Extracting Visually Presented Element Relationships from Web Documents

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