Autocomplete as Relevancy

Rimple Shah • Location: Theater 4Back to Haystack 2019

View the Slides View the Video

Autocomplete is a staple feature for search applications. This feature (also called auto-suggest, search-as-you-type, or type-ahead) has become an expected part of an engaging, user-friendly search experience. Solr provides "out-of-the-box" autocomplete support called Solr Suggester, but we found this not flexible enough to meet application-specific requirements.

Instead, autocomplete can be cast as a traditional search relevance problem, which enables us to leverage Solr's customizability. This talk presents an approach to the autocomplete problem by treating suggestions as individual documents. The practical challenges of creating this kind of autocomplete system are addressed, including data collection, data quality, suggestion ranking, and system configuration. As with most search problems, there is a high degree of subjectivity in evaluating the relevance of autocomplete suggestions.

These issues are not domain-specific, but rather are of interest to a wide range of search application developers. In this presentation, we discuss our implementation of the autocomplete system in Lexis Advance. We also discuss multiple methods for evaluating the relevance of our autocomplete suggestions.

Rimple Shah