Handling Compound Words

Compound words (such as "skyscraper") are words actually compound from two or more other words ("sky" and "scraper"). For searching these kind of words are quite relevant, as most search engines will easily be able to look up words beginning with a string ("sky") but not with a string at other places (such as "scraper" in "skyscraper").

For SES there are several mechanisms in place to handle this kind of words.

Dictionaries

When indexing your product catalogue, SES will search for language dictionaries at several places. By default it will handle dictionaries in /usr/share/hunspell/ and %PLUGIN_DIR%/Resources/assets/. Furthermore you are able to maintain custom dictionaries in %PLUGIN_DIR%/Resources/assets/custom_dictionaries/.

During the indexing process, SES will build the ngram of each indexable field and will compare each ngram with the dictionary. This will make indexing a bit slower - but makes sure that only high quality search terms are indexed. With this mechanism, "skyscaper" can be decompound into "sky" and "scraper" and therefore be indexed.

Dictionary format

By default SES expects the dictionaries to be available in the three folders mentioned above. Each dictionary should be named by the convention %LANGUAGE_CODE%.dic, for example de_DE.dic. Within the dictionary SES expects the words to be line separated, for example:

sky
scraper

Extending the dictionary

The default dictionaries SES considers are the dictionaries provided by hunspell. SES also comes with dictionaries for de_DE and en_GB if hunspell is not available on your system. In order to extend these dictionaries or add custom ones, you can just create the corresponding files in %PLUGIN_DIR%/Resources/assets/custom_dictionaries/. SES will handle those files as additions to the default dictionaries. For that reason there should be no need to modify any other dictionary files in SES directly.

Custom dictionary handler

If you want to provide dictionaries in another format, you can do so by implementing the interface \SwagEnterpriseSearch\Bundle\EnterpriseSearchBundle\Dictionary\DictionaryInterface. Then just set your implementation into \SwagEnterpriseSearch\Bundle\EnterpriseSearchBundle\Dictionary\DictionaryManager using setDictionaries. DictionaryManager will - for every given language key - search for an implementation supporting the given language key. The first implementation matching, will be responsible for handling the language in question.

If you want your implementation to be considered before SES's default dictionary implementation, you should make it the first one in the array passed to setDictionaries.

Using the dictionary index

In order to use the decompound words from the dictionaries, make use of the search field attributes.enterprise.product_suggestion in the relevance configuration of SES.

n-gram

ElasticSearch also allows to index n-grams directly. On the one hand, this will not require an additional lookup against dictionaries while indexing, on the other hand, this will massively bloat the search index with non meaningful search terms such as "ysc" or "ape" for "skyscraper". By default Shopware will index ngrams with a width from 3 to 8. If you want to change this (to have larger ngrams or to turn them of generally in order to speed up indexing), you can do this as follows:

// config.php
return [
    'db' => // your database config,
    'es' => [
        // your default ES config
        'compound_filter' => [
            'enabled' => true,                  // enable / disable ngrams
            'type' => 'ngram',                  // type of ngram, e.g. 'ngram' or 'edge_ngram'
            'min_gram' => 3,                    // minimum width
            'max_gram' => 8,                    // maximum width
            'token_chars' => ['letter'],        // characters to tokenize
        ],
    ]
];

Using n-grams

In order to use the n-gram index, make use of the *.ngram fields in the relevance configuration of SES.

Top