Warning: Altis v11 is no longer supported.

Tuning Relevancy

Altis provides a set of default configuration for relevancy scoring based on experience and customer feedback.

Depending on your use case and requirements, you may wish to tune relevancy to provide more relevant results to visitors. The following built-in configuration options aim to provide some high-level levers for tuning.

Note: Relevancy scoring is a highly subjective process, and Altis support cannot assist with this except for clearly identified bugs. The Altis team can help you find partners with Elasticsearch experience if needed.

Search Mode

The mode setting determines whether or not the search will allow the use of advanced search syntax. The default mode is "simple" search. To enable advanced mode you would use the configuration below.

{
	"extra": {
		"altis": {
			"modules": {
				"search": {
					"mode": "advanced"
				}
			}
		}
	}
}

By setting mode to "advanced" it is possible to write more complex search terms using operators:

  • Quoted strings will force an exact match
  • Brackets ( and ) denote precedence of terms
  • AND or a plus + character will match both the terms
  • OR or a pipe | character will match either of the terms
  • A word followed by * will match anything with that prefix
  • A word preceded by - will be excluded from the results

For example the following search would return results with exact matches for "the law", or the words "judge" and "dredd" and excluding the word "marvel".

"the law" OR (judge AND dredd) -marvel

By default searching will match all the provided search terms.

This means the more specific a search query is the fewer results will be provided. For example "the quick brown fox" will be interpreted as "the AND quick AND brown AND fox".

Setting strict to false will change the behaviour to match each individual word as well as the full search query, with complete matches being scored more highly.

{
	"extra": {
		"altis": {
			"modules": {
				"search": {
					"strict": false
				}
			}
		}
	}
}

Strict matching is recommended if you have a user interface that allows sorting search results by anything other than relevance such as date.

Field Boosting

By default the post title, excerpt, content, author name and taxonomy terms are searched against. Field boosting allows you to modify the importance of those fields with regards to search results. The following configuration will increase the important of the post title, excerpt and a custom meta field above the default value of 1.

{
	"extra": {
		"altis": {
			"modules": {
				"search": {
					"field-boost": {
						"post_title": 3,
						"post_excerpt": 2,
						"meta.custom_field.value": 2
					}
				}
			}
		}
	}
}

It is important to note that the field names should match the fields in the elasticsearch index and do not always correspond 1:1 with database fields.

Fuzzy Matching

By default some degree of fuzzy matching is allowed so that simple spelling errors can still return some results, for example "breif" would match "brief". This can sometimes result in unwanted search results however with short words and acronyms.

Fuzzy matching works by providing an edit distance as an integer from 0-2. The number indicates how many edits are allowed for a term to match. Edits can be one of the following:

  • Changing a character (box → fox)
  • Removing a character (click → lick)
  • Inserting a character (sic → sick)
  • Transposing two adjacent characters (act → cat)

Note: when using "advanced" search mode quoted strings will not use fuzzy matching.

Note: if you have synonyms configured they will not use fuzzy matching.

Below is the default fuzzy search configuration.

{
	"extra": {
		"altis": {
			"modules": {
				"search": {
					"fuzziness": {
						"distance": "auto:4,7",
						"prefix-length": 1,
						"max-expansions": 40,
						"transpositions": true
					}
				}
			}
		}
	}
}

distance

This is the Elasticsearch fuzziness value and determines the maximum edit distance for fuzzy matching. This value can also be set as the value for fuzziness directly if you do not need to edit the other options.

  • If an integer is provided this is used as a fixed edit distance eg. 0, 1 or 2 edits allowed
  • If a string is provided it must be in the form auto:[min],[max]
    • If min is 3 all terms from 0-2 characters long will have an edit distance of 0
    • If max is 6 all terms from min to 6 characters long will have an edit distance of 1
    • Any terms longer than max will have an edit distance of 2

prefix-length

This value determines how many characters at the start of a search term must match before fuzzy matching is applied. For instance a prefix length of 1 will mean that "lotion" will match "lotoin" but not "potion" or "motion".

max-expansions

This value is the number of fuzzy terms generated from a search term when used for matching. Higher values will result in slower searches but more matches and lower values will result in faster searches but fewer matches.

If you have a prefix length of 0 you may wish to increase this value to get more matches.

transpositions

This option allows you to prevent transpositions from being counted as a single edit. In the above example the transposition "act" to "cat" has an edit distance of 1. Setting this value to false would mean the same transposition would have an edit distance of 2 because 2 letters have been replaced.

Max Query Length

Typically search queries can be as long as the user wishes. As a preventative measure against slow queries and search request spamming, Altis limits search query strings to 100 characters by default.

You can configure this value in 2 ways.

Using the Altis Search config option max-query-length:

{
	"extra": {
		"altis": {
			"modules": {
				"search": {
					"max-query-length": 150
				}
			}
		}
	}
}

Using the filter altis.search.max_query_length:

Note that the filter method will take precedence and can be used for applying conditional logic.

add_filter( 'altis.search.max_query_length', function ( int $max_length, string $search ) {
	// Allow long search strings in the admin.
	if ( is_admin() ) {
		return 1000;
	}

	// Allow long search queries if they are URLs.
	if ( preg_match( '#https?://[a-z0-9/_-.]+#', $search ) ) {
		return mb_strlen( $search );
	}

	return 50;
} );

Date Decay

For most use cases, you'll want to treat newer results as more relevant than older ones. Altis includes date decay functionality which automatically adjusts relevancy scores based on dates.