Personalised Interactive Behaviour Embedding

Description: Many works have learnt generalised representations of interactive behaviour. However, most of them did not preserve the key characteristics of users themselves. User information is essential in developing personalised intelligent user interfaces or recommendation systems. This project is to learn personalised interactive behaviour embeddings.

Goal: Leveraging Transformer-based or Diffusion models to learn user embeddings through explicit (e.g., questionnaires) or implicit (e.g., behaviour) user information.

Supervisor: Guanhua Zhang

Distribution: 20% Literature, 30% Data preparation, 50% Deep learning

Requirements: Strong deep learning skills or mathematics, experience in Pytorch

Literature:

[1] Zhang et al. 2023. Exploring Natural Language Processing Methods for Interactive Behaviour Modelling. IFIP INTERACT'23.</i>

[2] Gan et al. 2022. HiGAN+: Handwriting Imitation GAN with Disentangled Representations. TOG

[3] Yu et al. 2023. Untargeted Attack against Federated Recommendation Systems via Poisonous Item Embeddings and the Defense. AAAI'23.

[4] Liang et al. 2023. Cross-Attribute Matrix Factorization Model with Shared User Embedding.