In this talk, I propose a robust approach for recommending privacy labels by extracting features based on TF-RF method using small data sets. Random Forest classifier predicts privacy settings of images using TF-RF method. However, if a tag of a test image does not exist in the set of unique tags of the training data, the classifier does not have any information about the privacy status of this tag. Each tag has vector representations by using pre-trained word embedding models such as BERT, Word2Vec, GloVe.
Affiliation: Utrecht University
Gönül Aycı is a Ph.D. candidate at Bogazici University and a researcher at Utrecht University (from Dec. 2021 to Dec. 2022). Her research interests include privacy in online social networks and uncertainty modeling. She is a content creator about AI on YouTube. She is a co-organizer of Django Girls Istanbul and PyCon Turkey. She is also Grace Hopper Celebration scholar for 2020 and 2021.