Web unit mining

In web classification, most researchers assume that theobjects to classify are individual web pages from one ormore web sites. In practice, the assumption is toorestrictive since a web page itself may not alwayscorrespond to a concept instance of some semantic concept(or category) given to the classification task. In thispaper, we want to relax this assumption and allow aconcept instance to be represented by a subgraph of webpages or a set of web pages. We identify several newissues to be addressed when the assumption is removed, andformulate the web unit mining problem. We also propose aniterative web unit mining (iWUM) method that first findssubgraphs of web pages using some knowledge about web sitestructure. From these web subgraphs, web units areconstructed and classified into semantic concepts (orcategories) in an iterative manner. Our experiments usingthe WebKB dataset showed that iWUM improves the overallclassification performance and works very well on the morestructured parts of a web site.