To use these relevance feedback algorithms, the retrieval
system should be aware of the non-relevant information as
well as relevant information. However, our system allows a
user to select related term cluster from the term clusters
provided by retrieval system only once to reduce the frequent
participation of the user’s judgment. This makes it
impossible to get the non-relevant information. In this
reason, this research assumes that each selected group and
no-selected group are mutually exclusive. That is, if a user
selects a cluster group considered to be ‘relevant’ to his
purpose of retrieval, the other groups that are not selected are
assumed to be ‘non-relevant’ to the purpose.