Learning Decomposed Contextual Token Representations from Pretrained and Collaborative Signals for Generative Recommendation
📰 ArXiv cs.AI
Learn how to improve generative recommenders by combining pretrained and collaborative signals for contextual token representations
Action Steps
- Tokenize items into semantic IDs using a pretrained tokenizer
- Train large language models (LLMs) to generate the next item via sequence-to-sequence modeling
- Combine pretrained and collaborative signals to learn decomposed contextual token representations
- Optimize the two-stage paradigm for a unified objective
- Evaluate the performance of the generative recommender using metrics such as precision and recall
Who Needs to Know This
Data scientists and AI engineers working on generative recommenders can benefit from this approach to improve their models' performance and accuracy
Key Insight
💡 Combining pretrained and collaborative signals can improve the accuracy and performance of generative recommenders
Share This
🤖 Improve generative recommenders with combined pretrained and collaborative signals for contextual token representations! 🚀
Key Takeaways
Learn how to improve generative recommenders by combining pretrained and collaborative signals for contextual token representations
Full Article
Title: Learning Decomposed Contextual Token Representations from Pretrained and Collaborative Signals for Generative Recommendation
Abstract:
arXiv:2509.10468v2 Announce Type: replace-cross Abstract: Recent advances in generative recommenders adopt a two-stage paradigm: items are first tokenized into semantic IDs using a pretrained tokenizer, and then large language models (LLMs) are trained to generate the next item via sequence-to-sequence modeling. However, these two stages are optimized for different objectives: semantic reconstruction during tokenizer pretraining versus user interaction modeling during recommender training. This
Abstract:
arXiv:2509.10468v2 Announce Type: replace-cross Abstract: Recent advances in generative recommenders adopt a two-stage paradigm: items are first tokenized into semantic IDs using a pretrained tokenizer, and then large language models (LLMs) are trained to generate the next item via sequence-to-sequence modeling. However, these two stages are optimized for different objectives: semantic reconstruction during tokenizer pretraining versus user interaction modeling during recommender training. This
DeepCamp AI