User Semantic Embedding in Social Media Graphs.
Develop Self-Supervised method to learn high quality semantic embedding of social media users, purely based on their activity on the graph and without relying on labelled data.
Requirements
- M.Sc. in Machine Learning, Computer Science, Mathematics, Physics, Engeneering, or similar
- Basic knowledge of graph theory
- Decent familiarity with transformers architecture and self-attention mechanism
- Familirity with Python and Pytorch
Description
Characterizing Social Media Users is an important task that can be used for several goals, the less appealing (to us) being advertisment while the most being online content moderation and mitigating online hate spreading. The vast majority of methods addressing this problem rely solely on the text posted by the User online. Hoewer, when building a User embedding the topology of the graph of connections it is also of great importance. In this thesis we want to develop a novel** graph-attention mechanism** tailored for social media graphs, together with and efficient method for self-learning that requires no label data and is able to generate high quality user semantic embeddings.
This thesis can be divided into three main parts: (1) litterature review of transformers methods on graph and self learning methods; (2) set up of appropriate testing framework (i.e. appropriate datasets) and (3) development of novel social media graph attention mechanism and self-learning methods.