Transformer attention in generative recommenders can create positional, popularity, latent-driver, and synthetic-data biases that concentrate exposure and reduce diversity.
If all entriesbpjk are positive, choose logits (W ⊤ j,:)k = logbpjk (up to an additive constant), such that softmax(W ⊤ j,:) =bpj
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CY 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
LLM Biases
Transformer attention in generative recommenders can create positional, popularity, latent-driver, and synthetic-data biases that concentrate exposure and reduce diversity.