Improving retrieval performance by relevance feedback

A study of the previously mentioned parallel processing application reveals that the relevance feedback process is easily implemented by using windowing and information display techniques to establish communications between system and users. In particular, ranked lists of retrieved documents can be graphically displayed for the user, and screen pointers can be used to designate certain listed items as relevant to the user’s needs. These relevance indications are then further used by the system to construct modified feedback queries.

It includes probabilistic feedback and vector processing models.

Twelve different relevance feedback methods are used for evaluation purposes with the six sample collections.

The advantage of all probabilistic feedback models compared with the conventional vector modification methods, is that the feedback process is directly related to the derivation of a weight for query terms. Indeed, the document similarity function of expression (8) increases by a weighting factor of log[p,(l - u,)/u,(l - pi)] for each query term i that matches a document, and this term weight is optimal under the assumed conditions of term independence and binary document indexing. residual collection