Direct approach for mining association rules from structured XML data

• Main Author: Abazeed, Ashraf Riad
• Format: Thesis
• Language: English
• Published: 2012
• Online Access: http://psasir.upm.edu.my/id/eprint/27118/1/FSKTM%202012%2021R.pdf; http://psasir.upm.edu.my/id/eprint/27118/

XML has become the standard for data representation on the internet. This expansion in reputation has prompt the need for a technique to access XML documents for particular information and to manipulate repositories of documents represented in XML to find specific documents. Having the ability to extract information from XML data would answer the problem of mining the web contents which is a very useful and required power nowadays. Efforts are made to develop a new tool or method for extracting information from XML data directly without any preprocessing or post processing of the XML documents. Association rules express the probability of the existing of a set of items when another set of items exists. It searches for similarities among large database. “Web mining” refer to how we can apply the traditional mining techniques that works on relational data and bind it to new data input represented in XML data which might be semi structure or unstructured. There are several techniques to mine association rules from XML data. The basic approach is to map the XML documents to relational data model and to store them in a relational database. This allows us to apply the standard tools that are in use to perform rule mining from relational databases. Even though it makes use of the existing technology, this approach is often time consuming and involves manual intervention because of the mapping process. The focus of this study is to propose an enhancement on memory consumption by reducing the number of candidates generated for the existing FLEX algorithm which will reduce the amount of memory needed to execute the algorithm. Another aim of this study is to do an enhancement on the current structure of FLEX algorithm in terms of elimination of the candidate generation step. The thesis also provides a two different implementation of the modified FLEX algorithm using a java based parsers and XQuery implementation. The thesis outlines the two different implementation techniques of the existing FLEX algorithm using java based parsers and using a query language for XML. The implementation details shows the difference in accessing and manipulating XML v documents using java based parsers and query languages for XML and the steps needed to access an XML document until we produce a list of association rules . The proposed method, XiFLEX has been implemented using two different techniques (java based & XQuery) and compared with the original FLEX algorithm in its basic implementation and the Apriori algorithm for frequent patterns generation. The experiments were conducted on self generated data sets (7 different sets) and well known datasets (Mushroom & Cars Data set). The results have shows that the proposed method, XiFLEX, has a better performance in terms of the time it takes to generate frequent patterns and the number of candidates generated (memory consumption).