Especially through the increasing popularity of the Web, more and more hyperdocuments have been created. A major task in creating hypertext document is link generation. Therefore, at least some support for link generation is highly desirable. We have created linking services that create semantic hyperlinks automatically among Web information contextualized. Using those services, users are able to exploit relationships among Web documents without having to suffer all cognitive overhead implied by manual authorship of documents. Therefore those services did not make any preconceived assumptions about the information needs of their users. This paper presents an infrastructure for generating links by considering relevance feedback given by users. The relevance feedback information is a information retrieval method which is required by the proposed infrastructure during the presentation of semantic links identified. At that time, users indicate which of returned semantic links are useful. The original context is automatically reformulated upon those relevance judgments and the new "feedback context" is then compared to the collection of documents, returning an improved set of documents to the user. This process can continue until the users considered all presented links are relevant. To demonstrate the utility of our infrastructure, we have run an experiment where latent semantic links extracted from the Sport section of two online versions of Brazilian newspapers are evaluated via relevance feedback information by users.