amounts of information scattered across multiple domains, technical engineering can be incorporated to auto-consume and categorize.  . For example, PubMed alone indexes more than 1 million publications per year. At this scale, assistance from computers through machine reading, can help researchers keep pace with current information.

A machine reading system that has the capacity to extract and aggregate influence relations at scale was created by M. Surdeanu from the Computational Language Understanding (CLU) Lab at University of Arizona. In the early  version of the tool,  approximately 115 thousand open-access publications were processed to construct a searchable database of over two million concepts relevant to children health connected by more than 2.5 million influence relations.

The database could be searched for direct and indirect influence interrelations (e.g., how does the bacteria campylobacter indirectly influence malnutrition?). Subject matter experts could assemble the results of these searches into explicit visualizations of influence models that could be edited and shared.