Korean Search Optimization

Prerequisites
Before reading this document, understand these concepts first:
Data Modeling - Analyzer, text vs keyword types
Query DSL - match, multi_match queries

This document covers how to optimize Korean language search in Elasticsearch. We’ll implement the Nori analyzer, autocomplete, and initial consonant (chosung) search.

Korean has fundamentally different linguistic characteristics from English, making proper search impossible with the default English analyzer. Searching “Samsung Electronics” (삼성전자) won’t find “Samsung Electronics Inc.” (삼성전자가) or “Samsung Electronics Co.” (삼성전자를), and it can’t extract “purchase” (구매) from “purchased” (구매했습니다). When users can’t find results they expect to find, they lose trust in the search service.

The core of Korean search is morphological analysis. You must separate particles, properly decompose compound nouns, and extract stems. Elasticsearch provides the Nori analyzer as an official plugin for this purpose. By implementing autocomplete, initial consonant search, and synonym handling on top of this, you can provide the search experience Korean users expect. This document covers step-by-step implementation methods for Korean search that can be used in production services.

Challenges of Korean Search#

English vs Korean#

Characteristic	English	Korean
Word Separation	Space-separated	Particles attached
Stem Changes	running → run	먹었다 (ate) → 먹다 (eat)
Synonyms	car, automobile	자동차, 차, 차량 (car, vehicle)
Typos	helo → hello	ㅎㅏㄴ글 → 한글 (Korean)

Example: "삼성전자 갤럭시를 구매했습니다" (Purchased Samsung Electronics Galaxy)

English analyzer: ["삼성전자", "갤럭시를", "구매했습니다"]  ❌ Not searchable
Korean analyzer: ["삼성", "전자", "갤럭시", "구매"]  ✅ Individual terms searchable

Nori Analyzer Configuration#

Nori Plugin Installation#

# In Elasticsearch container
bin/elasticsearch-plugin install analysis-nori

# Docker Compose
services:
  elasticsearch:
    image: docker.elastic.co/elasticsearch/elasticsearch:8.11.0
    command: >
      bash -c "
        bin/elasticsearch-plugin install analysis-nori &&
        /usr/local/bin/docker-entrypoint.sh
      "

Basic Nori Analyzer#

PUT /products
{
  "settings": {
    "analysis": {
      "analyzer": {
        "nori_analyzer": {
          "type": "custom",
          "tokenizer": "nori_tokenizer",
          "filter": ["nori_readingform", "lowercase"]
        }
      }
    }
  },
  "mappings": {
    "properties": {
      "name": {
        "type": "text",
        "analyzer": "nori_analyzer"
      }
    }
  }
}

Verify Analysis Results#

POST /products/_analyze
{
  "analyzer": "nori_analyzer",
  "text": "삼성전자 갤럭시를 구매했습니다"
}

// Result
{
  "tokens": [
    {"token": "삼성", "start_offset": 0, "end_offset": 2},
    {"token": "전자", "start_offset": 2, "end_offset": 4},
    {"token": "갤럭시", "start_offset": 5, "end_offset": 8},
    {"token": "구매", "start_offset": 10, "end_offset": 12}
  ]
}

Advanced Nori Configuration#

Compound Noun Decomposition#

PUT /products
{
  "settings": {
    "analysis": {
      "tokenizer": {
        "nori_mixed": {
          "type": "nori_tokenizer",
          "decompound_mode": "mixed",  // none, discard, mixed
          "discard_punctuation": true,
          "user_dictionary": "userdict_ko.txt"
        }
      },
      "analyzer": {
        "nori_mixed_analyzer": {
          "type": "custom",
          "tokenizer": "nori_mixed",
          "filter": [
            "nori_readingform",
            "nori_part_of_speech",
            "lowercase",
            "trim"
          ]
        }
      },
      "filter": {
        "nori_part_of_speech": {
          "type": "nori_part_of_speech",
          "stoptags": [
            "E", "IC", "J", "MAG", "MAJ", "MM",
            "SP", "SSC", "SSO", "SC", "SE",
            "XPN", "XSA", "XSN", "XSV",
            "UNA", "NA", "VSV"
          ]
        }
      }
    }
  }
}

decompound_mode	Description	“삼성전자” (Samsung Electronics) Result
`none`	No decomposition	[“삼성전자”]
`discard`	Remove original	[“삼성”, “전자”]
`mixed`	Keep both	[“삼성전자”, “삼성”, “전자”]

User Dictionary#

# config/userdict_ko.txt
# Format: word POS reading

삼성전자 NNP
갤럭시 NNP
아이폰 NNP
에어팟 NNP
맥북프로 NNP 맥북프로

Autocomplete#

Edge N-gram Approach#

PUT /products
{
  "settings": {
    "analysis": {
      "tokenizer": {
        "nori_tokenizer": {
          "type": "nori_tokenizer",
          "decompound_mode": "mixed"
        }
      },
      "filter": {
        "edge_ngram_filter": {
          "type": "edge_ngram",
          "min_gram": 1,
          "max_gram": 10
        }
      },
      "analyzer": {
        "autocomplete_index": {
          "type": "custom",
          "tokenizer": "nori_tokenizer",
          "filter": ["lowercase", "edge_ngram_filter"]
        },
        "autocomplete_search": {
          "type": "custom",
          "tokenizer": "nori_tokenizer",
          "filter": ["lowercase"]
        }
      }
    }
  },
  "mappings": {
    "properties": {
      "name": {
        "type": "text",
        "analyzer": "autocomplete_index",
        "search_analyzer": "autocomplete_search"
      }
    }
  }
}

Autocomplete Search#

// Indexing
POST /products/_doc
{
  "name": "삼성 갤럭시 S24 울트라"
}

// Search with "갤럭" (Galax)
GET /products/_search
{
  "query": {
    "match": {
      "name": "갤럭"
    }
  }
}

// Result: "삼성 갤럭시 S24 울트라" matched

Completion Suggester (Faster Approach)#

PUT /products
{
  "mappings": {
    "properties": {
      "name": {
        "type": "text",
        "analyzer": "nori_analyzer"
      },
      "name_suggest": {
        "type": "completion",
        "analyzer": "nori_analyzer",
        "contexts": [
          {
            "name": "category",
            "type": "category"
          }
        ]
      }
    }
  }
}

// Indexing
POST /products/_doc
{
  "name": "삼성 갤럭시 S24 울트라",
  "name_suggest": {
    "input": ["삼성", "갤럭시", "S24", "울트라", "삼성 갤럭시"],
    "contexts": {
      "category": "스마트폰"
    }
  }
}

// Autocomplete query
POST /products/_search
{
  "suggest": {
    "product-suggest": {
      "prefix": "갤럭",
      "completion": {
        "field": "name_suggest",
        "size": 5,
        "contexts": {
          "category": "스마트폰"
        }
      }
    }
  }
}

Initial Consonant Search (Chosung)#

Chosung Filter Implementation#

This is a Korean-specific feature that allows searching by initial consonants only. For example, searching “ㄱㄹㅅ” would find “갤럭시” (Galaxy).

PUT /products
{
  "settings": {
    "analysis": {
      "char_filter": {
        "chosung_filter": {
          "type": "pattern_replace",
          "pattern": "([가-깋])",
          "replacement": "ㄱ"
        }
      },
      "tokenizer": {
        "chosung_tokenizer": {
          "type": "pattern",
          "pattern": ""
        }
      },
      "analyzer": {
        "chosung_analyzer": {
          "type": "custom",
          "char_filter": ["chosung_filter"],
          "tokenizer": "keyword"
        }
      }
    }
  },
  "mappings": {
    "properties": {
      "name": {
        "type": "text",
        "analyzer": "nori_analyzer",
        "fields": {
          "chosung": {
            "type": "text",
            "analyzer": "chosung_analyzer"
          }
        }
      }
    }
  }
}

Using Elasticsearch Plugin (Recommended)#

# Install jaso-analyzer plugin
bin/elasticsearch-plugin install https://github.com/.../jaso-analyzer.zip

PUT /products
{
  "settings": {
    "analysis": {
      "analyzer": {
        "jaso_search": {
          "type": "custom",
          "tokenizer": "keyword",
          "filter": ["jaso_filter"]
        }
      },
      "filter": {
        "jaso_filter": {
          "type": "jaso",
          "chosung": true
        }
      }
    }
  },
  "mappings": {
    "properties": {
      "name": {
        "type": "text",
        "analyzer": "nori_analyzer",
        "fields": {
          "jaso": {
            "type": "text",
            "analyzer": "jaso_search"
          }
        }
      }
    }
  }
}

// Search "ㄱㄹㅅ" to find "갤럭시"
GET /products/_search
{
  "query": {
    "match": {
      "name.jaso": "ㄱㄹㅅ"
    }
  }
}

Synonym Handling#

Synonym Filter#

PUT /products
{
  "settings": {
    "analysis": {
      "filter": {
        "korean_synonym": {
          "type": "synonym",
          "synonyms": [
            "자동차, 차, 차량, 카",
            "노트북, 랩탑, 휴대용컴퓨터",
            "핸드폰, 휴대폰, 스마트폰, 폰",
            "에어컨, 에어콘, 냉방기",
            "tv, 티비, 텔레비전"
          ]
        }
      },
      "analyzer": {
        "nori_synonym": {
          "type": "custom",
          "tokenizer": "nori_tokenizer",
          "filter": [
            "nori_readingform",
            "lowercase",
            "korean_synonym"
          ]
        }
      }
    }
  }
}

Using Synonym File#

# config/synonyms_ko.txt
자동차, 차, 차량, 카
노트북, 랩탑, 휴대용컴퓨터

"filter": {
  "korean_synonym": {
    "type": "synonym",
    "synonyms_path": "synonyms_ko.txt"
  }
}

Production Mapping Example#

Product Search Index#

PUT /products_v1
{
  "settings": {
    "number_of_shards": 3,
    "number_of_replicas": 1,
    "analysis": {
      "tokenizer": {
        "nori_mixed": {
          "type": "nori_tokenizer",
          "decompound_mode": "mixed",
          "user_dictionary": "userdict_ko.txt"
        }
      },
      "filter": {
        "edge_ngram_2_10": {
          "type": "edge_ngram",
          "min_gram": 2,
          "max_gram": 10
        },
        "korean_synonym": {
          "type": "synonym",
          "synonyms_path": "synonyms_ko.txt"
        },
        "nori_posfilter": {
          "type": "nori_part_of_speech",
          "stoptags": ["E", "J", "SC", "SE", "SF", "SP", "SSC", "SSO", "VCP", "XSN", "XSV"]
        }
      },
      "analyzer": {
        "korean_index": {
          "type": "custom",
          "tokenizer": "nori_mixed",
          "filter": ["nori_readingform", "nori_posfilter", "lowercase", "korean_synonym"]
        },
        "korean_search": {
          "type": "custom",
          "tokenizer": "nori_mixed",
          "filter": ["nori_readingform", "nori_posfilter", "lowercase"]
        },
        "autocomplete_index": {
          "type": "custom",
          "tokenizer": "nori_mixed",
          "filter": ["nori_readingform", "lowercase", "edge_ngram_2_10"]
        },
        "autocomplete_search": {
          "type": "custom",
          "tokenizer": "nori_mixed",
          "filter": ["nori_readingform", "lowercase"]
        }
      }
    }
  },
  "mappings": {
    "properties": {
      "name": {
        "type": "text",
        "analyzer": "korean_index",
        "search_analyzer": "korean_search",
        "fields": {
          "autocomplete": {
            "type": "text",
            "analyzer": "autocomplete_index",
            "search_analyzer": "autocomplete_search"
          },
          "keyword": {
            "type": "keyword"
          }
        }
      },
      "brand": {
        "type": "keyword",
        "fields": {
          "text": {
            "type": "text",
            "analyzer": "korean_index"
          }
        }
      },
      "category": {
        "type": "keyword"
      },
      "price": {
        "type": "integer"
      },
      "description": {
        "type": "text",
        "analyzer": "korean_index",
        "search_analyzer": "korean_search"
      },
      "created_at": {
        "type": "date"
      }
    }
  }
}

// Alias setup
POST /_aliases
{
  "actions": [
    { "add": { "index": "products_v1", "alias": "products" } }
  ]
}

Combined Search Query#

GET /products/_search
{
  "query": {
    "bool": {
      "must": [
        {
          "multi_match": {
            "query": "삼성 갤럭시",
            "fields": [
              "name^3",
              "name.autocomplete^2",
              "brand.text",
              "description"
            ],
            "type": "best_fields",
            "operator": "or",
            "minimum_should_match": "75%"
          }
        }
      ],
      "filter": [
        { "term": { "category": "스마트폰" } },
        { "range": { "price": { "gte": 500000, "lte": 1500000 } } }
      ]
    }
  },
  "highlight": {
    "fields": {
      "name": {},
      "description": {}
    },
    "pre_tags": ["<em>"],
    "post_tags": ["</em>"]
  },
  "sort": [
    { "_score": "desc" },
    { "created_at": "desc" }
  ]
}

Spring Data Elasticsearch Integration#

Repository Configuration#

@Document(indexName = "products")
public class Product {
    @Id
    private String id;

    @Field(type = FieldType.Text, analyzer = "korean_index", searchAnalyzer = "korean_search")
    private String name;

    @Field(type = FieldType.Keyword)
    private String brand;

    @Field(type = FieldType.Integer)
    private Integer price;

    @Field(type = FieldType.Text, analyzer = "korean_index")
    private String description;
}

@Repository
public interface ProductRepository extends ElasticsearchRepository<Product, String> {

    // Autocomplete
    @Query("""
        {
          "match": {
            "name.autocomplete": "?0"
          }
        }
        """)
    List<Product> findByNameAutocomplete(String query);

    // Combined search
    @Query("""
        {
          "bool": {
            "must": [
              {
                "multi_match": {
                  "query": "?0",
                  "fields": ["name^3", "brand.text", "description"],
                  "type": "best_fields"
                }
              }
            ],
            "filter": [
              { "term": { "brand": "?1" } }
            ]
          }
        }
        """)
    Page<Product> searchProducts(String query, String brand, Pageable pageable);
}

Performance Optimization Tips#

Item	Recommended Setting
`decompound_mode`	`mixed` (balance between precision and recall)
`edge_ngram min_gram`	2 (single character search has too much noise)
Synonyms	Apply at search time rather than index time
Autocomplete	Use Completion Suggester (speed optimized)
Initial consonant search	Separate field (`name.chosung`)

Next Steps#

Query DSL - Various search queries
Search Relevance - Score adjustment
Performance Tuning - High-volume processing