In 2018, Stanford University launched a text comprehension challenge SQuAD. The R-NET model submitted by Natural Language Computing Group of Microsoft Research Asia earned top marks 82.650 in EM value, which for the first time ever surpassed human performance. Furu Wei was the one who led the research and had worked in Microsoft Research Asia for eight years by then.
During his PhD, Wei served as a visiting scholar in the department of computer science in Hong Kong Polytechnic University and took up research related to natural language processing. Later he entered the Natural Language Computing Group at Microsoft Research Asia. In the past 10 years, Wei’s research has dealt with multiple key projects in natural language processing, including intelligent question answer dialogue, machine reading comprehension and abstraction, etc. Wei’s research findings have long been applied to practical areas, such as chatbot Xiaoice, Bing search engine, Office software and Microsoft Cognitive Services.
However, Wei’s findings do not stop at machine reading comprehension. They even include recognizing and comprehending feelings, opinions, and emotions from texts to predict changes in user emotions on social networks more precisely via computer, monitor attitudes and opinions from online comments towards monitoring targets such as products and brands, and build human-like interpersonal relationship in the conversation and interaction between chatbots and humans. All of this enabled Wei’s products to achieve more powerful functions and better user experiences.
Wei said, “Little is known about the rationale and interpretation of natural language understanding. I truly hope that with scientific efforts, we can make machines understand natural language to bring intelligent and convenient computing experiences and serve people and societies better.”