Invited Talk by Douglas Downey

Title: Mining Topics and Key Phrases from Scientific Documents

Douglas Downey

Abstract:Methods that can automatically discover topics from documents could help scholars find relevant scientific articles. However, existing methods for mining topics struggle to scale to the massive number of distinct topics found in the scientific literature. “Flat” topic models like Latent Dirichlet Allocation (LDA) have difficulty modeling sparsely expressed topics, and richer hierarchical models become computationally intractable as the number of topics increases. In this talk, I will describe an efficient method for inferring large topic hierarchies. The method is based on the Sparse Backoff Tree (SBT), which organizes the discovered topics as leaves in a tree with nearby leaves representing related topics. I will show how the model can scale to models of over a million topics, and describe experiments showing that SBT is more accurate than flat LDA. Lastly, I will talk about our recent efforts to mine “key phrases” to succinctly describe topics, and the application of the work within the Semantic Scholar scientific search engine.

Speaker: Doug Downey is an associate professor in the Electrical Engineering and Computer Science department at Northwestern University. His research focuses on natural language processing, machine learning, and artificial intelligence, with an emphasis on probabilistic models of Web content and their applications. Recent prototypes he has contributed to include www.semanticscholar.com and www.atlasify.com.

Invited Talk by Alex Wade

Title: Academic Knowledge: new research opportunities with the Microsoft Academic Graph

Alex Wade

Abstract: The Microsoft Academic Graph (MAG) is a freely available dataset that includes information about academic publications and citations, researchers, venues, and topics. The MAG is a heterogeneous graph that can be used to study the influential nodes of various types, including authors, affiliations, and venues. In 2015, Microsoft Research published this dataset, and organized the 2016 WSDM Cup challenge focused on new methods to ranking nodes in such a heterogeneous graph. More recently, the Microsoft Research launched Project Oxford (https://www.projectoxford.ai) a set of artificial intelligence based APIs, which now includes the Academic Knowledge API, a queryable REST interface over the ever-growing MAG dataset. This talk will cover the MAG data, how it is being surfaced in Bing and Cortana, and how researchers can leverage this data in their own Big Scholarly Data research.

Speaker: Alex Wade is Director for Scholarly Communication at Microsoft Research, and is currently responsible for Microsoft Academic, a rich entity-based discovery and navigation service centered on researchers, topics, conferences, and publications. During his career at Microsoft, Alex has managed Microsoft’s internal corporate search and taxonomy management services, and has served as Program Manager for Windows Search for multiple Windows OS releases. Prior to joining Microsoft, Alex has worked for the library systems at the University of Washington, the University of Michigan, and the University of California at Berkeley. Alex holds a Bachelor's degree in Philosophy from U.C. Berkeley, and a Masters of Librarianship degree from the University of Washington.