Mining informative patterns from the Web, object-oriented and multiple data sources

点击翻译按钮获取中文摘要

Ability to mine multiple diverse datasets will provide answers to more complex pattern queries, discover hidden, derived, multimedia and accurate knowledge from historical data, in a more efficient fashion, as well as provide better recommendations to end users. Some example queries that can be solved are: (i) given a set of web pages, get the list of products with lowest price ranges in them in the past five years, (ii) given a set of University web pages, present faculty by gender, ranks and area of specialization, (iii) find all frequently purchased products from a set of databases including their image, customer comments, number sold over a period of time. Thus, the objectives of the proposed research are to (1) build a set of techniques for multiple database mining of diverse datasets using object-oriented data modeling of such diverse data as web data, (2) integrate pattern mining techniques in end user tools for web recommendations, personal organization and system monitoring. Scientific methods for accomplishing research goals include using non-deterministic finite state automata (NFA) to generate web site schemas, and the relationships between them as object and class hierarchies. Then, use association rule and pattern tree mining techniques to develop object-oriented frequent pattern mining techniques among many other mining techniques for extracting derived, rare, frequent, relevant and other patterns. The proposed research contributes to overall quality of life of Canadians as well as their productivity by cutting down drastically, the amount of time spent on the web searching and sifting through hidden and incorrect answers to queries manually. It increases organizations' and users' productivity by making pattern mining process less manual, more available, more applicable and more effective.