Solr suggester example how to#

One shortcoming of this design is that there is no way to add a new item in the structure without rebuilding it completely. We then call the build method to clear the in-memory structure and populate it with all the items from the source. In general, this will either be a simple text file or the indexed terms of a field from a Lucene index. To build the structure, we have to set a source for suggestions. But all those classes share an interesting characteristic: the structure is held entirely in memory, making the lookup very fast! Those classes use one of two structures to hold the suggestions: a ternary search tree (TST) or a finite state automata (FST). In this module, Lucene provides several classes extending Lookup. The Lookup abstract class is a simple one that has a lookup method to return suggestions from an input. This is what Lucene provides in the suggest module. Users expect autocomplete to be fast!Ī better approach is to have a dedicated and optimized structure to provide those autocomplete suggestions from a given input. As the index grows bigger (all those n-grams add up!), we might experience performance issues. A Lucene index is not ideal for this task as it will have to look for a lot of terms before retrieving the n-grams. The main drawback of this technique is that it is using an index to hold the n-grams.

For example, you can read this popular tutorial by Jay Hill to implement it in Solr, and this one by Jon Tai for Elasticsearch.

Solr suggester example how to#

You can already find detailed instructions on how to implement the edge n-grams technique around the Web. This technique is very flexible as it allows for easy implementation of interesting functionalities, such as fuzzy search and infix matching. For example, if we send the query sea, the index should return the word search as sea is an n-gram of this stored word. When the user starts typing a word, we send what is typed so far as a query to the index containing the n-grams. The general principle when using them to support autocomplete is to index all those n-grams in the search index with the original word stored as is. For example, the edge n-grams of the word search are s, se, sea, sear, searc and search. Edge n-grams are subsets from one edge of a word (generally the beginning).

Both allow for implementing autocomplete using edge n-grams. Open source search platforms like Solr and Elasticsearch support this feature. It is generally used to return either query suggestion (à la Google Autocomplete) or to propose existing search results (à la Facebook). Autocomplete (also known as live suggestions or search suggestions) is very popular with Search applications.