Neighborhood-Aware Unlearning for Generative Recommendation

Key points

• If training interactions must later be removed, such as injected spam or interactions with items a user no longer wants, this sharing lets an unlearning update unintentionally affect the predictions for the interactions' neighbors.1
• We show that this effect is locally structured and hidden by global metrics: unlearning can leave recommendation utility on the whole test set nearly unchanged while substantially degrading it on neighborhoods of the forgotten interactions' SIDs.1
• We propose Neighborhood-Aware Unlearning (NAU), which augments interaction-level unlearning with explicit control over the probability mass assigned to the forgotten items' neighborhoods.1
• We identify local probability redistribution as an important failure mode of unlearning in SID-based generative recommenders and show that explicitly controlling this redistribution can mitigate its effects.1

Sources (1)

• [1]Neighborhood-Aware Unlearning for Generative Recommendation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 08:00 PM

  Generative recommendation predicts items by autoregressively generating semantic identifiers (SIDs), whose tokens are shared across semantically related items.

  If training interactions must later be removed, such as injected spam or interactions with items a user no longer wants, this sharing lets an unlearning update unintentionally affect the predictions for the interactions' neighbors.

Extractive summary: sentences quoted from the sources.