To correctly fill in a Web form, a software agent needs to understand what information the Web form requires. This paper proposes that understanding be sought by scraping the Web page to form an ontology, and then matching the ontology to related ontologies. This approach, which the authors call application semantics analysis, moves beyond the linguistic analysis of discourse domains used in data-model analysis.

Web form scraping results in terms (names and labels), values, compositions (form structure), and precedence. Precedence, an original contribution of this paper, represents the presentation order of items on a Web form.

Matching is done over terms and values. Term matching and evaluation involve maximal substring match techniques; value matching exploits terms (“date” probably labels a date), form structure (a button box of colors), and syntax (email addresses contain an “@” symbol). Precedence modulates matching with presentation consistency; match likelihood is proportional to precedence similarity across ontologies.

Term- and value-matching experiments show term matching to be more precise than value matching. Precedence matching apparently works with less precision than value matching does, although it is unclear whether precedence matching is being tested alone or in combination with term matching. Citations reference more complete experiments, as well as systems using these techniques.

This paper is twice as long as it should have been; most of the overage is artificial intelligence boilerplate. It would be interesting to see the result of having a Perl programmer rewrite this paper.