Place Ranking in Nominatim

Nominatim uses two metrics to rank a place: search rank and address rank. Both can be assigned a value between 0 and 30. They serve slightly different purposes, which are explained in this chapter.

Search rank

The search rank describes the extent and importance of a place. It is used when ranking search results. Simply put, if there are two results for a search query which are otherwise equal, then the result with the lower search rank will be appear higher in the result list.

Search ranks are not so important these days because many well-known places use the Wikipedia importance ranking instead.

The following table gives an overview of the kind of features that Nominatim expects for each rank:

The extent column describes how far a feature is assumed to reach when it is mapped only as a point. Larger features like countries and states are usually available with their exact area in the OpenStreetMap data. That is why no extent is given.

Address rank

The address rank describes where a place shows up in an address hierarchy. Usually only administrative boundaries and place nodes and areas are eligible to be part of an address. Places that should not appear in the address must have an address rank of 0.

The following table gives an overview how ranks are mapped to address parts:

The country rank 4 usually doesn't show up in the address parts of an object. The country is determined indirectly from the country code.

Ranks 5-24 can be assigned more or less freely. They make up the major part of the address.

Rank 25 is also an addressing rank but it is special because while it can be the parent to a POI with an addr:place of the same name, it cannot be a parent to streets. Use it for place features that are technically on the same level as a street (e.g. squares, city blocks) or for places that should not normally appear in an address unless explicitly tagged so (e.g place=locality which should be uninhabited and as such not addressable).

The street ranks 26 and 27 are handled slightly differently. Only one object from these ranks shows up in an address.

For POI level objects like shops, buildings or house numbers always use rank 30. Ranks 28 is reserved for house number interpolations. 29 is for internal use only.

Rank configuration

Search and address ranks are assigned to a place when it is first imported into the database. There are a few hard-coded rules for the assignment:

• postcodes follow special rules according to their length
• boundaries that are not areas and railway=rail are dropped completely
• the following are always search rank 30 and address rank 0:
  • highway nodes
  • landuse that is not an area

Other than that, the ranks can be freely assigned via the JSON file defined with CONST_Address_Level_Config according to their type and the country they are in.

The address level configuration must consist of an array of configuration entries, each containing a tag definition and an optional country array:

[ {
    "tags" : {
      "place" : {
        "county" : 12,
        "city" : 16,
      },
      "landuse" : {
        "residential" : 22,
        "" : 30
      }
    }
  },
  {
    "countries" : [ "ca", "us" ],
    "tags" : {
      "boundary" : {
        "administrative8" : 18,
        "administrative9" : 20
      },
      "landuse" : {
        "residential" : [22, 0]
      }
    }
  }
]

The countries field contains a list of countries (as ISO 3166-1 alpha 2 code) for which the definition applies. When the field is omitted, then the definition is used as a fallback, when nothing more specific for a given country exists.

tags contains the ranks for key/value pairs. The ranks can be either a single number, in which case they are the search and address rank, or an array of search and address rank (in that order). The value may be left empty. Then the rank is used when no more specific value is found for the given key.

Countries and key/value combination may appear in multiple definitions. Just make sure that each combination of country/key/value appears only once per file. Otherwise the import will fail with a UNIQUE INDEX constraint violation on import.