PrefixMem encoder for Semantic IDs improves deepest-level accuracy by up to 46% relative and full-SID retrieval recall by up to 22% relative on Pinterest data across LLM families.
Openonerec technical report
13 Pith papers cite this work. Polarity classification is still indexing.
13
Pith papers citing it
years
2026 13representative citing papers
Semantic-ID tokenizers produce collisions affecting up to 30.5% of items across four datasets, inflating Hit@10 by up to 103.36% and making prior tokenizer comparisons unreliable.
RecRM-Bench is a new large-scale benchmark dataset and framework for multi-dimensional reward modeling in agentic recommender systems, spanning instruction following, factual consistency, query-item relevance, and user behavior prediction.
A single LLM trained to emit semantic item codes can fulfill complex shopping intents with fewer tool hand-offs, improving multi-turn follow-up on Taobao-derived tasks.
IntuRec anchors LLM latent reasoning for recommendation by deriving an intuition embedding from top-K candidates via self- and cross-attention to initialize more accurate trajectories.
NaviGen encodes user behavior via dual collaborative-textual identifiers and applies SFT+RL to produce personalized multimodal outputs and better instructions from interaction history.
LLMs for generative recommendation show heavy one-hop memorization that accounts for most gains over baselines, and IIRG training that incorporates multi-hop co-occurrences and semantic relations yields larger gains on non-memorizable cases.
A Llama-based model trained on serialized user stories unifies item, carousel, and search ranking and outperforms specialist baselines offline while improving some online metrics and reducing latency.
Uni-OPD improves on-policy distillation via student-side data balancing for informative rollouts and teacher-side outcome-guided margin calibration that restores order consistency with rewards.
GR2 applies mid-training on semantic IDs, reasoning distillation, RL with conditional verifiable rewards, and a context compressor to re-ranking in industrial recsys, reporting +18.7% R@1 over baselines.
VirtualMLE deploys an LLM agent with execution-reflection-memory to tune sequential recommenders, reaching competitive quality on Amazon benchmarks with fewer trials and transferring heuristics across datasets.
Ghost mitigates popularity bias in generative recommenders through asymmetric unlikelihood optimization and skeleton-founded tokenization.
TriAlignGR introduces cross-modal alignment, deep interest mining via CoT, and triangular multitask training to fix semantic degradation and opacity in SID-based generative recommendation.
citing papers explorer
-
LLMs Need Encoders for Semantic IDs Too
PrefixMem encoder for Semantic IDs improves deepest-level accuracy by up to 46% relative and full-SID retrieval recall by up to 22% relative on Pinterest data across LLM families.
-
How Reliable Are Semantic-ID Tokenizer Comparisons in Generative Recommendation?
Semantic-ID tokenizers produce collisions affecting up to 30.5% of items across four datasets, inflating Hit@10 by up to 103.36% and making prior tokenizer comparisons unreliable.
-
RecRM-Bench: Benchmarking Multidimensional Reward Modeling for Agentic Recommender Systems
RecRM-Bench is a new large-scale benchmark dataset and framework for multi-dimensional reward modeling in agentic recommender systems, spanning instruction following, factual consistency, query-item relevance, and user behavior prediction.
-
ShopX: A Foundation Model for Intent-to-Item Fulfillment in Agentic Shopping
A single LLM trained to emit semantic item codes can fulfill complex shopping intents with fewer tool hand-offs, improving multi-turn follow-up on Taobao-derived tasks.
-
Intuition-Guided Latent Reasoning for LLM-Based Recommendation
IntuRec anchors LLM latent reasoning for recommendation by deriving an intuition embedding from top-K candidates via self- and cross-attention to initialize more accurate trajectories.
-
Navigating User Behavior toward Personalized Multimodal Generation
NaviGen encodes user behavior via dual collaborative-textual identifiers and applies SFT+RL to produce personalized multimodal outputs and better instructions from interaction history.
-
On the Memorization Behavior of LLMs in Generative Recommendation: Observations, Implications, and Training Strategies
LLMs for generative recommendation show heavy one-hop memorization that accounts for most gains over baselines, and IIRG training that incorporates multi-hop co-occurrences and semantic relations yields larger gains on non-memorizable cases.
-
TubiFM: Unified Item, Carousel, and Search Ranking for Streaming Discovery
A Llama-based model trained on serialized user stories unifies item, carousel, and search ranking and outperforms specialist baselines offline while improving some online metrics and reducing latency.
-
Uni-OPD: Unifying On-Policy Distillation with a Dual-Perspective Recipe
Uni-OPD improves on-policy distillation via student-side data balancing for informative rollouts and teacher-side outcome-guided margin calibration that restores order consistency with rewards.
-
GR2 Technical Report
GR2 applies mid-training on semantic IDs, reasoning distillation, RL with conditional verifiable rewards, and a context compressor to re-ranking in industrial recsys, reporting +18.7% R@1 over baselines.
-
VirtualMLE: A Virtual ML Engineer that Optimizes Sequential Recommenders
VirtualMLE deploys an LLM agent with execution-reflection-memory to tune sequential recommenders, reaching competitive quality on Amazon benchmarks with fewer trials and transferring heuristics across datasets.
-
Echoes in Filter Bubble: Diagnosing and Curing Popularity Bias in Generative Recommenders
Ghost mitigates popularity bias in generative recommenders through asymmetric unlikelihood optimization and skeleton-founded tokenization.
-
TriAlignGR: Triangular Multitask Alignment with Multimodal Deep Interest Mining for Generative Recommendation
TriAlignGR introduces cross-modal alignment, deep interest mining via CoT, and triangular multitask training to fix semantic degradation and opacity in SID-based generative recommendation.