archive
Every paper Pith has read. Search by title, abstract, or pith.
1286 papers in cs.IR · page 15
-
Semantic trimming speeds generative rec training by 23-38%
Semantic Trimming and Auxiliary Multi-step Prediction for Generative Recommendation
-
Tree-based next-scale generator reranks lists scale by scale
Next-Scale Generative Reranking: A Tree-based Generative Rerank Method at Meituan
-
Hidden sequence split inflates recent recommendation scores
Pay Attention to Sequence Split: Uncovering the Impacts of Sub-Sequence Splitting on Sequential Recommendation Models
-
Immersive recommenders repeat visible info and miss next questions
Evaluating Scene-based In-Situ Item Labeling for Immersive Conversational Recommendation
-
MaxSim pooling spikes gradients and hurts long-document retrieval
Spike Hijacking in Late-Interaction Retrieval
-
DotRAG makes retrieval a reasoning process over paths
DOTRAG: Retrieval-Time Reasoning Along Paths
-
Entity signals add no MAP value in open-world retrieval
Entities as Retrieval Signals: A Systematic Study of Coverage, Supervision, and Evaluation in Entity-Oriented Ranking
-
RAG-enhanced MedGemma scores 89% on EHR trial screening
Retrieval-Augmented LLMs for Evidence Localization in Clinical Trial Recruitment from Longitudinal EHR Narratives
-
RL policies cut retrieval steps by 47% at 92% accuracy
Offline RL for Adaptive Policy Retrieval in Prior Authorization
-
Codebook rebalancing cuts popularity bias in generative recommenders
CRAB: Codebook Rebalancing for Bias Mitigation in Generative Recommendation
-
RL optimizes documents to boost black-box retriever accuracy
Document Optimization for Black-Box Retrieval via Reinforcement Learning
-
Skill library from strong agents lifts weaker ones
SkillX: Automatically Constructing Skill Knowledge Bases for Agents
-
DisastRAG lifts disaster query accuracy 12-23 points
DisastRAG: A Multi-Source Disaster Information Integration and Access System Based on Retrieval-Augmented Large Language Models
-
RAG system gains 12-23 points on disaster info tasks
DisastRAG: A Multi-Source Disaster Information Integration and Access System Based on Retrieval-Augmented Large Language Models
-
Stratified sampling beats hard negatives in retrieval distillation
Beyond Hard Negatives: The Importance of Score Distribution in Knowledge Distillation for Dense Retrieval
-
Ruling out mimics boosts medical QA retrieval
Ruling Out to Rule In: Contrastive Hypothesis Retrieval for Medical Question Answering
-
Contrastive learning disentangles long- and short-term interests
SLSREC: Self-Supervised Contrastive Learning for Adaptive Fusion of Long- and Short-Term User Interests
-
Human-inspired forgetting lifts AI agent memory to 70% on benchmarks
SuperLocalMemory V3.3: The Living Brain -- Biologically-Inspired Forgetting, Cognitive Quantization, and Multi-Channel Retrieval for Zero-LLM Agent Memory Systems
-
RL aligns retrieval and generation stages in recommender systems
Retrieval Augmented Conversational Recommendation with Reinforcement Learning
-
One-step flow model delivers 10x faster sequential recommendations
FAVE: Flow-based Average Velocity Establishment for Sequential Recommendation
-
Small models structure kidney biopsy reports on ordinary computers
A Semi-Automated Annotation Workflow for Paediatric Histopathology Reports Using Small Language Models
-
Formalized topics raise LLM judgment agreement
Formalized Information Needs Improve Large-Language-Model Relevance Judgments
-
Single network matches ensemble performance in sequential recs
FLAME: Condensing Ensemble Diversity into a Single Network for Efficient Sequential Recommendation
-
Dual-hypergraph RAG raises math teacher feedback quality by 15%
MisEdu-RAG: A Misconception-Aware Dual-Hypergraph RAG for Novice Math Teachers
-
Circuit edits enable effective unlearning in LLM recommenders
CURE:Circuit-Aware Unlearning for LLM-based Recommendation
-
Two datasets give generative recommenders 10 million real ad users
Tencent Advertising Algorithm Challenge 2025: All-Modality Generative Recommendation
-
Popularity outranks AI models in item-level explanation ranking
Rank, Don't Generate: Statement-level Ranking for Explainable Recommendation
-
LLM embeddings fused and aligned boost tail-item recommendations
Fusion and Alignment Enhancement with Large Language Models for Tail-item Sequential Recommendation
-
LightThinker++ cuts LLM peak tokens by 70% while raising accuracy
LightThinker++: From Reasoning Compression to Memory Management
-
Specialized retrievers match top effectiveness with competitive throughput
Are LLM-Based Retrievers Worth Their Cost? An Empirical Study of Efficiency, Robustness, and Reasoning Overhead
-
Co-training aligns process rewards in search agents
OASES: Outcome-Aligned Search-Evaluation Co-Training for Agentic Search
-
RL with simulator guidance reduces bias in multi-turn recommendations
User Simulator-Guided Multi-Turn Preference Optimization for Reasoning LLM-based Conversational Recommendation
-
Multimodal paths boost LLM explainable recommendations
MMP-Refer: Multimodal Path Retrieval-augmented LLMs For Explainable Recommendation
-
Conditional diffusion denoises multimodal recs and augments pairs
Joint Behavior-guided and Modality-coherence Conditional Graph Diffusion Denoising for Multi Modal Recommendation
-
MG²-RAG speeds multimodal graph construction 43 times with SOTA accuracy
MG$^2$-RAG: Multi-Granularity Graph for Multimodal Retrieval-Augmented Generation
-
TRACE-KG induces reusable schemas while building traceable graphs
Beyond Predefined Schemas: TRACE-KG for Context-Enriched Knowledge Graphs from Complex Documents
-
TF-IDF SVM routes RAG queries at 93% accuracy
Lightweight Query Routing for Adaptive RAG: A Baseline Study on RAGRouter-Bench
-
Two-stage adapter aligns query and document spaces for low-label retrieval
Align then Train: Efficient Retrieval Adapter Learning
-
Bridge conditioning lifts multi-hop retrieval without training
BridgeRAG: Training-Free Bridge-Conditioned Retrieval for Multi-Hop Question Answering
-
Sparse LLM labels train sharper topic clusters than big models
PRISM: LLM-Guided Semantic Clustering for High-Precision Topics
-
User-specific diffusion filtering boosts multi-modal recommendation
User-Aware Conditional Generative Total Correlation Learning for Multi-Modal Recommendation
-
Multi-agent systems self-optimize to match expert prompts
Self-Optimizing Multi-Agent Systems for Deep Research
-
Prompt compression cuts LLM inference time up to 18% when conditions match
Prompt Compression in the Wild: Measuring Latency, Rate Adherence, and Quality for Faster LLM Inference
-
BIPCL lifts sequential recs via bilateral intent prototypes and perturbations
BIPCL: Bilateral Intent-Enhanced Sequential Recommendation via Embedding Perturbation Contrastive Learning
-
AI now builds video trailers instead of selecting clips
Generative AI for Video Trailer Synthesis: From Extractive Heuristics to Autoregressive Creativity
-
Auto-generated annotations replace embeddings for cheaper document search
AnnoRetrieve: Efficient Structured Retrieval for Unstructured Document Analysis
-
MBGR fixes seesaw effect in multi-business generative recs
MBGR: Multi-Business Prediction for Generative Recommendation at Meituan
-
Agentic affordance reasoning improves context-sensitive POI ranking
Agent4POI: Agentic Context-Conditioned Affordance Reasoning for Multimodal Point-of-Interest Recommendation
-
LLM agents verify technical claims without experts
AutoVerifier: An Agentic Automated Verification Framework Using Large Language Models
-
Quadratic program retrieves diverse RAG passages with speedup
Principled and Scalable Diversity-Aware Retrieval via Cardinality-Constrained Binary Quadratic Programming