Custom analyzers in Elasticsearch - Step-by-Step Execution

Concept Flow - Custom analyzers

Define tokenizer

↓

Define filters

↓

Create custom analyzer

↓

Apply analyzer to text

↓

Tokenize text

↓

Apply filters to tokens

↓

Output processed tokens

Custom analyzers combine a tokenizer and filters to process text into tokens for searching.

Execution Sample

Elasticsearch

PUT /my_index
{
  "settings": {
    "analysis": {
      "analyzer": {
        "my_custom_analyzer": {
          "type": "custom",
          "tokenizer": "standard",
          "filter": ["lowercase", "asciifolding"]
        }
      }
    }
  }
}

This code creates an index with a custom analyzer that tokenizes text, lowercases it, and removes accents.

Execution Table

Step	Action	Input Text / Tokens	Output Tokens
1	Input text to analyze	"Café Déjà Vu"	"Café Déjà Vu"
2	Tokenize with standard tokenizer	"Café Déjà Vu"	["Café", "Déjà", "Vu"]
3	Apply lowercase filter	["Café", "Déjà", "Vu"]	["café", "déjà", "vu"]
4	Apply asciifolding filter	["café", "déjà", "vu"]	["cafe", "deja", "vu"]
5	Output final tokens	["cafe", "deja", "vu"]	["cafe", "deja", "vu"]

💡 All filters applied; final tokens ready for indexing or searching.

Variable Tracker

Variable	Start	After Tokenize	After Lowercase	After Asciifolding	Final
tokens	N/A	["Café", "Déjà", "Vu"]	["café", "déjà", "vu"]	["cafe", "deja", "vu"]	["cafe", "deja", "vu"]

Key Moments - 3 Insights

Why do tokens change after applying the lowercase filter?

Because the lowercase filter converts all tokens to lowercase, as shown in step 3 of the execution_table where tokens change from ["Café", "Déjà", "Vu"] to ["café", "déjà", "vu"].

What does the asciifolding filter do to tokens?

It removes accents and converts characters to their ASCII equivalents, as seen in step 4 where ["café", "déjà", "vu"] becomes ["cafe", "deja", "vu"].

Why is the tokenizer step important before filters?

Because the tokenizer splits the input text into tokens, which filters then process. Without tokenization (step 2), filters cannot work on individual words.

Visual Quiz - 3 Questions

Look at the execution_table, what are the tokens immediately after tokenization (step 2)?

A ["cafe", "deja", "vu"]

B ["Café", "Déjà", "Vu"]

C ["café", "déjà", "vu"]

D ["Cafe", "Deja", "Vu"]

💡 Hint

Check the Output Tokens column at step 2 in the execution_table.

At which step do tokens lose their accents?

A Step 4 - Asciifolding filter

B Step 3 - Lowercase filter

C Step 2 - Tokenize

D Step 5 - Output final tokens

💡 Hint

Look at the change from accented to unaccented tokens in the Output Tokens column.

If we remove the lowercase filter, what would be the tokens after asciifolding?

A ["cafe", "deja", "vu"]

B ["Café", "Déjà", "Vu"]

C ["Cafe", "Deja", "Vu"]

D ["café", "déjà", "vu"]

💡 Hint

Consider that asciifolding removes accents but does not change case; lowercase filter changes case.

Concept Snapshot

Custom analyzers in Elasticsearch combine a tokenizer and filters.
Tokenizer splits text into tokens.
Filters modify tokens (e.g., lowercase, asciifolding).
Define in index settings under analysis.analyzer.
Used to control how text is indexed and searched.

Full Transcript

Custom analyzers in Elasticsearch let you control how text is broken into tokens and processed. First, a tokenizer splits the text into words or tokens. Then, filters change these tokens, for example by making them lowercase or removing accents. In the example, the text "Café Déjà Vu" is tokenized into ["Café", "Déjà", "Vu"]. The lowercase filter changes these to ["café", "déjà", "vu"]. The asciifolding filter then removes accents, resulting in ["cafe", "deja", "vu"]. This process helps make searching more flexible and accurate. You define custom analyzers in the index settings under the analysis section, specifying the tokenizer and filters you want to use.