In this paper we suggest how hypertext links and the content of HTML pages can be used to cluster pages into what we call Web documents We put forward a method to automatically construct a hierarchy ofWeb doc uments and with the help of an abstraction function the context hierarchy of a site This hierarchy is represented by a graph whose links are structural typed Structural links between nodes reveal a context relationship The con text hierarchy along with the graph of the pages underlying the site are used to better index and retrieve the pages Furthermore it permits a new operator to be added in the IRS Information Retrieval System query language whereby the user will be able to di erentiate the context from the subject of his queries .