Retrieval strategies for hypertext

The links in a hypertext database distinguish it from a typical document database used in information retrieval applications. We show how these links can be incorporated into a probabilistic retrieval model based on inference nets. Retrieval experiments using a strategy based on this model show that it produces results at least as good as a spreading activation strategy. The results also show that citations can be a better source of links for a hypertext database than nearest neighbors, even though the latter appear to connect related documents. For this test collection, the overall improvement obtained by including hypertext links in the retrieval model was low. The hypertext model may, however, have additional benefits in multimedia environments.