How to Index Fields for Autocompletion

string

required

Human-readable label that identifies this field type. Value must be string.

string

optional

Name of the analyzer to use with this autocomplete mapping. You can use any Atlas Search analyzer except the lucene.kuromoji language analyzer and the following custom analyzer tokenizers and token filters:

• nGram Tokenizer

• edgeGram Tokenizer

• daitchMokotoffSoundex Token Filter

• nGram Token Filter

• edgeGram Token Filter

• shingle Token Filter

lucene.standard

int

optional

Maximum number of characters per indexed sequence. The value limits the character length of indexed tokens. When you search for terms longer than the maxGrams value, Atlas Search truncates the tokens to the maxGrams length.

15

int

optional

Minimum number of characters per indexed sequence. We recommend 4 for the minimum value. A value that is less than 4 could impact performance because the size of the index can become very large. We recommend the default value of 2 for edgeGram only.

2

enum

optional

Tokenization strategy to use when indexing the field for autocompletion. Value can be one of the following:

• edgeGram - create indexable tokens, referred to as grams, from variable-length character sequences starting at the left side of the words as delimited by the analyzer used with this autocomplete mapping.

• rightEdgeGram - create indexable tokens, referred to as grams, from variable-length character sequences starting at the right side of the words as delimited by the analyzer used with this autocomplete mapping.

• nGram - create indexable tokens, referred to as grams, by sliding a variable-length character window over a word. Atlas Search creates more tokens for nGram than edgeGram or rightEdgeGram. Therefore, nGram takes more space and time to index the field. nGram is better suited for querying languages with long, compound words or languages that don't use spaces.

edgeGram, rightEdgeGram, and nGram are applied at the letter-level. For example, consider the following sentence:

The quick brown fox jumps over the lazy dog.

When tokenized with a minGrams value of 2 and a maxGrams value of 5, Atlas Search indexes the following sequence of characters based on the tokenization value you choose:

th
the
the{SPACE}
the q
qu
qui
quic
uick
...

og
dog
{SPACE}dog
y dog
zy
azy
lazy
{SPACE}lazy
he
the
{SPACE}the
r the
er
ver
over
{SPACE}over
...

th
the
the{SPACE}
the q
he
he{SPACE}
he q
he qu
e{SPACE}
e q
e qu
e qui
{SPACE}q
{SPACE}qu
{SPACE}qui
{SPACE}quic
qu
qui
quic
quick
...

Indexing a field for autocomplete with an edgeGram, rightEdgeGram, or nGram tokenization strategy is more computationally expensive than indexing a string field. The index takes more space than an index with regular string fields.

For the specified tokenization strategy, Atlas Search applies the following process to concatenate sequential tokens before emitting them. This process is sometimes referred to as "shingling". Atlas Search emits tokens between minGrams and maxGrams characters in length:

• Keeps tokens less than minGrams.

• Joins tokens greater than minGrams but less than maxGrams to subsequent tokens to create tokens up to the specified maximum number of characters in length.

edgeGram

boolean

optional

Flag that indicates whether to perform normalizations such as including or removing diacritics from the indexed text. Value can be one of the following:

• true - perform normalizations such as ignoring diacritic marks in the index and query text. For example, a search for cafè returns results with the characters cafè and cafe because Atlas Search returns results with and without diacritics.

• false - don't perform normalizations such as ignoring diacritic marks in the index and query text. So, Atlas Search returns only results that match the strings with or without diacritics in the query. For example, a search for cafè returns results only with the characters cafè. A search for cafe returns results only with the characters cafe.

true