Add DIGER (SIGIR 2026) to Semantic Identifiers Augmentation

[junchen-fu]

Hi, thank you for maintaining this useful survey and taxonomy of pure and augmented LLM recommenders!

We would like to suggest adding our SIGIR 2026 full paper:

DIGER: Differentiable Semantic ID for Generative Recommendation

• Paper: https://dl.acm.org/doi/10.1145/3805712.3809641
• Code: https://github.com/junchen-fu/DIGER

DIGER introduces differentiable semantic IDs that are jointly optimized with the generative recommendation model. Compared with approaches that generate semantic IDs using a separately trained and frozen tokenizer, DIGER allows recommendation supervision to directly refine the item identifiers.

The complete implementation and reproduction artifacts are publicly available, including code, processed data, semantic embeddings, RQ-VAE checkpoints, and reproduction scripts.

We believe DIGER fits the following section:

Augmented LLM Recommenders → Semantic Identifiers Augmentation

Suggested entry:

Venue	Code	Paper
SIGIR'26	Code	Differentiable Semantic ID for Generative Recommendation

Thank you for considering this addition!