Maximizing the Spread of Influence through a Social Network Kempe, Kleinberg, Tardos 2003 Ellie Clougherty Goal: Which subset of individuals are the most influential? How can we select a subset of nodes that will have the greatest impact? Name: Ellie Clougherty Can you find degree centrality? Can you find distance centrality? Network Data Co-authorships of high-energy physics papers Node: Author Edges: Degree of co-authorship Goal: Select the most influential nodes An NP-Hard optimization problem! Solution: Naturally-greedy, hill-climbing, submodular algorithm obtains 63% of the optimal solution Two Models Weight: Number of Edges / Co-Authorships Linear Threshold Model A node becomes active when the total weight of it’s active neighbors exceeds a random threshold Independent Cascade Model An activated node has one chance to activate an inactive neighbor with a weighted probability Context Matters Consider the dynamics of information in a network, not just the structure Weighting the influence of one node to another depends on the context and nature of the data Future Steps: Generalized Framework: combine both models Non-progressive models: activation is bi- directional Additional Sources Node Grid:http://en.wikipedia.org/wiki/Centrality#mediaviewer/File:Centrality.svg Indian Grid:http://www.mnn.com/earth-matters/energy/stories/whats-wrong-withindias-electric-grid John Kelly Protesting: http://www.npr.org/2014/10/01/352757107/campus-sexualassault-law-now-includes-language-on-same-sex-violence UVA Protestors: http://www.nytimes.com/2014/11/25/us/rocked-by-rape-reportuniversity-of-virginia-to-hold-special-meeting.html?smid=fbnytimes&smtyp=cur&bicmp=AD&bicmlukp=WT.mc_id&bicmst=1409232722000&bic met=1419773522000 EM Slide: http://sifaka.cs.uiuc.edu/~wang296/Course/IR_Fall/docs/PDFs/Language%20Mod els%20II.pdf Truthy: http://truthy.indiana.edu/ Business Insider: http://www.businessinsider.com/what-makes-an-article-go-viral-20125