histograph

A user might enter the name “Robert Schuman” in a search field, for example, only to generate a list of results containing documents that feature the name “Robert Schuman” in relation to a street, a building or a hospital named after him, but none about the statesman himself. And finding photos or videos of Robert Schuman can also be tricky if his name is not mentioned in an accompanying text such as a caption. But searching for entities based on co-occurrences of people, organisations, places, dates, etc. in a given document can help target a search. This idea is the basis for an application developed by the C²DH known as “histograph”.

As its name suggests, the aim is to display results in graph form. Users can make an initial selection of a specific period or document type, then use histograph to perform an analysis of the co-occurrence of names, generating networks of more or less significant relationships between several people or other entities. These results are refined using mathematical models, and can be displayed in the form of a network graph, with lines extending out from a selected entity or person.

Another original feature of histograph is that the application has been designed to actively involve users. They are given the option of confirming or ruling out whether, for a given document, the result of an automatic analysis really represents an entity, whether the name is actually mentioned in the text or whether the person in question is really in the photo. They can also agree or disagree with the annotations of previous users. The developers came up with a system of multi-level validation based on user participation by means of annotations. “This human contribution is what makes the application so effective, because it enables any imperfections in automatic detection to be corrected,” explains Dr Lars Wieneke, responsible for the development of the project. Once it has been finalised, histograph will help facilitate searches in all types of multimedia collections.

…