Extracting Visually Presented Element Relationships from Web Documents

Many documents in the World Wide Web present structured information that consists of multiplepieces of data with certain relationships among them. Although it is usually not difficult to identifythe individual data values in the document text, their relationships are often not explicitly describedin the document content. They are expressed by visual presentation of the document content that isexpected to be interpreted by a human reader. In this paper, we propose a formal generic model oflogical relationships in a document based on an interpretation of visual presentation patterns in thedocuments. The model describes the visually expressed relationships between individual parts of thecontents independently of the document format and the particular way of presentation. Therefore, itcan be used as an appropriate document model in many information retrieval or extraction applica-tions. We formally define the model, we introduce a method of extracting the relationships betweenthe content parts based on the visual presentation analysis and we discuss the expected applications.We also present a new dataset consisting of programmes of conferences and other scientific eventsand we discuss its suitability for the task in hand. Finally, we use the dataset to evaluate results ofthe implemented system.