When Fashion Meets Machine Learning: Automatic Authorship Identification on Fashion Social Network

The Author profiling task identifies characteristics such as gender, age, or personal traits by analyzing their written works. With the growth of online social networks, multimodal genres such as website posts, tweets, or user comments have arisen. However, traditional approaches focus on text corpus only. Moreover, the author profiling task on reviews and comments corpus is hard to accurately predict using traditional methods. In this work, we are devoted to dealing with author profiling tasks on online fashion social networks. First, we construct a fashion comment dataset named NetiLook Authors - a user comments dataset from the Lookbook website for the author profiling task, and uncover useful insights. In addition, we leverage and investigate the merit of multimodal data integration, which is able to make up the insufficiency of conventional works. Furthermore, we proposed a Multimodal Comment Profiling Network (MCPN) for the author profiling task of user comments by utilizing multimodal tensor product fusion. Experimental results demonstrate our proposed approaches benefit author profiling task on fashion social networks.