The Center for Data Innovation spoke with Yong-Yeol Ahn, an assistant professor at the Center for Complex Networks and Systems Research, part of the School of Informatics and Computing at Indiana University. Ahn discussed some unusual examples of networks in biology and physics, and how he feels network science is similar to statistics in its broad applicability across other disciplines.
This interview has been lightly edited.
Travis Korte: Can you discuss some of your current research topics?
Yong-Yeol Ahn: I’m interested in network structures around us. For instance, I have been studying the structure of social networks and how the structure affects the diffusion of information and innovation in our society. At the same time, I have been exploring other types of networks in nature and society. One example is the flavor network, which exposes the connections between culinary ingredients.
TK: Most people have some intuitive understanding that their Facebook friends or their Twitter follows form a network, but your work covers networks that we may not always think about as such, like the network of food flavors. Can you give a couple of examples of some other less intuitive networks, and maybe some advice for identifying situations that might have underlying network structure?
YA: You can find networks in many systems, particularly if the systems consist of many elements. The links in networks may represent not only actual interactions but also similarity, co-occurrences, or other more abstract relationships. For food, there was a nice study that examined the co-occurrences of ingredients in an online recipe website. The key question is whether network representation provides insights or not. In many cases, you can build networks that are completely useless.
Some of the unconventional examples that I can think of right now are:
- A word-association network where researchers recorded how people associate words in their mind.
- Social networks of animals.
- Networks of complex molecular structure where transitions from one topological configuration of a molecule to another are recorded.
- Networks in cosmology.
- Networks between prime numbers.
TK: Network science has been around in some capacity for decades, but seems to have really blossomed in the last ten or fifteen years. What do you think spurred this change? Was it really just that more data became available, or is there something more going on?
YA: I think data and computing technologies are the most important driving forces. Most networks that we care about now—online social networks, biological networks, brain networks, etc.—were either born or digitized in the 90’s or later. At the same time, the computing power began to allow large-scale data analysis.
TK: What is the most significant bottleneck to your research into networks? Is it mainly an issue of data quality, technical resources, analytical tools, funding, or something else?
YA: I think the most significant bottlenecks are usually the availability of data (and privacy concerns) and data quality. For instance, although Facebook is recording amazing data on social behavior of people, only a handful of researchers can access the data in house because of the privacy issues. This also raises concerns about the reproducibility of research because the data cannot be released and no researchers outside Facebook can replicate the research done in Facebook (Facebook has claimed that they will create some validation programs to address this issue). Many other interesting datasets (personal genomes, mobile phone traces, etc.) have similar issues. In neuroscience and biology, the major issue seems like the quality of data. For instance, constructing a brain network (either functional or structural) requires lots of assumptions and statistical techniques to detect significant connections. The steps to create a network seem more challenging than the analyses on the network.
TK: Network science is often identified as a fundamentally interdisciplinary undertaking—your training was in physics, for example, but you have worked with biological and social networks. Why do you think this is? Do you think it will always be the case, or do you see the discipline solidifying over time?
YA: It is because one can find networks in so many systems and easily translate and apply methodologies that are developed for a specific domain to others. Moreover, even totally different types of networks often seem to share several universal characteristics and mechanisms of evolution. For instance, we analyzed social, biological, political, psychological, and consumer product networks in a single paper. These factors make network science fundamentally interdisciplinary. But at the same time, providing great insights or useful technologies often require strong domain knowledge and it is often essential for network scientists to collaborate with domain experts.
In some sense, I feel that network science is similar to statistics, providing knowledge and tools to various researchers who work with networks. If we look at the people who can be called “network scientists,” I expect that, as many universal features and methodological frameworks are developed, future network research will require more domain knowledge and thus we will see more network scientists who have strong disciplinary training. Nevertheless, I believe that the network framework will continue to bridge disciplines and spur interdisciplinary research.
Image credit: Lee et al.