UNIVERSITY OF HERTFORDSHIRE
             COMPUTER SCIENCE RESEARCH COLLOQUIUM

                            presents

      "Parsing Romanian Addresses with Ruby: Agile Hill Climbing"

            Dr. Peter Lane (University of Hertfordshire)
            W. S. (Bill) Metcalf  (Geo Strategies, Ltd.)

                  3 December 2014 (Wednesday)
                          1 pm - 2 pm

                  Hatfield, College Lane Campus
                       Seminar Room D102

                  Everyone is Welcome to Attend

                  Refreshments will be available

Abstract:

Address parsing is the process of assigning terms within a customer address
into fields, such as the street name, locality or county.  Parsing is a vital
step in uniquely identifying an address so that further information, including
postcodes or geographical information, can be associated with it.  In this
talk, we will cover the design and implementation of a system which parses
Romanian addresses.  Examples will demonstrate the complexity of the problem,
in part due to the lack of any standard address format, but also from the
intricacies of the typical Romanian address.  

The current implementation uses a form of hill-climbing algorithm, with a
three-tiered arrangement of experts.  Development was completed in JRuby, a
JVM-based implementation of the Ruby language.  Apart from its easy-on-the-eye,
compact syntax, Ruby has excellent support for testing and meta-programming.
Ruby's flexibility was required to agilely adapt the implementation as knowledge
on how to parse Romanian addresses was developed during the project's lifetime.
Results will be presented based on real-world use by Geo Strategies, who
process large volumes of Romanian addresses each month.  We will discuss overall
accuracy, false positive rates, and processing speed, as well as look at future
directions for improvement.



---------------------------------------------------
Hertfordshire Computer Science Research Colloquium
http://cs-colloq.stca.herts.ac.uk