Automated extraction of non <h>-tagged headers in webpages by decision trees

The guideline #5.2a in the JIS X 8341-3 reccomends to “represent headings with heading elements instead of difference in font size, etc”. Thus, in checking webpage accessibility, headers that are not tagged with heading tags (<;h1>;-<;h6>;) should be extracted as problems. In this paper, we propose a method for the extraction. Our idea is to let a machine learning method to automatically derive extraction rules from problem instances on the web. We define 26 attributes of HTML elements for deriving the rules. Values of these attributes are calculated by parsing the HTML source of the webpage. Accuracy of our method was evaluated by 10-fold cross validations with the data we collected from the web. The accuracy was 85-88% in average in terms of F-measure. Non <;h>;-tagged image headers were slightly better discriminated than non <;h>;-tagged text headers.