Mining Framework for Pervasive Applications

A pervasive web application is a server providingmany web services for its registered users. Nowadays,
three of basic services that a typical pervasive web application offers are membership management,
search service and map-enabled photo service. In this thesis, we design a data mining framework
composed of three different data mining techniques to improve the performance of three services. In
order to improve the performance of membership management, in the second chapter, we develop a
categorical decision tree classifier to classify users efficiently. It noted that the data of user profiles has
an unique phenomenon. Its characteristic is that few attributes of user profiles have higher information
gains to distinguish users. By exploiting this characteristic that a traditional decision tree classifier does
not consider, our designed classifier can reduce the execution time in generating a decision tree for user
classification. As a result, the decision tree generated by our classifier can identify users efficiently
for special marketing needs of an advertisement. For the improvement of a search service, in the third
chapter, we propose a sequential web search algorithm that leverages the sequential queries issued
by users to search the required information. Compared with previous works, our approach uses the
additional feedback data on result pages of sequential queries where prior works only use feedback
data of a query. Thus, our approach can provide a better ranking of result pages for sequential queries.
For the efficiency of retrieving geotagged photos, in the fourth chapter, we design a clustering algorithm
that incrementally clusters geotagged photos in accordance to thresholds of different scales. Compared
with other applications, we show the photo clusters instead of all photos where the number of photo
clusters is much less than that of all photos. As a result, the performance of map-enabled photo service
is improved efficiently.