log in  |  register  |  feedback?  |  help  |  web accessibility
Logo
resMBS: Information Extraction, Topic Modeling and Tensor Factors for Financial Contracts
Zheng Xu - University of Maryland
Wednesday, November 2, 2016, 11:00 am-12:00 pm Calendar
  • You are subscribed to this talk through .
  • You are watching this talk through .
  • You are subscribed to this talk. (unsubscribe, watch)
  • You are watching this talk. (unwatch, subscribe)
  • You are not subscribed to this talk. (watch, subscribe)
Abstract

This talk summarizes recent work on information extraction to create graph datasets from financial documents and further analysis. resMBS is a bipartite graph dataset respresenting a financial supply chain where financial institutions (FIs) play a Role (labeled edge) on financial contracts (FCs). resMBS was created using a domain-specific rule-based extractor based on the IBM SystemT platform. We apply and extend two unsupervised learning techniques, topic modeling and tensor decomposition to analyze the resMBS graph. One contribution is to efficiently solve a non-negative factorization of an occurrence
tensor (discrete values). We present some preliminary results. This is joint work with Louiqa Raschid. Multiple collaborators from UMD and IBM Almaden contributed to creating the resMBS dataset.

Bio

Zheng is a third year PhD student in the Department of Computer Science, University of Maryland, College Park. He works on optimization, machine learning and their applications. His recentl work automates a specific primal-dual optimization method, ADMM, by adaptively choosing the only free parameter in the algorithm. Before joining UMD, he worked on computer vision as a Project Officer at Nanyang Technological University, Singapore. He received his M.Eng. and B.Eng. from the University of Science and Technology of China, and collaborated with researchers from Microsoft Research Asia on large scale multimedia data.

This talk is organized by Naomi Feldman