The purpose of this exercise is to find ways to accurately query objects without having to use tags, just the objects themselves. This leads to very complex sorting and searching of a database that can be potentially very large. We need to pick and choose the correct "similarity" features to accurately query these objects. Unfourtunately, similarity choices can change, depending on what you need from each database. Also, with the higher number of features increasing (the dimensionality of the vectors increasing), we have poorer results. Most of these techniques are from the 80's, but there have been newer types of indexing that can handle higher dimensions with greater accuracy (X trees, R* trees).