Matthew Michelson Information Sciences Institute University of Southern California 4676 Admiralty Way Marina del Rey, CA 90292 Email: Voice: (310) 448-8719

My research interest lies in the areas of information integration and machine learning. I am especially interested in information extraction and record linkage and the application of these techniques to build queryable data sets from ungrammatical and unstructured data sources.

For a demo of this idea, check out Phoebus Car Search! This demo allows you to search and sort online listings of used cars. It also dynamically links to MSN autos as a form of integration, allowing users to view information such as a value rating and a price range. Further, as a form of true data integration, the demo links the used car listings with crash ratings for the given car using data from the NHTSA. The point is to show that creating queryable data sets allows us to deeply and accurately search the listings and also perform information integtation with outside sources.

• M.S., Computer Science, 2005, University of Southern California
• B.S., Computer Science, 2002, The Johns Hopkins University

2008

• Matthew Michelson and Craig A. Knoblock, Creating Relational Data from Unstructured and Ungrammatical Data Sources, Journal of Artificial Intelligence Research (JAIR), 31, p.543-590, 2008 [PDF, Abstract, bibtex]
• Matthew Michelson, Aman Goel, and Craig A. Knoblock, Identifyinig Maps on the World Wide Web, In Proceedings of the 5th International Conference on Geographic Information Science (GIScience), Park City, Utah, 2008 [PDF, bibtex]
• Craig A. Knoblock, Jose Luis Ambite, Mark Carman, Matthew Michelson, Pedro Szekely, and Rattapoom Tuchinda, Beyond the Elves: Making Intelligent Agents Intelligent, AI Magazine, 29 (2), p.33-42, 2008 [bibtex]

2007

• Matthew Michelson and Craig A. Knoblock, Unsupervised Information Extraction from Unstructured, Ungrammatical Data Sources on the World Wide Web, International Journal of Document Analysis and Recognition (IJDAR), Special Issue on Noisy Text Analytics, 10, p.211-226, 2007 [PDF, SpringerLink, bibtex]
• Matthew Michelson and Craig A. Knoblock, Mining Heterogeneous Transformations for Record Linkage, In Proceedings of the 6th International Workshop on Information Integration on the Web (IIWEB-07), Vancouver, Canada, 2007 [PDF, bibtex]
• Matthew Michelson and Craig A. Knoblock, Beginning to Understand Unstructured, Ungrammatical Text: An Information Integration Approach, In Proceedings of the AAAI Spring Symposium on Machine Reading, Stanford, CA, 2007 [PDF, bibtex]
• Matthew Michelson and Craig A. Knoblock, An Automatic Approach to Semantic Annotation of Unstructured, Ungrammatical Sources: A First Look, In Proceedings of the 1st IJCAI Workshop on Analytics for Noisy Unstructured Text Data (AND-07), Hyderabad, India, 2007 [PDF, bibtex]

2006

• Matthew Michelson and Craig A. Knoblock, Learning Blocking Schemes for Record Linkage, In Proceedings of the 21st National Conference on Artificial Intelligence (AAAI-06), Boston, MA, 2006 [PDF, bibtex]
• Matthew Michelson and Craig A. Knoblock, Phoebus: A System for Extracting and Integrating Data from Unstructured and Ungrammatical Sources, In Proceedings of the 21st National Conference on Artificial Intelligence (AAAI-06), Intelligent Systems Demonstrations, Boston, MA, 2006 [PDF, bibtex]

2005

• Matthew Michelson and Craig A. Knoblock, Semantic Annotation of Unstructured and Ungrammatical Text, In Proceedings of the 19th International Joint Conference on Artificial Intelligence (IJCAI-05), Edinburgh, Scotland, 2005 [PDF, bibtex]
• Steven N. Minton, Claude Nanjo, Craig A. Knoblock, Martin Michalowski, and Matthew Michelson, A Heterogeneous Field Matching Method for Record Linkage, In Proceedings of the 5th IEEE International Conference on Data Mining (ICDM-05), Houston, TX, 2005 [PDF, bibtex]
• Matthew Michelson and Craig A. Knoblock. Beyond keyword search for data sources on the World Wide Web Software Developer's Journal Extra, 1/2006: 44-48. [Check it out in French or Spanish!]
• Matthew Michelson, Building Queryable Datasets from Ungrammatical and Unstructured Sources, Master's thesis, University of Southern California, 2005 [PDF]

Data sets
• Bidding For Travel data set  [Record Linkage and Info. Extraction]

Software
• BSL: Blocking Scheme Learner
  This code implements the algorithm presented in the Learning Blocking Schemes for Record Linkage paper. Blocking schemes are rules to efficiently generate candidate matches in record linkage, without pruning too many true positives from the set. For more on this, please read the paper.

• My wife!
• IJCAI 2005. Matt in Edinburgh, Scotland!
• AAAI 2006. Matt, Hector and Rattapoom in Boston, MA!