Traditional information retrieval (IR) systems only allow users access to documents which match their current query and, therefore, users can only give relevance feedback on matching documents (or those with a matching strength greater than a set threshold). This paper shows that, in systems which allow access to non-matching documents (e.g. hybrid hypertext and information retrieval systems), the strength of the effect of giving relevance feedback varies between matching and non-matching documents. For positive feedback the results shown here are encouraging as they can be justified by an intuitive view of the process. However, for negative feedback the results show behaviour which cannot easily be justified. Taken with results from user evaluations, which showed that users tended to use negative feedback differently from that expected by IR engines, this indicates that negative feedback is complex and its use must be questioned.
Categories and subject descriptors: H.3.3 [Information storage and retrieval] information search and retrieval -- search process.
General terms: Experimentation, theory
Additional key words and phrases: Hypertext, relevance feedback, negative feedback, free text information retrieval, vector space model, probabilistic model.
Compressed postscript version
Mark Dunlop's Publications