Peter Gibbons Memorial Lecture Series: There's gold in them thar mountains
Speaker: Professor Gillian Dobbie, Dept of Computer Science, The University of Auckland
When: Refreshments at 5.30pm, lecture starts at 6.00pm.
Gillian Dobbie has undertaken lecturing and research at the University of Melbourne, Victoria University of Wellington and the National University of Singapore. Her main areas of interest pertain to data management and the web. She has worked in the foundations of database systems, defining logical models for various kinds of database systems, and reasoning about the correctness of algorithms in that setting.
Professor Dobbie is Head of the Department of Computer Science at The University of Auckland. This will be her inaugural lecture as professor.
Synopsis: Every day a tremendous amount of data is captured that tracks activities such as web-site accesses, sales, credit card usage, and movement using radio-frequency identification (RFID). One of today's challenges is to make good use of the data that is collected. Because of the amount of data, any techniques that are developed have to process the data not only accurately, but also efficiently.
Typically one does not want to look at each record or entry in detail - rather one is interested in grouping the data to find out what the trends are, or when there is some unusual activity. Techniques that find patterns in queries can be used to predict likely behaviour, and may be used for recommending future queries or for preloading potentially useful information. Examples of such "recommender" systems are in use at web sites where people buy books. The recommender finds all the books bought by the people who bought a particular book and recommends them to any new purchaser of that book. Similarly, the speed of web access may be improved if it can be predicted what other pages they are likely to access so that they can be cached locally. In contrast, techniques for detecting unexpected "outlier" behaviour may be used to detect abnormalities such as credit card fraud.
"Data clustering" provides a way to group data based on patterns within the data, and to find data that does not match the usual patterns. We have developed novel data clustering techniques and shown how they can be applied to web usage mining, recommender systems, and outlier detection. We have run experiments to demonstrate the accuracy of the techniques, and also their efficiency compared to traditional data clustering techniques.