Ideas worth spreading: How does network position influence the - PowerPoint PPT Presentation

Ideas worth spreading: How does network position influence the spread of research topics? Allison Morgan University of Colorado, Boulder with Dimitrios Economou, Samuel Way, Aaron Clauset

How do research ideas spread within a discipline? How does network position affect the spread of new ideas?

Prestige and Publication Sociology of Education, 375-397 (1971) American Journal of Sociology 76(2), 286-306 (1970) Am. Soc. Rev. 55, 469-478 (1990)

Prestigious universities are better connected Slope: 0.0163 5.0 4.5 Average Path Length ( 〈 l 〉 ) 4.0 3.5 3.0 MIT 2.5 New Mexico State University 2.0 University of Colorado, 1.5 Boulder 1.0 20 40 60 80 100 120 140 160 Universities Sorted by Prestige ( π ) Increasing Prestige

Methods How much does an idea's origination location matter for how far it spreads? First, we need to know how ideas spread. Then, we can use simulation to measure the consequences of prestige.

Data Publication records Faculty hiring network MIT Cornell Caltech UC Berkeley Harvard Stanford Washington Princeton Carnegie Mellon Yale Science Advances 1(1), e1400005, 2015. Proc. 25th Int'l World Wide Web Conf. (WWW), ( 2016 )

Data Faculty hiring network Publication records. Education and employment history for faculty at 205 U.S. MIT and Canadian CS depts. Cornell Caltech UC Berkeley Harvard Node represents an u institution with unique prestige. Stanford Washington Edge represents a PhD ( u, v ) Princeton Carnegie Mellon Yale candidate from who got an u assistant faculty position at . Proc. 25th Int'l World Wide Web Conf. (WWW), v Science Advances 1(1), e1400005 (2015) 2016.

Data Publication records Publication records for 2659 tenure-track faculty. Includes title, author list, venue, and date for each pub. Use first assistant professorship start dates to identify faculty who started a research area at their university. Proc. 25th Int'l World Wide Web Conf. (WWW), ( 2016 )

Do faculty bring research with them? { employment University A infection due relevant to hiring publication { infection not due to hiring University B 1985 1990 1995 2000 2005 2010 Comics by Jorge Cham: http://phdcomics.com/

Research ideas spread across hiring Compare real hiring infection rate to a null model where faculty randomly choose research topics. “topic modeling”: p = 0 . 01 ± 0 . 01 “incremental computing”: p = 0 . 01 ± 0 . 01 “deep learning”: p = 0 . 2 ± 0 . 01 Faculty hiring network shapes spread of ideas. 1985 1990 1995 2000 2005 2010

What role might prestige play in the spread of new ideas? University of Seed an epidemic at a single New Mexico CU Boulder university with unique prestige. Simulate an SI epidemic on the network. Examine resulting epidemic size as a function of prestige (network Stanford location) and quality of idea. UC Santa Cruz Slope: 0.0163 5.0 4.5 Average Path Length ( 〈 l 〉 ) 4.0 MIT 3.5 3.0 MIT 2.5 New Mexico State University 2.0 University of Colorado, 1.5 Boulder 1.0 20 40 60 80 100 120 140 160 Universities Sorted by Prestige ( π )

Network position affects epidemic size 1.0 Fraction of Network Infected Infection Rate 0.8 (“Idea Quality”) p Epidemic Size 0.1 0.10 0.6 0.3 0.30 0.5 0.50 0.4 0.7 0.70 0.9 0.90 0.2 0.0 0 50 100 150 University Prestige ( π ) Increasing Prestige

Network position affects epidemic size 1.0 Fraction of Network Infected Infection Rate 0.8 (“Idea Quality”) p Epidemic Size For a fixed infection rate 0.1, 0.1 0.10 0.6 0.3 0.30 increasing prestige by 10 units ( ∆ π = +10) 0.5 0.50 infects 7.5% more of the whole network. 0.4 0.7 0.70 0.9 0.90 0.2 0.0 0 50 100 150 University Prestige ( π ) Increasing Prestige

Conclusions Researchers appear to carry some research ideas from PhD to first assistant professorship. Under a model where ideas entirely spread via hiring, higher prestige universities have large influence.

Future Work Better approximate the spread of research by obtaining full text of publications. Explore more sophisticated models of idea- spreading.

1.0 Fraction of Network Infected Thanks! Infection Rate 0.8 (“Idea Quality”) p Epidemic Size 0.1 0.10 0.6 Collaborators: Dimitrios Economou, 0.3 0.30 Samuel Way, Aaron Clauset 0.5 0.50 0.4 0.7 0.70 0.9 0.90 0.2 @alliecmorgan allisonmorgan.github.io allison.morgan@colorado.edu 0.0 0 50 100 150 University Prestige ( π ) Increasing Prestige