Ahead of his presentation at the Sentiment Analysis Innovation Summit in San Francisco on April 29 & 30, we spoke to Ben Taylor, Chief Data Scientist at HireVue about his role and his thoughts on HR analytics.
What are the most important recent innovations in NLP?
Deep learning. Deep learning has enabled greater accuracy than previous modeling approaches. Deep learning has also allowed for text to be reduced to more accurate context summaries such as personality or subject matter for additional analytics.
Why have you focused on HR?
HR is a new frontier for data science. The problems are messy with missing data and unstructured problems. Doing digital interview ranking is the most challenging and exciting problem I have worked on in my career so far. It is also very rewarding to work to enable candidates, who are qualified, to get jobs that may have not been considered for previously due to subjective screening processes.
What is the potential for NLP in the future?
Skynet? Nirvana? NLP in the future can take the famous Turing test to the next level where humans can't distinguish the computer at all. If NLP is capable of dynamically engaging humans in conversation it opens up opportunities in multiple fields. Dynamic interviewing between a human and a computer psychologist or physician.
What are you going to discuss in your talk at the Sentiment Analysis Innovation Summit?
I will demonstrate some of the text features beyond just simple word or n-gram use. Automated feature engineering will also be demonstrated using deep learning techniques. Finally, after generating 100,000s of features how do you reduce the features effectively to reduce computation costs and increase accuracy? Metaheuristic feature reduction on distributed systems will be demonstrated to address automated feature reduction. Lastly, examples and data analysis will be presented with real interview data on one of the world's largest digital repositories of recorded interviews.
At HireVue we are doing digital interview prediction and analytics. Many companies use resume ranking algorithms to filter top candidates to the top of their sourcing lists. Doing the same with the interview is a much more challenging, and rewarding problem to solve. Also, would you ever hire someone from just a resume or an assessment sight unseen? No, of course not, you would use the interview, so why not focus our modeling and data science abilities on that.