Explanation of SKG traversal implementation (Chapter 05)

[labdmitriy]

Hi @treygrainger ,

Could you please explain why we have the following Solr query for the SKG implement in Chapter 05, based on the example from Listing 5.4:

{
  "limit": 0,
  "params": {
    "q": "*:*",
    "fore": "{!${defType} v=$q}",
    "back": "*:*",
    "defType": "edismax",
    "f0_0_query": "advil"
  },
  "facet": {
    "f0_0": {
      "type": "query",
      "sort": {
        "relatedness": "desc"
      },
      "facet": {
        "relatedness": {
          "type": "func",
          "func": "relatedness($fore,$back)"
        },
        "f1_0": {
          "type": "terms",
          "limit": 8,
          "sort": {
            "relatedness": "desc"
          },
          "facet": {
            "relatedness": {
              "type": "func",
              "func": "relatedness($fore,$back)"
            }
          },
          "mincount": 2,
          "field": "body"
        }
      },
      "field": "body",
      "query": "{!edismax q.op=AND qf=body v=$f0_0_query}"
    }
  }
}

I've just started exploring Solr features therefore I have the following questions about it:

• Why do we have foreground dataset initially as "{!${defType} v=$q}", and because "q": "*:*", so as I understand initially we have foreground set as entire collection, but somehow using top-level facet of the type query we will filter foreground set to only documents that match our query ("query": "{!edismax q.op=AND qf=body v=$f0_0_query}")? Could you please explain the mechanism?
  I didn't find enough information about it in the documentaition and on external resources.
• Why do we need to get multiple relatedness function calculations and sorting based on these calculations? As I understand this calculation and sorting on the level of the facet f1_0 is enough?

I tried to simplify this request manually and tested the query using Solr Admin, and found that we have the same results with the following JSON query:

{
  "limit": 0,
  "params": {
    "fore": "{!edismax q.op=AND qf=body v=$f0_0_query}",
    "back": "*:*",
    "defType": "edismax",
    "f0_0_query": "advil"
  },
  "facet": {
    "f1_0": {
      "type": "terms",
      "limit": 8,
      "sort": {
        "relatedness": "desc"
      },
      "facet": {
        "relatedness": {
          "type": "func",
          "func": "relatedness($fore,$back)"
        }
      },
      "mincount": 2,
      "field": "body"
    }
  }
}

• q parameter was removed
• fore parameter has already target query
• Top-level facet f0_0 was removed
• Only one calculation and sorting of the relatedness function scores is presented

Could you please share your thoughts about it?

Thank you.

[labdmitriy]

Also I am really confused about relatedness arguments, in the formula from the book we have 3 arguments: x, fg and bg, but in Solr we have function with 2 arguments: fore and back, and as we can see from the simplified JSON that we even don't need q parameter and need to define only foreground and background datasets.
Could you please explain what is the real meaning of x from the formula from the book and how it is defined in Solr implementation?

Thank you.