User identification in social media is of crucial interest for companies and organizations for purposes of marketing, e-commerce, security and demographics. In this paper, we aim to identify users from Pinterest, a platform where users post pins, a combination of an image and a short text. This type of multi-modal content is very common nowadays, since it is a natural way in which users express their interests, emotions and opinions. Thus, the goal is to identify the user that would post a particular pin. For solving the problem, we propose a two-phase classification model. In a first phase, we train independent classifiers from image data, using a deep learning representation, and from text data, using a bag-of-words representation. During testing we apply a cascade fusion of the classifiers. In a second phase, we refine the output of the cascade for each test pin by selecting the top most likely users for the test pin and re-weighting their corresponding output in the cascade by their similarity with the test pin. Our experiments show that the problem is very hard because several reasons with the data distribution, but they also show promising results.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.