elastic-phonatic-comparsion

comparison between elasticsearch phonatic algorithms

TLDR;

click me to see the comparison.

start elasticsearch 5 and kibana

  docker-compose up -d

elasticsearch plugings needed docker run -it elasticsearch bash then:

  # provide char_filter icu_normalizer and filter icu_folding
  bin/elasticsearch-plugin install analysis-icu
  # provide phonatic algorithms
  bin/elasticsearch-plugin install analysis-phonetic

  #### restart elasticsearch => docker-compose restart

Test:

• load kibana console with test date by clicking here

Notes:

• if you are using elasticsearch older than 5, you should check changes happened to suggester
• autocomplete filed does not need payload attribute any more, _source is used as a replacement
• autocomplete filed does not need output attribute, you could use _source to simulate it.
• consider using index/_search instead of the deprecated index/_suggest
• Fuzzy Matching to Search by Sound

Issues:

• suggester context do ORed thing not AND 21291

How to custom analyzers

• char_filters: Character filters are used to preprocess the stream of characters before it is passed to the tokenizer.

• tokenizer: A tokenizer receives a stream of characters, breaks it up into individual tokens (usually individual words), and outputs a stream of tokens.

• filter: Token filters accept a stream of tokens from a tokenizer and can modify tokens (eg lowercasing), delete tokens (eg remove stopwords) or add tokens (eg synonyms).

     {
        "analyzer":{
           "my_custom_analyzer":{
              "type":"custom",
              "tokenizer":"standard",
              "char_filter":[
                 "html_strip"
              ],
              "filter":[
                 "lowercase",
                 "asciifolding"
              ]
           }
        }
     }
  

results

• ✓ means match
• ☓ means doesn't match

cities/algorithms	cologne	soundex	metaphone	doublemetaphone
Kairo vs Cairo	✓	☓	✓	✓
Kairo vs Qairo	✓	☓	✓	✓
Koln vs Köln	✓	✓	✓	✓
Paris vs Biarritz	✓	☓	☓	☓
Berlin vs Paralimni	✓	☓	☓	☓

suggestion

• Metaphone algorithm produce the most expected result for my test
• Don't depend only on the phonatic algorithm, you should have a Weight for every suggestion