LinkedIn Researchers Present New Findings at KDD2016
August 15, 2016
This week, researchers and scientists in areas like machine learning, data mining, and analytics are meeting in San Francisco for the 22nd Annual ACM SIGKDD Conference (KDD 206). Several of our researchers and experts will be attending this year to present papers, workshops, and tutorials on topics from university rankings to generalized machine learning models. In the interest of promoting the great work that’s being done by our experts, we wanted to provide blog readers with a sampling of some of the research that LinkedIn employees are presenting this year.
First up, Ya Xu and Nanyu Chen discuss how to measure mobile app features individually through randomized A/B tests. They note how A/B tests on mobile apps are conducted very differently from tests on the web because of the lengthy build, review, and adoption process for app release. They also use data from a major launch at LinkedIn as a case study for their approach.
Next, Haishan Liu, David Pardoe, and Kun Liu introduce Audience Expansion. This is a technique developed at LinkedIn to identify new audiences similar to the original target group. With this technique, they simplified targeting process and increased reach for advertisers. This technique also realized better utilization of ads inventory and more efficient market participation.
Shipeng Yu and Abishek Gupta, in collaboration with Evangelia Christakopoulou from the University of Minnesota, are introducing a system for identifying decision makers within social networks. Their paper introduces the LinkedIn Decision Maker Score (LDMS) to quantify the ability of a member to make a sales decision. This is the key data-driven technology underlying Sales Navigator, a LinkedIn product that is designed for sales professionals.
On the machine learning front, we’re also presenting several papers and hands-on tutorials on tools like PhotonML and concepts like recommendation systems. Also of note is research on using Generalized linear mixed models (GLMix) for large-scale response prediction and how our machine learning team scaled these models to generate 20-40 percent more job applications by more accurately improving job recommendation data. Finally, Rupesh Gupta and several other researchers show how we used machine learning to dramatically and effectively scale back email communications while still maintaining the same quality of communication with our members.