Abstract
Most documents available in Textual Database or in Internet are strongly structured. This is the case for example for scientific papers or written documents using markup languages (HTML, XML). This information provided by the structure can be exploited by systems of information retrieval to define the granularity of elements to return in response to a request made by a user or to improve the relevance of these results. In this article, We are interested in recovering multimedia elements. Like this, we propose a new metric for multimedia retrieval in XML documents which is based on computing a geometric distance between XML nodes while taking into account kinship ties and proximities between them. This measure will introduce a new source of evidence for multimedia retrieval in structural documents which aims at finding relevant multimedia element that focus on the user information need. Experiments have been undertaken to show the effectiveness of our method.