An analysis of learned proximity functions

A lot of recent work has shown that the proximity of terms can be exploited to improve the performance of information retrieval systems. We review a recent approach that uses an intuitive framework to incorporate proximity functions into vector based information retrieval systems. More importantly, we present several proximity functions that were learned within this framework and show that they adhere to previously developed constraints regarding the shape of a good proximity function. Finally, we include results of all of the learned functions on unseen test data that shows the consistency of the learning approach used.