Elasticsearch - search for word with Apostrophe

Question

I want to be able to search for the following words

Vincent Vincents Vincent's

Currently the test in the Database and ES is Vincent's

Is it possible to detect the possessive and also ignore the apostrophe. I have looked at the Word-Delimiter but can't seem to find a decent explanation on this

Answer 1

You need to understand how elasticsearch's analyzers work. Analyzers perform a tokenization (split an input into a bunch of tokens, such as on whitespace), and a set of token filters (filter out tokens you don't want, like stop words, or modify tokens, like the lowercase token filter which converts everything to lower case).

Analysis is performed at two very specific times - during indexing (when you put stuff into elasticsearch) and, depending on your query, during searching (on the string you're searching for).

That said, the default analyzer is the standard analyzer which consists of a standard tokenizer, standard token filter (to clean up tokens from the standard tokenizer), lowercase token filter, and stop words token filter.

To put this to an example, when you save the string "I love Vincent's pie!" into elasticsearch, and you're using the default standard analyzer, you're actually storing "i", "love", "vincent", "s", "pie". Then, when you attempt to search for "Vincent's" with a term query (which is not analyzed), you will not find anything because "Vincent's" is not one of those tokens! However, if you search for "Vincent's" using a match query (which is analyzed), you will find "I love Vincent's pie!" because "vincent" and "s" both find matches.

The bottom line, either:

1. Use an analyzed query, such as match, when searching natural language strings.
2. Set up the analyzers to match your needs. You could set up set up a custom analyzer that performs a whitespace tokenizer or a letter tokenizer or a pattern tokenizer if you want to get complicated, as well as whatever filters your heart desires. It depends on your use case, but if you're dealing with natural language sentences I don't recommend this because the standard tokenizer was built for natural language searching.

See http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/analysis.html for further reading.

Answer 2

Use the "possessive_english" stemmer as described in the ES docs: http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/analysis-stemmer-tokenfilter.html

Example:

{
  "index" : {
    "analysis" : {
        "analyzer" : {
            "my_analyzer" : {
                "tokenizer" : "standard",
                "filter" : ["standard", "lowercase", "my_stemmer"]
            }
        },
        "filter" : {
            "my_stemmer" : {
                "type" : "stemmer",
                "name" : "possessive_english"
            }
        }
    }
  }
}

Untested code, but should work. Here's a tested example with "word_delimiter":

{
  "index" : {
    "analysis" : {
        "analyzer" : {
            "my_analyzer" : {
                "tokenizer" : "standard",
                "filter" : ["standard", "lowercase", "my_word_delimiter"]
            }
        },
        "filter" : {
            "my_word_delimiter" : {
                "type" : "word_delimiter",
                "preserve_original": "true"
            }
        }
    }
  }
}

Works for me :-) ES docs: http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/analysis-word-delimiter-tokenfilter.html