@_reachsumit Sumit

Sumit posts on X about generative, llm, kuaishou, bytedance the most. They currently have [-----] followers and [---] posts still getting attention that total [-----] engagements in the last [--] hours.

Engagements: [-----] #

• [--] Week [------] +62%
• [--] Month [------] +17%
• [--] Months [-------] +166%
• [--] Year [-------] -54%

Mentions: [--] #

• [--] Month [--] -67%
• [--] Months [---] +62%
• [--] Year [---] +81%

Followers: [-----] #

• [--] Week [-----] +0.74%
• [--] Month [-----] +3%
• [--] Months [-----] +14%
• [--] Year [-----] +29%

CreatorRank: [---------] #

Social Influence

Social category influence technology brands 39% social networks 14% stocks 5% finance 4% currencies 1%

Social topic influence generative #357, llm #363, kuaishou #56, bytedance 9%, tencent #196, alibaba #388, meituan #23, tokenization #373, systems 4%, preference #155

Top accounts mentioned or mentioned by @xiezhouhang @juliana42f9a @liuqidong07 @zombadin @danieltian97 @xincanfeng @qtasfagzcb @liwnhn964157 @baleccchen @ronjunchenfu @jhdylanju @jiayinwang @mvandenhirtz @xylinhaha @xtremesecurity @bclavie

Top assets mentioned Alphabet Inc Class A (GOOGL) Microsoft Corp. (MSFT) Snap, Inc. (SNAP)

Top Social Posts

Top posts by engagements in the last [--] hours

"RAG without Forgetting: Continual Query-Infused Key Memory Introduces a training-free framework that converts transient query-time adaptations into persistent retrieval improvements. 📝 https://arxiv.org/abs/2602.05152 https://arxiv.org/abs/2602.05152"
X Link 2026-02-06T03:43Z [----] followers, [---] engagements

"Internalizing Multi-Agent Reasoning for Accurate and Efficient LLM-based Recommendation Tencent distills multi-agent reasoning into a single model for recommendations using collaborative signal translation to verbalize user behavior patterns. 📝 https://arxiv.org/abs/2602.09829 https://arxiv.org/abs/2602.09829"
X Link 2026-02-11T04:45Z [----] followers, [---] engagements

"DiffuReason: Bridging Latent Reasoning and Generative Refinement for Sequential Recommendation Tencent combines latent reasoning with diffusion-based refinement for sequential recommendation using GRPO alignment to optimize for ranking performance. 📝 https://arxiv.org/abs/2602.09744 https://arxiv.org/abs/2602.09744"
X Link 2026-02-11T04:45Z [----] followers, [---] engagements

"KuaiSearch: A Large-Scale E-Commerce Search Dataset for Recall Ranking and Relevance Kuaishou releases the largest e-commerce search dataset with real user queries covering recall ranking and relevance tasks. 📝 👨🏽💻 https://github.com/benchen4395/KuaiSearch https://arxiv.org/abs/2602.11518 https://github.com/benchen4395/KuaiSearch https://arxiv.org/abs/2602.11518"
X Link 2026-02-13T05:14Z [----] followers, [---] engagements

"MTFM: A Scalable and Alignment-free Foundation Model for Industrial Recommendation in Meituan Meituan presents a transformer-based foundation model using heterogeneous tokenization for multi-scenario recommendation without input alignment. 📝 https://arxiv.org/abs/2602.11235 https://arxiv.org/abs/2602.11235"
X Link 2026-02-13T05:14Z [----] followers, [---] engagements

"OneMall: One Model More Scenarios -- End-to-End Generative Recommender Family at Kuaishou E-Commerce Kuaishou presents a unified generative recommendation framework for e-commerce combining semantic tokenizers Transformer architectures and RL. 📝 https://arxiv.org/abs/2601.21770 https://arxiv.org/abs/2601.21770"
X Link 2026-01-30T05:03Z [----] followers, [---] engagements

"Thinking Broad Acting Fast: Latent Reasoning Distillation from Multi-Perspective Chain-of-Thought for E-Commerce Relevance Alibaba distills multi-perspective CoT reasoning from LLMs into lightweight models for e-commerce relevance classification. 📝 https://arxiv.org/abs/2601.21611 https://arxiv.org/abs/2601.21611"
X Link 2026-01-30T05:05Z [----] followers, [---] engagements

"Zenith: Scaling up Ranking Models for Billion-scale Livestreaming Recommendation ByteDance presents a scalable ranking architecture using Prime Tokens and tokenwise processing modules achieving +9.93% quality watch sessions in production. 📝 https://arxiv.org/abs/2601.21285 https://arxiv.org/abs/2601.21285"
X Link 2026-01-30T05:08Z [----] followers, [---] engagements

"Farewell to Item IDs: Unlocking the Scaling Potential of Large Ranking Models via Semantic Tokens ByteDance presents a token-based ranking framework replacing item IDs with semantic tokens achieving reduction in storage with better scaling properties. https://arxiv.org/abs/2601.22694 https://arxiv.org/abs/2601.22694"
X Link 2026-02-02T04:24Z [----] followers, [----] engagements

"Rethinking Generative Recommender Tokenizer: Recsys-Native Encoding and Semantic Quantization Beyond LLMs Proposes a recommendation-native Semantic ID framework that rethinks representation learning and quantization for genrec https://github.com/FuCongResearchSquad/ReSID https://arxiv.org/abs/2602.02338 https://github.com/FuCongResearchSquad/ReSID https://arxiv.org/abs/2602.02338"
X Link 2026-02-03T05:59Z [----] followers, [---] engagements

"GRAB: An LLM-Inspired Sequence-First Click-Through Rate Prediction Modeling Paradigm Baidu presents an end-to-end generative CTR prediction framework with a novel Causal Action-aware Multi-channel Attention mechanism. 📝 https://arxiv.org/abs/2602.01865 https://arxiv.org/abs/2602.01865"
X Link 2026-02-03T06:01Z [----] followers, [---] engagements

"Beyond RAG for Agent Memory: Retrieval by Decoupling and Aggregation Introduces a hierarchical memory framework that disentangles agent memory into semantic components and retrieves top-down reducing redundancy while improving answer quality. 📝 https://arxiv.org/abs/2602.02007 https://arxiv.org/abs/2602.02007"
X Link 2026-02-03T06:10Z [----] followers, [----] engagements

"Distribution-Aware End-to-End Embedding for Streaming Numerical Features in Click-Through Rate Prediction Tencent presents a framework for numerical feature embedding in streaming CTR prediction using reservoir sampling for distribution estimation. 📝 https://arxiv.org/abs/2602.03223 https://arxiv.org/abs/2602.03223"
X Link 2026-02-04T05:10Z [----] followers, [---] engagements

"Training Multi-Turn Search Agent via Contrastive Dynamic Branch Sampling Introduces a reinforcement learning method that improves credit assignment in multi-turn search agents by sampling contrastive branches near trajectory tails. 📝 https://arxiv.org/abs/2602.03719 https://arxiv.org/abs/2602.03719"
X Link 2026-02-04T05:13Z [----] followers, [---] engagements

"DOS: Dual-Flow Orthogonal Semantic IDs for Recommendation in Meituan Meituan introduces a dual-flow framework that aligns Semantic ID codebook space with generation space using orthogonal residual quantization for generative recommendation. 📝 https://arxiv.org/abs/2602.04460 https://arxiv.org/abs/2602.04460"
X Link 2026-02-05T03:28Z [----] followers, [---] engagements

"AgenticTagger: Structured Item Representation for Recommendation with LLM Agents @XieZhouhang et al. at Google presents a multi-agent framework that generates low-cardinality natural language descriptors for items improving recommendation across tasks https://arxiv.org/abs/2602.05945 https://arxiv.org/abs/2602.05945"
X Link 2026-02-06T03:36Z [----] followers, [---] engagements

"GLASS: A Generative Recommender for Long-sequence Modeling via SID-Tier and Semantic Search Kuaishou integrates long-term user interests into generative recommendation through SID-Tier and semantic search enabling effective modeling of long sequences. https://arxiv.org/abs/2602.05663 https://arxiv.org/abs/2602.05663"
X Link 2026-02-06T03:40Z [----] followers, [---] engagements

"Scaling Laws for Embedding Dimension in Information Retrieval @Julian_a42f9a et al. introduce scaling laws that predict dense retrieval performance based on embedding dimension showing performance follows a power law. 📝 https://arxiv.org/abs/2602.05062 https://arxiv.org/abs/2602.05062"
X Link 2026-02-06T03:41Z [----] followers, [----] engagements

"I published Vol. [---] of "Top Information Retrieval Papers of the Week" on Substack. 🔗 https://recsys.substack.com/p/the-case-for-semantic-tokens-in-modern https://recsys.substack.com/p/the-case-for-semantic-tokens-in-modern"
X Link 2026-02-08T16:57Z [----] followers, [---] engagements

"RLED: Equipping Retrieval and Refinement in Lifelong User Modeling with Semantic IDs for CTR Prediction @LiuQidong07 et al. use semantic IDs to improve retrieval and refinement in lifelong user modeling. 📝 👨🏽💻 https://github.com/abananbao/R2LED https://arxiv.org/abs/2602.06622 https://github.com/abananbao/R2LED https://arxiv.org/abs/2602.06622"
X Link 2026-02-09T04:43Z [----] followers, [---] engagements

"TokenMixer-Large: Scaling Up Large Ranking Models in Industrial Recommenders ByteDance scales ranking models to 7-15 billion parameters through improved residual design sparse-pertoken MoE and efficient CUDA kernels for industrial recommender systems https://arxiv.org/abs/2602.06563 https://arxiv.org/abs/2602.06563"
X Link 2026-02-09T04:43Z [----] followers, [----] engagements

"Multimodal Generative Retrieval Model with Staged Pretraining for Food Delivery on Meituan Meituan presents a staged pretraining strategy that addresses modality neglect in multimodal retrieval by focusing on specialized tasks at each stage 📝 https://arxiv.org/abs/2602.06654 https://arxiv.org/abs/2602.06654"
X Link 2026-02-09T04:44Z [----] followers, [---] engagements

"OneLive: Dynamically Unified Generative Framework for Live-Streaming Recommendation Kuaishou presents a generative recommendation framework for live-streaming that handles dynamic content limited lifecycles and multi-objective optimization. 📝 https://arxiv.org/abs/2602.08612 https://arxiv.org/abs/2602.08612"
X Link 2026-02-10T05:31Z [----] followers, [---] engagements

"RankGR: Rank-Enhanced Generative Retrieval with Listwise Direct Preference Optimization in Recommendation Alibaba introduces a generative retrieval method that captures hierarchical user preferences through listwise direct preference optimization. 📝 https://arxiv.org/abs/2602.08575 https://arxiv.org/abs/2602.08575"
X Link 2026-02-10T05:32Z [----] followers, [---] engagements

"Towards Reliable Social A/B Testing: Spillover-Contained Clustering with Robust Post-Experiment Analysis Kuaishou proposes a framework for A/B testing that mitigates network interference through Balanced Louvain clustering for spillover containment. 📝 https://arxiv.org/abs/2602.08569 https://arxiv.org/abs/2602.08569"
X Link 2026-02-10T05:32Z [----] followers, [---] engagements

"QARM V2: Quantitative Alignment Multi-Modal Recommendation for Reasoning User Sequence Modeling Kuaishou bridges LLM semantic understanding with recommendation systems using reasoning-based item alignment for retrieval and a hybrid quantization method https://arxiv.org/abs/2602.08559 https://arxiv.org/abs/2602.08559"
X Link 2026-02-10T05:34Z [----] followers, [---] engagements

"PIT: A Dynamic Personalized Item Tokenizer for End-to-End Generative Recommendation Kuaishou presents a co-generative framework that dynamically aligns item tokenization with user preferences through minimum-loss selection and a one-to-many beam index. https://arxiv.org/abs/2602.08530 https://arxiv.org/abs/2602.08530"
X Link 2026-02-10T05:35Z [----] followers, [---] engagements

"SimGR: Escaping the Pitfalls of Generative Decoding in LLM-based Recommendation Bypasses token-level generation addressing biases in autoregressive and parallel decoding for LLM-based recommendations. 📝 👨🏽💻 https://anonymous.4open.science/r/SimGR-C408/ https://arxiv.org/abs/2602.07847 https://anonymous.4open.science/r/SimGR-C408/ https://arxiv.org/abs/2602.07847"
X Link 2026-02-10T05:37Z [----] followers, [---] engagements

"MSN: A Memory-based Sparse Activation Scaling Framework for Large-scale Industrial Recommendation ByteDance presents a memory-based sparse scaling framework that retrieves personalized representations from large parameterized memory. https://arxiv.org/abs/2602.07526 https://arxiv.org/abs/2602.07526"
X Link 2026-02-10T05:41Z [----] followers, [---] engagements

"MDL: A Unified Multi-Distribution Learner in Large-scale Industrial Recommendation through Tokenization ByteDance presents a tokenization-based framework that unifies multi-scenario & multi-task learning by treating scenarios and tasks as prompt tokens https://arxiv.org/abs/2602.07520 https://arxiv.org/abs/2602.07520"
X Link 2026-02-10T05:42Z [----] followers, [---] engagements

"High Fidelity Textual User Representation over Heterogeneous Sources via Reinforcement Learning LinkedIn presents an RL framework that synthesizes heterogeneous user data into concise interpretable textual representations for LLM-based recommendations https://arxiv.org/abs/2602.07333 https://arxiv.org/abs/2602.07333"
X Link 2026-02-10T05:45Z [----] followers, [---] engagements

"Semantic Search At LinkedIn LinkedIn presents an LLM-based semantic search framework combining multi-teacher distillation context compression and prefill-optimized inference to achieve 75x throughput gains while powering production search. 📝 https://arxiv.org/abs/2602.07309 https://arxiv.org/abs/2602.07309"
X Link 2026-02-10T05:46Z [----] followers, [----] engagements

"Principled Synthetic Data Enables the First Scaling Laws for LLMs in Recommendation Meta introduces a layered synthetic data framework that enables the first robust power-law scaling laws for LLMs in recommendation. 📝 https://arxiv.org/abs/2602.07298 https://arxiv.org/abs/2602.07298"
X Link 2026-02-10T05:46Z [----] followers, [---] engagements

"Kunlun: Establishing Scaling Laws for Massive-Scale Recommendation Systems through Unified Architecture Design Meta introduces a unified architecture that establishes predictable scaling laws for recommendation systems. 📝 https://arxiv.org/abs/2602.10016 https://arxiv.org/abs/2602.10016"
X Link 2026-02-11T04:42Z [----] followers, [---] engagements

"Efficient Learning of Sparse Representations from Interactions @zombadin et al. propose a strategy for learning high-dimensional sparse embeddings in collaborative filtering achieving up to 100x reduction in embedding size. 📝 👨🏽💻 https://github.com/zombak79/compressed_elsa https://arxiv.org/abs/2602.09935 https://github.com/zombak79/compressed_elsa https://arxiv.org/abs/2602.09935"
X Link 2026-02-11T04:43Z [----] followers, [---] engagements

"Query-Mixed Interest Extraction and Heterogeneous Interaction: A Scalable CTR Model for Industrial Recommender Systems Alibaba presents a scalable ranking model that combines interest extraction with token mixing for efficient feature interaction. 📝 https://arxiv.org/abs/2602.09387 https://arxiv.org/abs/2602.09387"
X Link 2026-02-11T04:49Z [----] followers, [---] engagements

"Predicting Retrieval Utility and Answer Quality in Retrieval-Augmented Generation @DanielTian97 et al. introduce two prediction tasks for RAG: retrieval performance prediction and generation performance prediction. 📝 https://arxiv.org/abs/2601.14546 https://arxiv.org/abs/2601.14546"
X Link 2026-01-22T03:57Z [----] followers, [---] engagements

"Keyword search is all you need: Achieving RAG-Level Performance without vector databases using agentic tool use Shows that tool-augmented LLM agents using simple keyword search can get over 90% of traditional RAG performance. https://github.com/amazon-science/aws-research-science/tree/main/ShellAgent https://www.amazon.science/publications/keyword-search-is-all-you-need-achieving-rag-level-performance-without-vector-databases-using-agentic-tool-use https://github.com/amazon-science/aws-research-science/tree/main/ShellAgent"
X Link 2026-02-06T03:45Z [----] followers, 14.7K engagements

"Generative Reasoning Re-ranker Meta presents a three-stage LLM framework for recommendation re-ranking that combines semantic ID tokenization reasoning trace generation and reinforcement learning with custom rewards. 📝 https://arxiv.org/abs/2602.07774 https://arxiv.org/abs/2602.07774"
X Link 2026-02-10T05:38Z [----] followers, [----] engagements

"SMES: Towards Scalable Multi-Task Recommendation via Expert Sparsity Kuaishou presents a sparse Mixture-of-Experts framework for multi-task recommendation that uses progressive expert routing and cross-task load balancing. 📝 https://arxiv.org/abs/2602.09386 https://arxiv.org/abs/2602.09386"
X Link 2026-02-11T04:50Z [----] followers, [---] engagements

"Beyond the Unit Hypersphere: Embedding Magnitude in Contrastive Learning @xincanfeng et al. challenge the assumption that embedding magnitude is noise in contrastive learning and show document magnitude correlates with relevance and benefits retrieval https://arxiv.org/abs/2602.09229 https://arxiv.org/abs/2602.09229"
X Link 2026-02-11T04:53Z [----] followers, [---] engagements

"Knowledge Integration Decay in Search-Augmented Reasoning of Large Language Models Presents a study showing LLMs fail to integrate retrieved knowledge as reasoning chains grow longer and proposes a training-free method that anchors knowledge. 📝 https://arxiv.org/abs/2602.09517 https://arxiv.org/abs/2602.09517"
X Link 2026-02-11T04:54Z [----] followers, [---] engagements

"Diffusion-Pretrained Dense and Contextual Embeddings Perplexity AI introduces a family of multilingual embedding models using diffusion-pretrained language model backbones with multi-stage contrastive learning. 📝 👨🏽💻 https://huggingface.co/collections/perplexity-ai/pplx-embed https://arxiv.org/abs/2602.11151 https://huggingface.co/collections/perplexity-ai/pplx-embed https://arxiv.org/abs/2602.11151"
X Link 2026-02-12T05:31Z [----] followers, [----] engagements

"Training-Induced Bias Toward LLM-Generated Content in Dense Retrieval Investigates source bias in dense retrievers finding that pro-LLM preference is not inherent but emerges from supervised fine-tuning. 📝 👨🏽💻 https://github.com/williamx854/finetuning-source-bias https://arxiv.org/abs/2602.10833 https://github.com/williamx854/finetuning-source-bias https://arxiv.org/abs/2602.10833"
X Link 2026-02-12T05:32Z [----] followers, [---] engagements

"EST: Towards Efficient Scaling Laws in Click-Through Rate Prediction via Unified Modeling Alibaba proposes a scalable transformer for CTR prediction that achieves fully unified modeling through Lightweight Cross-Attention and Content Sparse Attention. 📝 https://arxiv.org/abs/2602.10811 https://arxiv.org/abs/2602.10811"
X Link 2026-02-12T05:33Z [----] followers, [---] engagements

"S-GRec: Personalized Semantic-Aware Generative Recommendation with Asymmetric Advantage Tencent uses LLMs as offline semantic judges to provide preference supervision for generative recommendation training while keeping online serving lightweight. 📝 https://arxiv.org/abs/2602.10606 https://arxiv.org/abs/2602.10606"
X Link 2026-02-12T05:35Z [----] followers, [---] engagements

"Compute Only Once: UG-Separation for Efficient Large Recommendation Models ByteDance introduces a masking mechanism that separates user and item information flows in dense feature interaction models enabling user-side computation reuse. 📝 https://arxiv.org/abs/2602.10455 https://arxiv.org/abs/2602.10455"
X Link 2026-02-12T05:35Z [----] followers, [---] engagements

"End-to-End Semantic ID Generation for Generative Advertisement Recommendation Tencent proposes a framework that jointly optimizes embeddings and Semantic IDs directly from raw advertising data bypassing the traditional two-stage compression paradigm. 📝 https://arxiv.org/abs/2602.10445 https://arxiv.org/abs/2602.10445"
X Link 2026-02-12T05:36Z [----] followers, [---] engagements

"Spend Search Where It Pays: Value-Guided Structured Sampling and Optimization for Generative Recommendation Tencent uses value-guided decoding and sibling-relative advantage learning to improve generative recommendation. https://arxiv.org/abs/2602.10699 https://arxiv.org/abs/2602.10699"
X Link 2026-02-12T05:37Z [----] followers, [---] engagements

"Self-Evolving Recommendation System: End-To-End Autonomous Model Optimization With LLM Agents Google presents a system where LLM agents autonomously evolve recommenders by generating hypotheses writing code & validating changes through A/B testing. 📝 https://arxiv.org/abs/2602.10226 https://arxiv.org/abs/2602.10226"
X Link 2026-02-12T05:38Z [----] followers, [----] engagements

"IntTravel: A Real-World Dataset and Generative Framework for Integrated Multi-Task Travel Recommendation Alibaba releases a 4.1B interaction dataset and decoder-only generative framework for multi-task travel recommendation. 📝 👨🏽💻 https://github.com/AMAP-ML/IntTravel https://arxiv.org/abs/2602.11664 https://github.com/AMAP-ML/IntTravel https://arxiv.org/abs/2602.11664"
X Link 2026-02-13T05:10Z [----] followers, [---] engagements

"Improving Neural Retrieval with Attribution-Guided Query Rewriting Uses token-level gradient attributions from retrievers to guide LLM-based query rewriting improving retrieval effectiveness without retraining the model. 📝 👨🏽💻 https://github.com/anonym-submission-code/Attribution-Guided-Query-Rewriting-for-Neural-Information-Retrieval https://arxiv.org/abs/2602.11841 https://github.com/anonym-submission-code/Attribution-Guided-Query-Rewriting-for-Neural-Information-Retrieval https://arxiv.org/abs/2602.11841"
X Link 2026-02-13T05:10Z [----] followers, [---] engagements

"Compress Cross and Scale: Multi-Level Compression Cross Networks for Efficient Scaling in Recommender Systems Bilibili introduces a framework for efficient feature interaction in CTR prediction achieving 26x fewer parameters https://github.com/shishishu/MLCC https://arxiv.org/abs/2602.12041 https://github.com/shishishu/MLCC https://arxiv.org/abs/2602.12041"
X Link 2026-02-13T05:12Z [----] followers, [---] engagements

"Recurrent Preference Memory for Efficient Long-Sequence Generative Recommendation Tencent introduces a framework that compresses long user interaction histories into compact Preference Memory tokens for efficient generative recommendation. 📝 https://arxiv.org/abs/2602.11605 https://arxiv.org/abs/2602.11605"
X Link 2026-02-13T05:13Z [----] followers, [---] engagements

"Query-focused and Memory-aware Reranker for Long Context Processing Tencent presents a lightweight listwise reranker using attention scores from retrieval heads. 📝 🤗 https://huggingface.co/MindscapeRAG/QRRanker https://arxiv.org/abs/2602.12192 https://huggingface.co/MindscapeRAG/QRRanker https://arxiv.org/abs/2602.12192"
X Link 2026-02-13T05:17Z [----] followers, [---] engagements

"Detecting Overflow in Compressed Token Representations for Retrieval-Augmented Generation Defines token overflow in soft compression for RAG and proposes lightweight probing classifiers to detect it without LLM inference. 📝 👨🏽💻 https://github.com/s-nlp/overflow-detection https://arxiv.org/abs/2602.12235 https://github.com/s-nlp/overflow-detection https://arxiv.org/abs/2602.12235"
X Link 2026-02-13T05:17Z [----] followers, [---] engagements

"AttentionRetriever: Attention Layers are Secretly Long Document Retrievers Proposes a training-free retrieval model using LLM attention layers and entity-based retrieval for context-aware long document retrieval in RAG. 📝 https://arxiv.org/abs/2602.12278 https://arxiv.org/abs/2602.12278"
X Link 2026-02-13T05:19Z [----] followers, [----] engagements

"PI2I: A Personalized Item-Based Collaborative Filtering Retrieval Framework Alibaba proposes a 2-stage retrieval framework that enhances collaborative filtering with personalized scoring. 📝 🤗 https://huggingface.co/datasets/PI2I/PI2I https://arxiv.org/abs/2601.16815 https://huggingface.co/datasets/PI2I/PI2I https://arxiv.org/abs/2601.16815"
X Link 2026-01-26T03:36Z [----] followers, [---] engagements

"PRISM: Purified Representation and Integrated Semantic Modeling for Generative Sequential Recommendation Addresses codebook collapse and information loss in generative sequential recommendation through collaborative denoising and semantics integration. https://arxiv.org/abs/2601.16556 https://arxiv.org/abs/2601.16556"
X Link 2026-01-26T03:37Z [----] followers, [---] engagements

"FastInsight: Fast and Insightful Retrieval via Fusion Operators for Graph RAG @qtasfagzcb et al. propose a graph retrieval method combining Graph-based Reranker and Semantic-Topological Expansion for fast effective retrieval. https://anonymous.4open.science/r/FastInsight-0F6C https://arxiv.org/abs/2601.18579 https://anonymous.4open.science/r/FastInsight-0F6C https://arxiv.org/abs/2601.18579"
X Link 2026-01-27T06:51Z [----] followers, [---] engagements

"ReflectiveRAG: Rethinking Adaptivity in Retrieval-Augmented Generation Amazon presents a training-free RAG framework using self-reflective query refinement and contrastive noise removal to improve factual grounding with only 18ms added latency. 📝 https://www.amazon.science/publications/reflectiverag-rethinking-adaptivity-in-retrieval-augmented-generation https://www.amazon.science/publications/reflectiverag-rethinking-adaptivity-in-retrieval-augmented-generation"
X Link 2026-01-27T07:11Z [----] followers, [---] engagements

"Iterative Reranking as a Compute-Scaling Method for LLM-based Rankers Proposes iteratively applying listwise LLM rankers to refine search results trading compute for quality gains on difficult queries like comparative and multi-attribute searches. https://www.amazon.science/publications/iterative-reranking-as-a-compute-scaling-method-for-llm-based-rankers https://www.amazon.science/publications/iterative-reranking-as-a-compute-scaling-method-for-llm-based-rankers"
X Link 2026-01-27T07:14Z [----] followers, [---] engagements

"PROMISE: Process Reward Models Unlock Test-Time Scaling Laws in Generative Recommendations Kuaishou introduces a process reward model framework for generative recommendation that mitigates semantic drift through step-by-step verification. 📝 https://arxiv.org/abs/2601.04674 https://arxiv.org/abs/2601.04674"
X Link 2026-01-09T04:05Z [----] followers, [---] engagements

"Deep GraphRAG: A Balanced Approach to Hierarchical Retrieval and Adaptive Integration Ant Group introduces a hierarchical retrieval framework using beam search and an RL method (DW-GRPO) that enables compact 1.5B models to approach 70B performance. 📝 https://arxiv.org/abs/2601.11144 https://arxiv.org/abs/2601.11144"
X Link 2026-01-19T06:42Z [----] followers, 10.8K engagements

"HyFormer: Revisiting the Roles of Sequence Modeling and Feature Interaction in CTR Prediction ByteDance presents a unified hybrid transformer that integrates long-sequence modeling and feature interaction through Query Decoding and Query Boosting. 📝 https://arxiv.org/abs/2601.12681 https://arxiv.org/abs/2601.12681"
X Link 2026-01-21T06:35Z [----] followers, [---] engagements

"Agentic-R: Learning to Retrieve for Agentic Search @liwnhn964157 et al. at Baidu present a retriever training framework for agentic search that uses both local relevance and global answer correctness to measure passage utility https://github.com/8421BCD/Agentic-R https://arxiv.org/abs/2601.11888 https://github.com/8421BCD/Agentic-R https://arxiv.org/abs/2601.11888"
X Link 2026-01-21T06:38Z [----] followers, [---] engagements

"RAGExplorer: A Visual Analytics System for the Comparative Diagnosis of RAG Systems Introduces a visual analytics system for comparing and diagnosing multiple RAG configurations. 📝 👨🏽💻 https://github.com/Thymezzz/RAGExplorer https://arxiv.org/abs/2601.12991 https://github.com/Thymezzz/RAGExplorer https://arxiv.org/abs/2601.12991"
X Link 2026-01-21T06:42Z [----] followers, [---] engagements

"Beyond Single-shot Writing: Deep Research Agents are Unreliable at Multi-turn Report Revision @BaleCcchen et al. reveal that Deep Research Agents struggle with iterative report revision. 📝 👨🏽💻 https://github.com/BaleChen/Mr-Dre https://arxiv.org/abs/2601.13217 https://github.com/BaleChen/Mr-Dre https://arxiv.org/abs/2601.13217"
X Link 2026-01-21T06:44Z [----] followers, [---] engagements

"CoNRec: Context-Discerning Negative Recommendation with LLMs Alibaba presents an LLM framework for modeling users' negative interests using hierarchical semantic IDs and progressive GRPO training to accurately predict items users are likely to dislike https://arxiv.org/abs/2601.15721 https://arxiv.org/abs/2601.15721"
X Link 2026-01-23T03:31Z [----] followers, [---] engagements

"Graph-Anchored Knowledge Indexing for Retrieval-Augmented Generation Presents an evolving graph structure that incrementally anchors key entities and relations to better integrate scattered evidence for multi-hop QA. 📝 👨🏽💻 https://github.com/NEUIR/GraphAnchor https://arxiv.org/abs/2601.16462 https://github.com/NEUIR/GraphAnchor https://arxiv.org/abs/2601.16462"
X Link 2026-01-26T03:45Z [----] followers, [----] engagements

"Token-level Collaborative Alignment for LLM-based Generative Recommendation Introduces a plug-and-play framework that bridges collaborative filtering signals with LLM generation through token-level soft label alignment. 📝 👨🏽💻 https://github.com/critical88/TCA4Rec https://arxiv.org/abs/2601.18457 https://github.com/critical88/TCA4Rec https://arxiv.org/abs/2601.18457"
X Link 2026-01-27T06:53Z [----] followers, [---] engagements

"Think When Needed: Model-Aware Reasoning Routing for LLM-based Ranking Presents a lightweight router that decides when LLMs should use reasoning for ranking tasks achieving better accuracy with fewer tokens. 📝 👨🏽💻 https://anonymous.4open.science/r/reasoning_router-561 https://arxiv.org/abs/2601.18146 https://anonymous.4open.science/r/reasoning_router-561 https://arxiv.org/abs/2601.18146"
X Link 2026-01-27T06:59Z [----] followers, [---] engagements

"Differentiable Semantic ID for Generative Recommendation @ron_junchen_fu et al. enable joint optimization of semantic IDs and generative recommenders through Gumbel noise-based exploration and uncertainty decay strategies to prevent codebook collapse. 📝 https://arxiv.org/abs/2601.19711 https://arxiv.org/abs/2601.19711"
X Link 2026-01-28T06:24Z [----] followers, [---] engagements

"Length-Adaptive Interest Network for Balancing Long and Short Sequence Modeling in CTR Prediction Presents a plug-and-play framework that uses sequence length as a conditioning signal to address perf imbalance between long- and short-sequence users. 📝 https://arxiv.org/abs/2601.19142 https://arxiv.org/abs/2601.19142"
X Link 2026-01-28T06:32Z [----] followers, [---] engagements

"When Vision Meets Texts in Listwise Reranking Introduces a lightweight 2B multimodal reranker that performs listwise ranking on mixed image-text documents using progressive cross-modal training and diversity-based data curation. 📝 https://arxiv.org/abs/2601.20623 https://arxiv.org/abs/2601.20623"
X Link 2026-01-29T06:13Z [----] followers, [---] engagements

"LANCER: LLM Reranking for Nugget Coverage @JHDylanJu et al. introduce an LLM-based reranking method that generates sub-questions and predicts document answerability to optimize information coverage for long-form RAG tasks. 📝 👨🏽💻 https://github.com/DylanJoo/LANCER https://arxiv.org/abs/2601.22008 https://github.com/DylanJoo/LANCER https://arxiv.org/abs/2601.22008"
X Link 2026-01-30T04:55Z [----] followers, [---] engagements

"LEMUR: Learned Multi-Vector Retrieval Reduces multi-vector similarity search to single-vector search via supervised learning achieving an order of magnitude speedup over existing ColBERT-style retrieval methods. 📝 👨🏽💻 https://github.com/ejaasaari/lemur https://arxiv.org/abs/2601.21853 https://github.com/ejaasaari/lemur https://arxiv.org/abs/2601.21853"
X Link 2026-01-30T05:02Z [----] followers, [---] engagements

"UniDex: Rethinking Search Inverted Indexing with Unified Semantic Modeling Kuaishou presents a model-based framework that replaces traditional term-matching with semantic modeling for inverted indexing improving retrieval effectiveness in search. https://arxiv.org/abs/2509.24632 https://arxiv.org/abs/2509.24632"
X Link 2025-09-30T05:10Z [----] followers, [---] engagements

"SPAD: Seven-Source Token Probability Attribution with Syntactic Aggregation for Detecting Hallucinations in RAG Introduces a method to detect hallucinations in RAG systems by mathematically attributing token probabilities to seven distinct sources. 📝 https://arxiv.org/abs/2512.07515 https://arxiv.org/abs/2512.07515"
X Link 2025-12-09T06:08Z [----] followers, [---] engagements

"Unleashing the Native Recommendation Potential: LLM-Based Generative Recommendation via Structured Term Identifiers Kuaishou uses structured textual keywords as item identifiers to enable generative recommendation. https://github.com/ZY0025/GRLM https://arxiv.org/abs/2601.06798 https://github.com/ZY0025/GRLM https://arxiv.org/abs/2601.06798"
X Link 2026-01-13T05:26Z [----] followers, [---] engagements

"SGR: Stepwise Semantic-Guided Reasoning in Latent Space for Generative Recommendation Kuaishou presents a generative recommendation framework with stepwise thinking tokens that align with hierarchical semantic IDs enabling reliable reasoning paths. 📝 https://arxiv.org/abs/2601.18664 https://arxiv.org/abs/2601.18664"
X Link 2026-01-27T06:50Z [----] followers, [---] engagements

"Orchestrating Specialized Agents for Trustworthy Enterprise RAG Atlassian introduces an agentic framework that replaces linear RAG with memory-locked synthesis and evidence-coverage-guided execution for traceable enterprise reporting. 📝 https://arxiv.org/abs/2601.18267 https://arxiv.org/abs/2601.18267"
X Link 2026-01-27T06:55Z [----] followers, [---] engagements

"Unleashing the Potential of Sparse Attention on Long-term Behaviors for CTR Prediction Meituan presents a sparse attention model that captures long-term user behaviors through personalized chunking and temporal encoding. https://github.com/laiweijiang/SparseCTR https://arxiv.org/abs/2601.17836 https://github.com/laiweijiang/SparseCTR https://arxiv.org/abs/2601.17836"
X Link 2026-01-27T07:01Z [----] followers, [---] engagements

"Masked Diffusion Generative Recommendation Alibaba presents a framework that reformulates generative recommendation as masked diffusion enabling parallel decoding with curriculum noise scheduling. 📝 https://arxiv.org/abs/2601.19501 https://arxiv.org/abs/2601.19501"
X Link 2026-01-28T06:25Z [----] followers, [---] engagements

"MERGE: Next-Generation Item Indexing Paradigm for Large-Scale Streaming Recommendation ByteDance proposes a dynamic item indexing method that adaptively constructs clusters from scratch and monitors occupancy in real-time. 📝 https://arxiv.org/abs/2601.20199 https://arxiv.org/abs/2601.20199"
X Link 2026-01-29T06:16Z [----] followers, [---] engagements

"Attribution Techniques for Mitigating Hallucinated Information in RAG Systems: A Survey Surveys attribution-based techniques to reduce hallucinations in RAG systems proposing a unified four-module pipeline and taxonomy. 📝 https://arxiv.org/abs/2601.19927 https://arxiv.org/abs/2601.19927"
X Link 2026-01-29T06:20Z [----] followers, [---] engagements

"Retrieval-Infused Reasoning Sandbox: A Benchmark for Decoupling Retrieval and Reasoning Capabilities ByteDance introduces a benchmark that isolates retrieval from reasoning via four evaluation regimes. 📝 👨🏽💻 https://retrieval-infused-reasoning-sandbox.github.io/ https://arxiv.org/abs/2601.21937 https://retrieval-infused-reasoning-sandbox.github.io/ https://arxiv.org/abs/2601.21937"
X Link 2026-01-30T04:57Z [----] followers, [---] engagements

"I published Vol. [---] of "Top Information Retrieval Papers of the Week" on Substack. 🔗 https://recsys.substack.com/p/empirical-patterns-in-real-world https://recsys.substack.com/p/empirical-patterns-in-real-world"
X Link 2026-02-01T16:32Z [----] followers, [----] engagements

"BEAR: Towards Beam-Search-Aware Optimization for Recommendation with Large Language Models Introduces a beam-search-aware fine-tuning objective for LLM-based recommender systems that addresses the training-inference inconsistency. 📝 https://arxiv.org/abs/2601.22925 https://arxiv.org/abs/2601.22925"
X Link 2026-02-02T04:24Z [----] followers, [---] engagements

"Understanding Internal Representations of Recommendation Models with Sparse Autoencoders @Jiayin_Wang_ et al. propose a framework using sparse autoencoders to extract interpretable latent concepts from recommendation models. 📝 👨🏽💻 https://github.com/Alice1998/RecSAE https://dl.acm.org/doi/10.1145/3795529 https://github.com/Alice1998/RecSAE https://dl.acm.org/doi/10.1145/3795529"
X Link 2026-02-02T04:27Z [----] followers, [---] engagements

"ChunkNorris: A High-Performance and Low-Energy Approach to PDF Parsing and Chunking Presents a heuristic-based PDF parsing and chunking method that outperforms ML-based approaches in retrieval accuracy. 📝 👨🏽💻 https://github.com/wikit-ai/chunknorris https://arxiv.org/abs/2602.00010 https://github.com/wikit-ai/chunknorris https://arxiv.org/abs/2602.00010"
X Link 2026-02-03T06:07Z [----] followers, [---] engagements

"AdNanny: One Reasoning LLM for All Offline Ads Recommendation Tasks Microsoft introduces a unified 671B-parameter reasoning LLM fine-tuned from DeepSeek-R1 that consolidates multiple offline ads tasks into a single model. 📝 https://arxiv.org/abs/2602.01563 https://arxiv.org/abs/2602.01563"
X Link 2026-02-03T06:07Z [----] followers, [---] engagements

"Multimodal Generative Recommendation for Fusing Semantic and Collaborative Signals @MVandenhirtz et al. introduces a multimodal generative recommender that fuses text image and collaborative signals with a novel DINO-based image quantization method. https://arxiv.org/abs/2602.03713 https://arxiv.org/abs/2602.03713"
X Link 2026-02-04T04:54Z [----] followers, [---] engagements

"Bringing Reasoning to Generative Recommendation Through the Lens of Cascaded Ranking @xylinhaha et al. introduce a cascaded reasoning framework that mitigates bias amplification in generative recommendation. 📝 👨🏽💻 https://github.com/Linxyhaha/CARE https://arxiv.org/abs/2602.03692 https://github.com/Linxyhaha/CARE https://arxiv.org/abs/2602.03692"
X Link 2026-02-04T04:59Z [----] followers, [---] engagements

"A-RAG: Scaling Agentic Retrieval-Augmented Generation via Hierarchical Retrieval Interfaces Introduces an agentic RAG framework with hierarchical retrieval tools enabling models to autonomously decide retrieval strategies. 📝 👨🏽💻 https://github.com/Ayanami0730/arag https://arxiv.org/abs/2602.03442 https://github.com/Ayanami0730/arag https://arxiv.org/abs/2602.03442"
X Link 2026-02-04T05:06Z [----] followers, [---] engagements

"PolyG: Effective and Efficient GraphRAG with Adaptive Graph Traversal Proposes a classification system for graph queries that adaptively selects the most suitable graph traversal strategy achieving 75% win rate on answer quality and up to [--] speedup. https://arxiv.org/abs/2504.02112 https://arxiv.org/abs/2504.02112"
X Link 2025-04-04T05:53Z [----] followers, [---] engagements

"QZhou-Embedding Technical Report Kingsoft AI presents a unified multi-task framework with LLM-powered data synthesis achieving sota results on MTEB and CMTEB benchmarks through specialized training strategies. https://github.com/Kingsoft-LLM/QZhou-Embedding https://arxiv.org/abs/2508.21632 https://github.com/Kingsoft-LLM/QZhou-Embedding https://arxiv.org/abs/2508.21632"
X Link 2025-09-01T06:35Z [----] followers, [----] engagements

"Hierarchical Token Prepending: Enhancing Information Flow in Decoder-based LLM Embeddings Snap Inc improves decoder-based LLM embeddings by creating block-level summary tokens that enable backward information flow. 📝 👨🏽💻 https://github.com/snap-research/HTP https://arxiv.org/abs/2511.14868 https://github.com/snap-research/HTP https://arxiv.org/abs/2511.14868"
X Link 2025-11-20T05:35Z [----] followers, [---] engagements

"Is Cosine-Similarity of Embeddings Really About Similarity Netflix cautions against blindly using cosine similarity as a measure of semantic similarity between learned embeddings as it can yield arbitrary and meaningless results. https://arxiv.org/abs/2403.05440 https://arxiv.org/abs/2403.05440"
X Link 2024-03-11T04:32Z [----] followers, 377.1K engagements

"In the final post of the Adaptive RAG series we explore how to treat selective retrieval as a core learned skill moving from passive observation to active intelligent decision-making. https://blog.reachsumit.com/posts/2025/10/learning-to-retrieve/ https://blog.reachsumit.com/posts/2025/10/learning-to-retrieve/"
X Link 2025-10-08T06:49Z [----] followers, [----] engagements

Limited data mode. Full metrics available with subscription: lunarcrush.com/pricing