Pretty close to what I am trying to do, hierarchical "soft/overlapping" clustering is what I am investigating. I have the design ready for matching user cluster

Pretty close to what I am trying to do, hierarchical "soft/overlapping" clustering is what I am investigating. I have the design ready for matching user cluster to corpus cluster. Then I use okapiBM25 to get primary results from this cluster and rank them (with an undecided algorithm). Then, second step, I augment the rank (increase or decrease) subset of top results with predetermined queries that match these results. For example among two equal documents & search query "Who is Tommen" if the second document has more people clicking then I increase the pagerank for second document (by a function of how many more people prefer the second document)