Relevance feedback has been used to improve the performance of retrieval systems [40]. For vector space representations, the method for query reformulation in response to user feedback is vector adjustment. Since queries and documents are both vectors, the query vector is moved closer to the vectors representing documents which received positive feedback and away from the vectors representing documents which received negative feedback.
The feedback mechanism used here is a generalization of the above method. Each of the field-vectors in the profile is modified in response to user feedback. The process of modification for each of the field-vectors is similar to the classical vector adjustment method.
Consider a profile , which contributed document
for
presentation to the user. The user provides feedback
, which is a
positive or negative integer indicating the amount of feedback. Each
field-vector in the profile is changed in proportion to the feedback
received,
i.e. the weight of each term in each field is modified
proportional to the learning rate and the feedback.
is the learning rate which indicates the sensitivity of the
profile to user feedback.
Addition in the context of equation 
means
where is the weight of term
in field
of the
profile and
is the weight of the same term in field
of
the document. The resulting effect is that, for those terms already
present in the profile, the term-weights are modified in proportion to
the feedback. Terms not already in the profile are added to the
profile. The fitness of profile
is also modified in proportion to
the feedback received
where is the sensitivity of the fitness to user feedback.
When feedback is provided for the document, the implicit assumption is that
the feedback refers to all the features of the document, as can be seen from
equation .
If the user provides selective feedback for certain terms in the document, only
the fields corresponding to the term are modified. For example, if the user
likes a certain author of the document and provides selective feedback, only
the author field is modified using equation .
Similarly, the user can give selective feedback for a portion of the text.