Customizing Ranking Models for Enterprise Search - Ammar Haris & Joe Zeimen, Salesforce

https://bit.ly/2A0dGnp Enter to see the secret sale page at GSA-Online.
Solr provides a suite of built-in capabilities that offers a wide variety of relevance related parameter tuning. Index and/or query time boosts along with function queries can provide a great way to tweak various relevance related parameters to help improve the search results ranking. In the enterprise space however, given the diversity of customers and documents, there is a much greater need to be able to have more control over the ranking models and be able to run multiple custom ranking models. At Salesforce, we have a multi-level ranking pipeline, first ranker (L1), is the basic lucene scoring based on tf-idf and the second ranker (L2), implements more complex ranking models ranging from something as trivial as a linear regression to the more complex models such as a boosted decision tree. This L2 ranker inside Solr enables us to extract features for every document from within the Solr Index and leverage them during ranking model execution. This talk discusses the motivation behind creating an L2 ranker and the use of Solr Search Component for running different types of ranking models. Presented at Lucene/Solr Revolution. Learn more: https://activate-conf.com/