Entity Resolution Interface

[shreyashankar]

Users need an interface to inspect and validate results of the resolve operator. This interface should be designed around:

• Assessing precision of the clusters; should all elements in a cluster actually be in that cluster?
• Assessing recall of the clusters; do we need to merge clusters together because a single cluster does not actually contain all documents for that cluster?

The interface's design goals are as follows:
DG1: See a "10000-foot" view of the clusters; how many clusters there are, and the distribution of # documents per cluster.
DG2: Assess cluster precision and recall, across all clusters.
DG3: For a single cluster, assess precision and recall. This may involve, highlighting nearby clusters and exemplar documents (to aid in recall), as well as creating clusters-on-the-fly via k-means for documents within a cluster, to see if the cluster itself contains outliers.