The Mancheser Ontology of Structural Semantics (MOSS)

The Manchester Ontology of Structural Semantics is a supporting project of SADIe. The aim of MOSS is to investigate ways in which the ontology used by the SADIe transcoding tool – or indeed other ontology based tools – can be created semi-automatically. This will decrease the time and effort needed to make the ontology and also open up the knowledge captured within the ontology to other applications.

About MOSS

Before SADIe can transcode a webpage, the structure of the page needs to be captured so that the transcoder is aware of the roles that each element on the page plays. This process at the moment suffers from two drawbacks:

1. The ontology creation process is manual. It takes a significant amount of effort to create an ontology that SADIe can use to transcode a page.
2. The ontology that is created is tied to the SADIe application. The effort required to build the ontology produces a model that only one tool can use. This creates a scenario where a significant amount of effort produces limited gains.

The principle idea behind MOSS is that several of our projects, notably SADIe and DANTE, use ontologies as the driving force behind their Web page manipulation. However, the ontologies that they use are closed, preventing the sharing of the knowledge captured within the ontology for reuse elsewhere. With MOSS, we will attempt to create a generic structural semantics ontology, which will be able to define the roles and characteristics of a diverse range of Website elements. With the structure of the Website defined, other applications, such as SADIe, DANTE, a Web harvester or any other Web application, will be able to extend and make use of the knowledge held within the MOSS ontology and manipulate the page accordingly. This facilitates the reuse of knowledge and allows multiple applications to exploit the information within a Web page without the time consuming and costly effort of redesigning and populating an ontology.

Proposed MOSS Architecture

Figure 1: Architecture For The MOSS Project
Click here to view an enlarged version of the figure.

1. The Ontology Generator takes as input the Cascading Style Sheet (CSS) of the website and the Document Object Model (DOM) of a sample webpage from the website. The DOM provides the structure of the page whilst the CSS provides the positioning and rendering of the elements on the page.
2. At this stage, the Ontology Generator uses predetermined heuristics, combined with the DOM and CSS, to try and determine what role the elements play within the webpage. For example, if the DOM tells the Ontology Generator that there is a list of links and the CSS tells the Ontology Generator that the list of links is rendered on the left side of the screen with a different colour background than the main content of the webpage, then heuristics tell us that the list of links is probably a menu.
3. The automated analysis of the webpage can only provide educated suggestions as to what the element roles within the webpage are. Therefore, the designer of the webpage provides feedback to the Ontology Generator, confirming or correcting the roles assigned to the elements of the webpage.
4. The final stage involves generating the ontology and then storing it in the ontology database. The database contains a link to the ontology and a reference to the website that it describes, to enable us to have different ontologies for different websites.

Prototype Experimental Tools

Tentative steps towards creating a generic ontology that can be used by multiple tools have been made. These tools have been made available for public use. However, they are experimental test platforms and we cannot guarantee that they will be fully functional, robust or accurate.

Generating the MOSS Ontology Through Heuristics

Omair Qadir has looked into creating the MOSS ontology through the use of heuristics (Figure 1 Section 2). His experimental prototype tool will be made publicly available soon.

Generating the MOSS Ontology By a Drag and Drop Interface

Wenhua Tang has looked into creating a drag and drop interface (Figure 1 Section 3). Designers can highlight areas of the page, drag the area to the appropriate structural tag and then from that generate the MOSS ontology. His experimental prototype tool will be made publicly available soon.