Lucene/Solr Revolution 2017 has ended
View analytic
Thursday, September 14 • 2:20pm - 3:00pm
Learning to Rank from Clicks
This talk introduces the mathematics behind the effort at Salesforce to improve search relevance using machine learning.

First, we discuss a new approach for search ranking. Much of the active research in this field has focused on the use of a large amount of labeled data. Specifically, human beings need to sift through tens of thousands of queries---along with returned results---and mark each result on a scale from `not relevant' to `extremely relevant'. At Salesforce, we have developed a novel machine learning approach, for learning a ranking function, that does not require labeled data.
Second, we discuss the mathematical foundations of our AB testing program. Since user clicks do not follow any standard distribution, we rely on extensive simulations to estimate the power of AB tests and evaluate the significance of the results.

avatar for Zach Alexander

Zach Alexander

Principal Data Scientist, Salesforce.com
Zach has extensive experience as a data scientist. He has worked in a variety of fields, including: hard drive reliability (Seagate Technology), voice-over-IP (Skype) and enterprise search (Salesforce.com). He holds a Ph.D. in Applied Math and has been heavily involved in data science... Read More →
avatar for Tracy Backes

Tracy Backes

Data Scientist, Salesforce.com
Tracy Backes is a Data Scientist with the Data Science for Communities, Service and Search team at Salesforce. She has been at Salesforce for one and a half years, where she works to build data-driven solutions for projects related to search relevance, sales forecasting and customer... Read More →

Thursday September 14, 2017 2:20pm - 3:00pm
South Seas A