Speaker: Srikanta Bedathur, IBM IRL

Date/Time/Venue: Aug11/12noon/SIT001

Title: Relationship Analysis over Large Graphs

Abstract:  Last years have seen the emergence of large-scale graphs comprising of upto billions of vertices and edges from a number of domains including knowledge graphs, social networks, and the Web graphs. Identifying and characterizing the relationship between vertices is a natural way to understand the structure of these graphs.  In this talk, I will discuss two such relationship analysis problems we have considered: first, to efficiently find the best relationship path between two given vertices; and second, to generate structural explanations for relationships between two sets of vertices. For the first problem, I will present the PathSketches graph indexing  that efficiently estimates the shortest distance and the corresponding path for any two nodes with near-zero error.  Next, I will present our ESPRESSO algorithm that takes two sets of vertices in a semantic graph as input and extracts an explanation of the relationship between these two sets as small subgraphs that represent events (e.g., political scandals, meetings, etc.).

Finally, I will briefly present some of the recent research results from our ongoing work in query processing and mining of semantic graphs and social networks.

Speaker Bio: Srikanta Bedathur is currently a senior researcher in the Knowledge Engineering and Data Platforms group of IBM India Research Labs and an adjunct faculty member at IIIT-Delhi and IIT-Delhi. Before joining IBM in 2014 he worked at IIIT-Delhi as an Assistant Professor where he established and headed the Max-Planck Partner Group on Large-scale Graphs.  Earlier, he was a senior researcher at Max-Planck Institute for Informatics, and simultaneously held the positions ofadjunct junior faculty member at the Saarland University and at the Cluster of Excellence on Multimodal Computing and Interaction. He received his Ph.D. from the Indian Institute of Science in 2005.

His current research interests are broadly in the fusion of ideas from databases, information retrieval and machine learning, and recently with emphasis on large-scale graph management, graph analytics and visualization.