Oana Ignat Towards Human Action Understanding in Social Media Videos Using Multimodal Models University of Michigan, USA 2022 https://hdl.handle.net/2027.42/174613