Abstract of the paper : 

             In domains like patents and biomedical information retrieval
a set of keywords as a query may not be sufficient but a document as a
query could be more useful. Entire content of the document as a query
will not be feasible as the results might be noisy and irrelevant. An
optimal query can be formed by extracting important keywords from
the document. We propose an unsupervised approach to extract and
rank candidate key phrases from a document. The goal is to maximize a
retrieval score like MAP or Recall depending upon the nature of search
task. A learning to rank method is used to select top phrases represented
by a set of pre-retrieval features used for query performance prediction.
The proposed keyphrase extraction technique is used to construct queries
from patents for Invalidity Search task in patent retrieval. Experiments
on a collection of USPTO Patents to maximize Recall and MAP show
that our approach is comparable to the baseline methods.
`