CODEWITHSUNDEEP
searchlocationelasticsearch
NathanAldenSr
NathanAldenSr

Reputation: 7961

What is an effective way to search world-wide location names with ElasticSearch?

I have location information provided by GeoNames.org parsed into a relational database. Using this information, I am attempting to build an ElasticSearch index that contains populated place (city) names, administrative division (state, province, etc.) names, country names and country codes. My goal is to provide a location search that is similar to Google Maps':

Google Maps

I don't need the cool bold highlighting, but I do need the search to return similar results in a similar way. I've tried creating a mapping with a single location field consisting of the entire location name (e.g., "Round Rock, TX, United States") and I've also tried having five separate fields consisting of each piece of a location. I've tried keyword and prefix queries and edgengram analyzers; I have been unsuccessful in finding the correct configuration to get this working properly.

What kinds of analyzers--both index and search--should I be looking at to accomplish my goals? This search doesn't have to be as perfected as Google's but I'd like it to be at least similar.

I do want to support partial-name matches, which is why I've been fiddling with edgengram. For example, a search of "round r" should match Round Rock, TX, United States. Also, I would prefer that results whose populated place (city) names begin with the exact search term be ranked higher than other results. For example, a search of "round ro" should match Round Rock, TX, United States before Round, Some Province, RO (Romania). I hope I've made this clear enough.

Here is my current index configuration (this is an anonymous type in C# that is later serialized to JSON and passed to the ElasticSearch API):

settings = new
{
    index = new
    {
        number_of_shards = 1,
        number_of_replicas = 0,
        refresh_interval = -1,
        analysis = new
        {
            analyzer = new
            {
                edgengram_index_analyzer = new
                {
                    type = "custom",
                    tokenizer = "index_tokenizer",
                    filter = new[] { "lowercase", "asciifolding" },
                    char_filter = new[] { "no_commas_char_filter" },
                    stopwords = new object[0]
                },
                search_analyzer = new
                {
                    type = "custom",
                    tokenizer = "standard",
                    filter = new[] { "lowercase", "asciifolding" },
                    char_filter = new[] { "no_commas_char_filter" },
                    stopwords = new object[0]
                }
            },
            tokenizer = new
            {
                index_tokenizer = new
                {
                    type = "edgeNGram",
                    min_gram = 1,
                    max_gram = 100
                }
            },
            char_filter = new
            {
                no_commas_char_filter = new
                {
                    type = "mapping",
                    mappings = new[] { ",=>" }
                }
            }
        }
    }
},
mappings = new
{
    location = new
    {
        _all = new { enabled = false },
        properties = new
        {
            populatedPlace = new { index_analyzer = "edgengram_index_analyzer", type = "string" },
            administrativeDivision = new { index_analyzer = "edgengram_index_analyzer", type = "string" },
            administrativeDivisionAbbreviation = new { index_analyzer = "edgengram_index_analyzer", type = "string" },
            country = new { index_analyzer = "edgengram_index_analyzer", type = "string" },
            countryCode = new { index_analyzer = "edgengram_index_analyzer", type = "string" },
            population = new { type = "long" }
        }
    }
}

Upvotes: 3

Views: 3528

Answers (1)

aash
aash

Reputation: 1323

This might be what you are looking for:

  "analysis": {
    "tokenizer": {
      "name_tokenizer": {
        "type": "edgeNGram",
        "max_gram": 100,
        "min_gram": 2,
        "side": "front"
      }
    },
    "analyzer": {
      "name_analyzer": {
        "tokenizer": "whitespace",
        "type": "custom",
        "filter": ["lowercase", "multi_words", "name_filter"]
      },
    },
    "filter": {
      "multi_words": {
        "type": "shingle",
        "min_shingle_size": 2,
        "max_shingle_size": 10
      },
      "name_filter": {
        "type": "edgeNGram",
        "max_gram": 100,
        "min_gram": 2,
        "side": "front"
      },          
    }
  }

I think using name_analyzer will replicate the google search that you are talking about. You can tweak the configuration a bit to suit your needs.

Upvotes: 2

Related Questions