@Halfak and I found that the scripts in https://github.com/Wikidata-lib/PropertySuggester-Python, are used for generating the data in wbs_propertypairs.
However, the current scripts can only be used for property-pairs "instance of" and "subclass of". We might need to rewrite the scripts in order to include other property-pairs.
Below are the steps that proposed by @Halfak in order to do this:
(1) Run the set of scripts for generating propertypairs against a database dump.
(2) Write a new script that generates a proposed set of pid1, qid1 for inclusion based on some version of optimal coverage/clustering
(3) review the results and iterate if necessary (e.g. we might find pairs that we want to exclude because their coverage doesn't help us)
(4) Estimate how much more space it will take to store the extended property tables and propose a switch
(5) review with hoo et al. and get it deployed.
(6) Build features that use this table to assess features of completeness through API calls.