Abstract: Choosing the correct optimization objective is key to success of a search engine. Unlike traditional web searches, where clicks are clearly the main objective to optimize, many emerging search engines like product search may require a different goal to achieve, such as more conversions, revenue, and higher quality. Selection of a good metric may depend on many factors such as query type (transactional vs. navigational vs. informational), and goal of a business (profitability vs. growth). For example, a typical product search engine may focus on maximizing the number of transactions and total revenue, while navigational search may aim at minimizing the total number of clicks. In this talk, we will investigate factors needed to be considered when we design objective functions of a product search engine, and we will also walk through a case study based on a particular business, including how an objective can be selected, mathematically defined, and optimized with a machine learning framework.
Bio: Liang Wu, PhD. is a machine learning data scientist at Airbnb. His research focuses on product search and web search. His dissertation work concentrated on building robust machine learning systems with noisy, inaccurate and biased data in the context of social media. He has published over 30 papers in major artificial intelligence conferences including AAAI, CIKM, ICDM, ICWSM, SDM, SIGIR and WSDM, and his solution has won the third place on the leaderboard in KDD Cup 2012. He serves as a Program Committee member for AAAI, SIGIR, KDD, WSDM, CIKM, BIGDATA, etc, and he was the main lecturer for tutorials in SBP'16 and ICDM'17. He obtained his Ph.D. from Arizona State University, and his master and bachelor from Univ. of Chinese Academy of Sciences, and Beijing Univ. of Posts and Telecom, respectively. He is an author of 2 book chapters and 4 patents in China.