Login
Computer Science @ Rochester
Monday, October 08, 2012
10:45 AM
Computer Studies Building, Room 209
Daisy Zhe Wang
University of Florida
Probabilistic Knowledge Base Systems over Large scale Unstructured Text
In the past 10 years, data has gone from scarce to superabundant. Much of this data is unstructured text data from the Web, emails, social media, digitized documents, and transcribed speeches. The state-of-the-art tool to explore text data is a keyword search engine, which uses string-matching techniques with little understanding that keywords and text documents refer to entities and relations in real-life. This lack of knowledge can and often does lead to poor search results. A knowledge base (KB) containing extracted entities and relations should be the backbone of any application that is fueled by text. Given a large amount of text data, a system is needed that can automatically construct a knowledge base using statistical machine learning (SML) methods, manage the uncertainty inherent in the extracted knowledge, and maintain them over time.

In this talk, I summarize the major results from BayesStore, a probabilistic database system that natively supports SML models and various inference algorithms to perform advanced statistical data analysis, including knowledge extraction from text. Results show that using in-database SML methods to extract and query probabilistic knowledge can significantly improve performance and answer quality.

Build on BayesStore, I propose to build a probabilistic knowledge base (ProbKB) system with a deep integration of the SML methods with scalable data processing frameworks. A ProbKB system should be designed to support various aspects in the life of a knowledge base (KB) including KB extraction, expansion, evolution, and integration. I will discuss in detail the challenges and our current progress in the following three research directions: (1) scalable statistical information extraction; (2) probabilistic deductive inference and incremental maintenance over large uncertain KBs; and (3) probabilistic knowledge integrate from both SML and crowd-sourcing.

Bio: Daisy Zhe Wang is an Assistant Professor in the CISE department at the University of Florida. She obtained her Ph.D. degree from the EECS Department at the University of California, Berkeley in 2011. Her research interest is in building systems and designing algorithms to support "better" data analysis, where "better" can mean: more efficient/scalable, more advanced, more accurate, more interactive, self-improving, and easier to use. She is particularly interested in bridging data management systems with statistical and probabilistic models and tools. She currently pursues research topics such as probabilistic databases, probabilistic knowledge bases, large-scale statistical text analysis, query-driven interactive machine learning, and crowd assisted machine learning.

Refreshments will be provided at 10:30 AM