Eugene Yan • 9/14/2025

Training an LLM-RecSys Hybrid for Steerable Recs with Semantic IDs

This technical article details the process of creating a hybrid model that combines a language model with a recommender system using Semantic IDs. It explains how extending an LLM's vocabulary with semantic tokens and training on user behavior sequences enables a single model to generate recommendations and be steered via natural language conversation, offering reasoning and explanations.

0 comments

#llm #Language Models #Recommender Systems