Elasticsearch: generating terms from array using script

I would probably go with a nested type. I don't know all the details of your setup, but here is a proof of concept, at least. I took out the "items" property because I didn't need that many layers, and just used "tags" as the nested type. It could be added back in if needed, I think.

So I set up an index with a "nested" property:

DELETE /test_index

PUT /test_index
{
   "settings": {
      "number_of_shards": 1,
      "number_of_replicas": 0
   },
   "mappings": {
      "doc": {
         "properties": {
            "created_time": {
               "type": "date"
            },
            "tags": {
               "type": "nested",
               "properties": {
                  "tag_type": {
                     "type": "string",
                     "index": "not_analyzed"
                  },
                  "tag_id": {
                     "type": "string",
                     "index": "not_analyzed"
                  }
               }
            }
         }
      }
   }
}

Then added a couple of docs (notice that the structure differs slightly from yours):

PUT /test_index/doc/1
{
   "created_time": 1412988495000,
   "tags": [
      {
         "tag_type": "Placement",
         "tag_id": "id1"
      },
      {
         "tag_type": "Product",
         "tag_id": "id2"
      }
   ]
}

PUT /test_index/doc/2
{
   "created_time": 1412988475000,
   "tags": [
      {
         "tag_type": "Type3",
         "tag_id": "id3"
      },
      {
         "tag_type": "Type4",
         "tag_id": "id3"
      }
   ]
}

Now a scripted terms aggregation inside a nested aggregation seems to do the trick:

POST /test_index/_search?search_type=count
{
   "query": {
      "match_all": {}
   },
   "aggs": {
      "tags": {
         "nested": { "path": "tags" },
         "aggs":{
             "tag_vals": {
                 "terms": {
                     "script": "doc['tag_type'].value+':'+doc['tag_id'].value"
                 }
             }
         }
      }
   }
}
...
{
   "took": 3,
   "timed_out": false,
   "_shards": {
      "total": 1,
      "successful": 1,
      "failed": 0
   },
   "hits": {
      "total": 2,
      "max_score": 0,
      "hits": []
   },
   "aggregations": {
      "tags": {
         "doc_count": 4,
         "tag_vals": {
            "buckets": [
               {
                  "key": "Placement:id1",
                  "doc_count": 1
               },
               {
                  "key": "Product:id2",
                  "doc_count": 1
               },
               {
                  "key": "Type3:id3",
                  "doc_count": 1
               },
               {
                  "key": "Type4:id3",
                  "doc_count": 1
               }
            ]
         }
      }
   }
}

Here is the code I used:

http://sense.qbox.io/gist/4ceaf8693f85ff257c2fd0639ba62295f2e5e8c5