Search papers, labs, and topics across Lattice.
The paper introduces VaLiDRec, a generative recommendation framework that utilizes variable-length, LLM-aligned semantic identifiers (SIDs) to enhance item representation. By constructing SIDs from native LLM vocabulary tokens and employing techniques like token importance estimation and semantic-quality-aware pruning, VaLiDRec adapts identifier lengths to the complexity of item semantics. Experimental results reveal that VaLiDRec outperforms existing recommendation systems across multiple datasets, achieving significant improvements in cold-start performance and inference speed.
LLM-aligned variable-length identifiers can revolutionize generative recommendation by providing a more nuanced and efficient representation of item semantics.
Generative recommendation commonly represents items using fixed-length semantic identifiers (SIDs) constructed through clustering and quantization. However, these artificial codes may overcompress item semantics, remain misaligned with pretrained LLM vocabularies, and require costly autoregressive decoding. In light of this, we propose VaLiDRec, a generative recommendation framework based on variable-length, LLM-aligned semantic identifiers. VaLiDRec constructs SIDs directly from informative native LLM vocabulary tokens via token importance estimation, semantic-quality-aware pruning, and collision-aware refinement, allowing identifier lengths to adapt to item semantic complexity. To model user preferences, VaLiDRec incorporates graph-aware soft prompts and reformulates recommendation as token-set prediction with token-level item scoring, eliminating autoregressive SID generation and beam search. Experiments on four real-world datasets show that VaLiDRec consistently outperforms strong sequential and generative recommendation baselines across all evaluation metrics. It further achieves superior zero-shot item cold-start performance and 87.49$\times$ faster inference than LC-Rec. These results demonstrate that LLM-native variable-length semantic identifiers provide a more expressive and efficient paradigm for generative recommendation.
