Implement causal reweighting and forgetting methods

[sarahxdean]

Papers about updating recommender models online:

• Factorization in Recommender Systems https://dl.acm.org/doi/pdf/10.1145/2695664.2695820?casa_token=2pQnZSBN5pYAAAAA:5Iio1nLAUzohhcKLEv6z4gC8DACvlKL7rHx-JRUOuDg0TEo7SEpol3u7RFK0LnNfJSwA5MG8Jw5w8mM
  • May want to implement a simple forgetting method (natural way to deal with drifting preferences)
• Causal Embeddings for Recommendation https://arxiv.org/abs/1904.05165
  • May want to implement something simple discussed in the related work for this paper (rather than the more complicated method), like clipped inverse propensity scoring
• Unbiased Recommender Learning from Missing-Not-At-Random Implicit Feedback https://dl.acm.org/doi/pdf/10.1145/3336191.3371783?casa_token=kqBfDSmF_G8AAAAA:tnEMQea84s363LpEBHIDfagc9VEyqxOQ4cAT4E0saEYN6wwmYeWSq1lqb0B7MFzJc3rhIaC7kI8c6P4
  • Looks like it has a decent review of related reweighting methods -- we can pick a simple one or two to try