paperlist.ai

AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Can LLMs Truly Forget? Revealing Unlearning Gaps Through Adversarial Evaluation

Ayush Gupta, Hima Varshini Surisetty, Sreevidya Bollineni, Varad Ingale, Tuhina Tripathi, Abhishek Lalwani, Somya Chatterjee, Sadid Hasan

Machine unlearning aims to remove the influence of targeted training data from a model while preserving its remaining capabilities, but evaluating whether such information has truly become inaccessible remains challenging. Existing benchmarks primarily assess unlearning under clean, non-adversarial queries, leaving open whether information that appears forgotten can still be recovered through strategic prompting. We address this gap through a unified evaluation of prompt-based and fine-tuning-based unlearning methods on TOFU using Llama-3.2-3B-Instruct, followed by an adversarial robustness evaluation of methods that perform strongly under standard metrics. We introduce Attack Success Rate (ASR), an LLM-as-judge metric that measures the fraction of adversarial responses whose leakage score exceeds $0.2$, and evaluate recovery across eight attack suites. Our results reveal a substantial gap between clean-query forgetting and adversarial robustness. Although several fine-tuning-based methods achieve Forget Quality above $0.91$, targeted information remains recoverable with ASRs between $72.8\%$ and $84.3\%$, close to the $87.5\%$ ASR of the unprotected base model. In contrast, clean multilingual reformulations yield only $2.95\%$ measured leakage. A manual audit further finds agreement between binary ASR decisions and human factual assessments in seven of ten cases, indicating that ASR provides a useful, though imperfect, signal of behavioral recoverability. These findings show that strong standard-metric performance alone is insufficient to establish robustness after unlearning and motivate adversarial stress-testing as a complementary component of unlearning evaluation.

arXiv

K-Bench: measuring model performance on real scientific agent requests

Aubrey Brueckner, Aubrey M. Brueckner, Darshil Patel, Yuhuan He, Timothy Kassis

Benchmarks for scientific artificial intelligence are mostly written to be scored: multiple-choice questions, curated agent tasks with reference solutions, or simulators with a known generative structure. Real scientific requests arrive differently. They are underspecified, they carry attachments, and they lack ground truth. We report K-Bench 01, an evaluation built from first-turn requests sampled from live user traffic on K-Dense Web and run end to end by nine frontier models in identical sandboxes, yielding 1,602 completed agent runs. Three blinded language-model judges scored every run against an eight-dimension rubric. On a rubric whose 8-anchor is defined as work a domain scientist would accept with minor edits, no model clears the line under all three judges. gpt-5.6-sol has the highest pooled mean, 8.04, but its 95% interval [7.80, 8.23] spans the threshold, and two of the three judges rank claude-opus-5 first instead. We therefore report the ordering of systems as the reproducible quantity, the absolute level as an attribute of the instrument, and the top of the table as unresolved. Across all 39,934 scored judgments -- the eight dimension scores plus a holistic overall for each assessment, excluding not-applicable cells -- 47.6% fall below the 8-point threshold. Difficulty is not uniform across the rubric: scientific accuracy averages 6.22 against 7.33 for communication, on identical denominators and in the same direction within every one of the nine models. The single leading failure tag is overclaiming, on 31.4% of assessments. We argue that the informative quantity for scientific agents is not a leaderboard position but the joint distribution of what was delivered, what was claimed, and what artifacts were produced.

arXiv

Evidence-State Reliability Under Controlled Degradation: Parser-Validity Divergence in a Multi-Stage LLM Pipeline

Naimur Rahman

Multi-stage LLM pipelines can remain structurally valid even when evidence available to downstream stages becomes incomplete, compressed, or conflicting. This paper introduces and operationalizes Evidence-State Reliability (ESR), an evaluation layer concerned with whether intermediate evidence remains sufficiently complete, grounded, internally consistent, and usable for a stage's assigned function. ESR is evaluated separately from parser validity, which measures structural conformance. We evaluate the framework using GLM-5.2 on 60 sanitized base cases under four evidence conditions: clean, compressed-lossy, partial-dropout, and noisy-conflicting. Each condition was processed through decision, audit, and escalation stages. The design comprised 720 planned and ledgered calls, with 713 retained, sanitized execution rows. Across nine matched degraded-minus-clean condition-stage comparisons, all operational stage-success estimates were negative, and all 95% bootstrap intervals remained below zero. All nine parser-validity point estimates were positive, although the three partial-dropout intervals included zero. Among parser-valid degraded audit outputs, degradation detection was 1.0 in each degraded condition, while false-assurance rates remained non-zero; among parser-valid degraded escalation outputs, recovery was 0.0 in every degraded condition. The results show a bounded reliability-layer divergence in the evaluated pipeline: structural conformance can improve directionally while evidence-sensitive stage success deteriorates under the same controlled intervention. They also separate detection of degraded evidence from recovery. The conclusions are limited to the evaluated model configuration, pipeline design, selected sanitized cases, scoring procedure, and single scaled run.

arXiv

Automating Multi-Hop RAG Evaluation via TRIAD: From Context Extraction to Validated Dataset Generation

Lorenz Brehme, Adam Jatowt

Recent advances in LLMs and the adoption of RAG systems in industry have created a need for domain-specific question-answer datasets that can assess RAG performance on proprietary data. Existing datasets, such as HotpotQA, challenge current RAG systems on Wikipedia-based knowledge, but they cannot be transferred directly to domain-specific settings. A comprehensive evaluation of RAG system quality requires both multi-hop queries and unanswerable questions. This paper introduces TRIAD, a three-stage automated dataset generation approach. First, it generates question--answer (QA) pairs for the domain-specific knowledge base of a RAG system. Second, a validator checks each QA-pair in a feedback loop. Third, the QA pairs are extended with relevance-labeled context documents for downstream evaluation. We evaluate this approach against the established MuSiQue and HotpotQA datasets. The results show that the generated dataset exhibits similar performance trends across different RAG setups, while human validation indicates that the questions are suitable for evaluating a domain-specific RAG system. The code used to generate the dataset and all validation results are available in our GitHub repository(https://github.com/lorenzbrehme/triad).

arXiv

Forgotten in Weights, Recovered by Tools: Agentic Tool Unlearning for LLM Agents

Baicheng Chen, Zheyuan Liu, Jingyu Zhang, Kaize Ding, Ningshan Ma, Yue Huang, Meng Jiang

Large language models (LLMs) are increasingly deployed as tool-augmented agents, where responses can depend on tool calls and external observations rather than model parameters alone. This creates an evaluation mismatch for LLM unlearning: previous unlearning methods may suppress direct parametric recall, but an agent can still recover the same forget target through tools such as web search, retrieval, or database lookup. We identify this failure mode as tool-mediated recovery and study agentic tool unlearning, which aims to reduce both parametric recall and tool-mediated recovery while preserving normal tool use for retained knowledge. To address this challenge, we propose Agentic Tool Unlearning (ATU), a two-stage framework. The first stage applies parametric knowledge unlearning to suppress direct recall, while the second stage performs trajectory-level reinforcement learning in simulated tool-augmented environments to penalize target-seeking tool behavior and final-answer leakage. Experiments on RWKU and MUSE across different LLM architectures show that ATU achieves a better balance between target forgetting and retained utility, making unlearning more robust under tool-augmented agent deployment.

arXiv

Beyond Sparse Weights: When Is Attention Compressible?

Chiwun Yang, Xiaoyu Li

KV-cache compression is often justified by attention maps with a few large weights. This is incomplete: large weights may not contain most of the mass, omitted values can cancel, and preserving the attention output may not preserve the task. We separate these questions. Global score gaps -- not threshold counts -- determine how many tokens are needed to retain a target mass. For a realized row, the weighted sum of omitted values is the exact missing statistic. A controlled retrieval--aggregation model explains when truncation helps and when it hurts. These results motivate CertKV, a training-free compressor that reserves one tail-summary slot per head and allocates the rest by value dispersion. Under matched budgets, CertKV is top-two in seven of nine LongBench-v2 settings, remains in the leading compressed tier on 128K RULER, and realizes a ten-fold cache budget in a packed Llama prototype. Compressibility depends on the mass, values, future queries, and task -- not on a sparse-looking map alone.

arXiv

DamageScope: Vision-Language Retrieval at Scale for Disaster Damage Assessment from Satellite Imagery

Ravi K. Rajendran, Biplob Debnath, Murugan Sankaradas, Srimat T. Chakradhar

Timely and accurate assessment of property damage is critical following natural disasters. Traditional on-site inspections are labor-intensive, costly, and often pose safety risks. Advances in satellite imagery and vision-language models (VLMs) enable scalable remote damage assessment; however, integrating VLMs into large-scale Earth observation pipelines presents challenges in computational efficiency, data organization, and information retrieval. To address these challenges, we present DamageScope, a retrieval-augmented framework that combines satellite imagery with Vision-Language Models (VLMs) and Large Language Models (LLMs) to automate property damage analysis. Built on a Retrieval-Augmented Generation (RAG) framework, DamageScope extracts structured visual representations from satellite imagery to support interactive natural language queries for damage assessment. To address scalability, we introduce a novel multi-vector embedding-based clustering algorithm that outperforms traditional single-vector embedding approaches while reducing indexing time by up to 14x. Furthermore, a dual-store data architecture minimizes LLM API calls, reducing both operational cost and response latency by up to approximately 3x. By effectively balancing scalability and operational efficiency, DamageScope provides a robust and practical solution for real-world damage assessment tasks.

arXiv

Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems

Xi Chu, Yupeng Hou

Large language models (LLMs) are becoming a major way for consumers to find products, but we do not yet understand how brands compete in this new channel. We study brand dynamics in LLM recommendations using skincare products -- a category where consumers cannot easily judge quality before buying and must rely on brand reputation -- across three commercial LLMs (GPT-4o-mini, Claude Sonnet, Gemini 3 Flash), with a robustness check on search goods. In three experiments, we find: (1) a Conditional Monopoly where well-known brands get recommended 100% of the time (IAI = 10.0) when all products have the same specifications, but this dominance disappears with less than a +0.1-star rating advantage for a competitor; (2) authority-style marketing language, including fabricated clinical-evidence claims, breaks this monopoly at a Bias Surplus Value equal to +0.17 rating points, with each model responding differently; and (3) a social dilemma in multi-brand GEO competition: when all brands adopt the same optimization strategy, individual payoff falls from +0.802 to +0.007 in our payoff proxy, and non-participating brands receive zero recommendations in our tests. Our results suggest that generative engine optimization (GEO) should be studied not only as a security risk, but also as an emerging marketing practice that shapes market competition.

arXiv

Same Payload, Different Channel: Measuring Trust Asymmetry in Tool-Using Language Models

Mohammed Sameer Syed, Rozhin Yasaei

As language models take on agentic roles that call APIs, read tool outputs, and act on third-party content, their attack surface expands beyond what users type. Whether they treat a malicious instruction the same way regardless of where it arrives has not been studied systematically. We introduce the Safety Asymmetry Score (SAS), measuring how a model's susceptibility to adversarial content shifts depending on whether it arrives in the user message, tool metadata, or tool output, using matched payload pairs that hold the malicious text identical and vary only the channel. Across 10 production LLMs and three attack families, general-purpose models sharply discount instructions arriving as tool metadata relative to identical instructions in the user message, while agent-native models discount them far less. This differential survives an affordance-matched control equalizing tool availability and scoring, and a size-controlled mixed-effects analysis. Within the tool surface the pattern reverses: the same content is far more instruction-like in a tool's description than in its output, with affordances identical. Models treat tool metadata as instructions and tool results as data. In Llama 3.3-70B and GPT-OSS-120B this signal is causally present at mid-to-late depths but non-linearly encoded, so linear probes miss what activation patching recovers.

arXiv

GIM: Evaluating models via tasks that integrate multiple cognitive domains

Rohit Patel, Alexandre Rezende, Steven McClain

As LLM benchmarks saturate, the evaluation community has pursued two strategies to increase difficulty: escalating knowledge demands (GPQA, HLE) or removing knowledge entirely in favor of abstract reasoning (ARC-AGI). The first conflates memorization with capability; the second divorces reasoning from the practical contexts in which it matters. We take a different approach. The Grounded Integration Measure (GIM) is a benchmark of 820 original problems (615 public, 205 private) where difficulty comes from integration; individual problems require coordinating multiple cognitive operations (constraint satisfaction, state tracking, epistemic vigilance, audience calibration) over broadly accessible knowledge, so that reasoning stays grounded in realistic tasks without being gated on specialized expertise. Each problem is an original expert-authored composition, majority with rubric-decomposed scoring. We calibrate a judge-aware continuous response 2-parameter logistic (2PL) IRT model across 53 test-configurations (unique model x thinking-level pairs) and five calibrated judges, using 203,800 epoch-averaged prompt-judge cells derived from >1M raw judge-scored observations, producing robust ability estimates that correctly order test-configurations even when raw accuracy is distorted by errors, missing data, or judge leniency differences. Using this framework, we present a comprehensive leaderboard spanning 22 models and 47 reporting test-configurations, and conduct what is to our knowledge the most extensive published study of how test-time compute trades off against model capability on a fixed benchmark: 11 models swept across 35 test-configurations. We observe that within-family configuration choices, such as thinking budget and quantization, matter as much as model selection, and increasing thinking tokens has diminishing marginal returns.

arXiv

SafeLens: Deliberate and Efficient Video Guardrails with Fast-and-Slow Screening

Shahriar Kabir Nahin, Hadi Askari, Muhao Chen, Anshuman Chhabra

The rapid growth of online video platforms and AI-generated content has made reliable video guardrails a key challenge for safety and real-world deployment. While most videos can be screened through fast pattern recognition, a small subset requires deeper reasoning over temporally complex content and nuanced policy constraints. Existing approaches typically rely on large vision-language models applied uniformly across all inputs, resulting in high inference costs and inefficient allocation of computation. We propose SafeLens, a video guardrail framework that introduces a fast-and-slow inference architecture for efficient and accurate content moderation with variable computational cost across inputs. Additionally, we construct a high-quality dataset by applying influence-guided filtering to the SafeWatch Dataset, retaining only 2.4% of the original data. To further address limitations of training-time scaling, we enable test-time reasoning by augmenting the filtered data with structured Chain-of-Thought traces. Across real-world and AI-generated video benchmarks, SafeLens achieves state-of-the-art performance, outperforming strong open-source video guardrails (e.g., SafeWatch-8B, OmniGuard-7B) and closed-source models (e.g., GPT-5.4, Gemini-3.1-pro) while significantly reducing inference cost, demonstrating that efficient design serves to be more effective than scaling data or model size alone.

arXiv

Revisiting the Effectiveness of LLM Pruning for Test-Time Scaling

Ocean Monjur, Shahriar Kabir Nahin, Anshuman Chhabra

Large Language Models (LLMs) now exhibit remarkable reasoning capabilities through test-time compute scaling (TTS), with impressive performance across math and coding benchmarks. In parallel, research in model compression has developed pruning methods that seek to remove redundant/detrimental parameters without sacrificing task performance. The intersection of these two research advancements lays the foundation for our work. Specific to reasoning LLMs, prior work has shown that structured pruning (methods which remove entire set of layer blocks), significantly degrades TTS reasoning performance. However, in this work, we revisit this assumption and investigate whether unstructured pruning (methods that carefully remove only certain redundant/detrimental weights) exhibits similar limitations. Surprisingly, our extensive experiments across four reasoning benchmarks on two reasoning LLMs: s1.1-7B and Qwen3-8B, consistently show that unstructured pruning augments TTS performance compared to structured pruning, and at times can even outperform the unpruned full-weight LLMs. Furthermore, we also empirically study the impact of different layer-wise sparsity allocation strategies, which are an important parametric choice for instantiating these unstructured methods. These findings challenge the conventional notion that pruning always reduces TTS performance and in fact, suggest that carefully undertaken pruning can retain TTS effectiveness.

arXiv

Alignment midtraining for animals

Jasmine Brazilek, Miles Tidmarsh

We investigate the robustness of value alignment via midtraining with synthetic documents, using animal compassion as a value that is both important in its own right and orthogonal to existing alignment efforts. To evaluate compassionate reasoning, we develop and publicly release Animal Norms In Moral Assessment (ANIMA), a 26-question evaluation spanning 13 ethical dimensions, publicly available as a dataset and Inspect evaluation. On ANIMA, training with 3000 documents achieves 77% compared to 40% for instruction-tuning approaches, with generalization to human compassion and no degradation in standard safety benchmarks or capabilities. However, subsequent unrelated instruction-tuning degrades the intervention, with the advantage disappearing after 5000 samples. Our exploratory results suggest document-based value interventions may require explicit preservation strategies to remain effective through typical training pipelines.

arXiv

Opportunities and Challenges of Natural Language Processing for Low-Resource Senegalese Languages in Social Science Research

Derguene Mbaye, Tatiana D. P. Mbengue, Madoune R. Seye, Moussa Diallo, Mamadou L. Ndiaye, Dimitri S. Adjanohoun, Cheikh S. Wade, Djiby Sow, Jean-Claude B. Munyaka, Jerome Chenal

Natural Language Processing (NLP) is rapidly transforming research methodologies across disciplines, yet African languages remain largely underrepresented in this technological shift. This paper provides the first comprehensive overview of NLP progress and challenges for the six national languages officially recognized by the Senegalese Constitution: Wolof, Pulaar, Sérère, Diola, Mandingue, and Soninké. We synthesize linguistic, socio-technical, and infrastructural factors that shape their digital readiness and identify gaps in data, tools, and benchmarks. Building on existing initiatives and research works, we analyze ongoing efforts in various tasks, covering both text and speech modalities. We also provide a centralized GitHub repository that compiles publicly accessible resources for a range of NLP tasks across these languages, designed to facilitate collaboration and reproducibility. A special focus is devoted to the application of NLP to the social sciences, where multilingual transcription, translation, and retrieval pipelines can significantly enhance the efficiency and inclusiveness of field research. The paper concludes by outlining a roadmap toward sustainable, community-centered NLP ecosystems for Senegalese languages, emphasizing ethical data governance, open resources, and interdisciplinary collaboration.

arXiv

Towards Safer Social Media Platforms: Scalable and Performant Few-Shot Harmful Content Moderation Using Large Language Models

Akash Bonagiri, Lucen Li, Rajvardhan Oak, Zeerak Babar, Magdalena Wojcieszak, Anshuman Chhabra

The prevalence of harmful content on social media platforms poses significant risks to users and society, necessitating more effective and scalable content moderation strategies. Current approaches rely on human moderators, supervised classifiers, and large volumes of training data, and often struggle with scalability, subjectivity, and the dynamic nature of harmful content (e.g., violent content, dangerous challenge trends, etc.). To bridge these gaps, we utilize Large Language Models (LLMs) to undertake few-shot dynamic content moderation via in-context learning. Through extensive experiments on multiple LLMs, we demonstrate that our few-shot approaches can outperform existing proprietary baselines (Perspective and OpenAI Moderation) as well as prior state-of-the-art few-shot learning methods, in identifying harm. We also incorporate visual information (video thumbnails) and assess if different multimodal techniques improve model performance. Our results underscore the significant benefits of employing LLM based methods for scalable and dynamic harmful content moderation online.

arXiv

TurboBias 2.0: Streaming Context-Biasing for Production-Efficient ASR Systems

Vladimir Bataev, Lilit Grigoryan, Andrei Andrusenko, Nikolay Karpov, Vitaly Lavrukhin, Boris Ginsburg

Contextualization is essential for production automatic speech recognition (ASR) systems, where user-provided phrases must be recognized accurately under strict latency constraints. Although many context-biasing methods improve recognition accuracy, they often do not address the practical requirements of modern production ASR systems: streaming inference, efficient batched decoding, user-specific context lists, and low runtime overhead. We propose TurboBias 2.0, a production-oriented framework for efficient phrase boosting in Transducer-based ASR systems. The framework extends GPU-accelerated TurboBias with a case-insensitive boosting graph and per-stream batched decoding, allowing each utterance in a batch to use an independent context-biasing configuration. This enables personalized context biasing for multiple simultaneous users without sharing or mixing their context lists. The proposed framework supports both offline and streaming inference and can be used with greedy and beam-search decoding. Experiments show that TurboBias 2.0 improves contextual phrase recognition while preserving low latency and high throughput.

arXiv

Move by Move: Measuring and Steering How LLMs Conduct Psychotherapy

Afonso Baldo, Hugo Pitorro, Areti Vassilopoulos, Anabela C. Areias, Maya D'Eon, Fabíola Costa, Ricardo Rei, Nuno M. Guerreiro

Users increasingly turn to large language models for emotional support, yet little is known about how these models actually conduct a psychotherapy interaction. We introduce an ontology of ten therapeutic moves: compact, function-based categories grounded in the MULTI-60 inventory, validated through an annotation campaign with five licensed psychologists, and scaled with a judge-based approach that matches expert agreement. Applying it to real counseling transcripts and model-led sessions, we compare the move distributions between human clinicians and a panel of frontier models. Models over-use inquiry at up to three times the human rate, neglect psychoeducation, and are strongly context-anchored: they carry forward strategies initiated by a human clinician but rarely initiate them themselves. Exposing the ontology as a set of tools roughly halves the mean deviation from the human move distribution and improves turn-level alignment with human therapist by 7-9 percentage points, without any fine-tuning.

arXiv

Prompt-Model Interaction Reaches the Fixed Points: A deterministic, task-free structural readout -- and the factorizations of it that failed

Nicolás Vera Zúñiga

That a prompt's effect is not a property of the prompt is established: prompts optimised for one model degrade on another, and rankings reorder under neutral reformatting. That evidence is about task accuracy, which cannot say whether the interaction is a fact about task machinery or about the conditional distribution itself. We ask on a readout with no task in it: the fixed-point structure of the short-window argmax map x_{t+1} = argmax_x p(x | x_{t-1}, x_t), censused from 96 starts. It is deterministic, so nothing can be helped or hurt, and it exists only at short windows -- four of six models lose it entirely by window 16 -- so everything here concerns how a model reads a fragment. Two results. First, the interaction reaches this readout at full magnitude: nine tokens of conditioning move the fixed-point fraction across most of its range, change a four-way structural class, and reorder models, while instruction tuning worth 60.5 IFEval points moves the class by zero. Second, nothing we proposed carries it. Prefix length fails: the effect is not monotone. Four phenomenological factors -- prose-versus-markup, a universal direction, bidirectionality, instruct-resistance -- were each withdrawn within one run of being proposed, dissolved by widening the sample. And the nearest mechanistic account, attention-sink dominance of early tokens, predicts the sign of the shift on 2 of 5 models -- chance -- while a length-by-content cross shows it holds on real text and fails on our probe's uniformly random input, so we are outside its regime, not against it. One fixed nine-token prefix drives four models toward 0 and two toward 1; the bidirectionality survives in-distribution starts. On this readout the unit of explanation is the prompt-model pair. The recurring error it caught in us has a name: a criterion with a shape applied to a quantity with no room to vary.

arXiv

Memory Augmentation Unlocks Efficient Chain-of-Thought Reasoning

Simeng Zhang, Yilong Chen, Wenyuan Zhang, Zhenyu Zhang, Yao Chen, Junyuan Shang, Tingwen Liu

Large language models often rely on Chain-of-Thought (CoT) reasoning to solve complex tasks, but verbose reasoning traces introduce substantial inference overhead. CoT compression shortens generation, yet aggressive compression may disrupt logical coherence and degrade performance. We formalize this trade-off as the Context-Generation Substitution Law, where explicit reasoning context substitutes for part of decode-time generation. Based on this principle, we propose Memory-Augmented Compression, a training-free framework that constructs reusable reasoning memories from historical traces and retrieves them as prefill-side scaffolds. Rather than using raw demonstrations, these memories summarize reusable reasoning patterns, key constraints, and critical operations to compensate for information lost during compression. Experiments show that Memory consistently improves prompt-based Chain-of-Draft (CoD) compression across mathematical reasoning, complex reasoning, and science question answering tasks, yielding accuracy gains of 21.4, 28.0, 29.5, and 6.61 points over CoD on GSM8K, MATH, BBH, and MMLU-Sci, while achieving a 1.14-1.49x latency speedup latency speedup over standard CoT. Memory is also compatible with token-level, reasoning-trace-level, and inference-state compression mechanisms.

arXiv

EnSI-RAG: Entity-Structure-Indexed Retrieval-Augmented Generation for Long-Document Question Answering

Xuanyu Meng, Jiashuo Sun, Jash Rajesh Parekh, Jiawei Han

Question answering (QA) over long, connected documents remains challenging because relevant evidence may span multiple entities and their relationships. Existing retrieval-augmented generation (RAG) methods typically index documents as raw chunks and retrieve them through embedding similarity. Their performance degrades when chunk boundaries separate entities from supporting evidence or when a question requires multi-hop reasoning across the corpus. We propose EnSI-RAG (Entity-Structure-Indexed Retrieval-Augmented Generation), a framework that constructs a query-independent, entity-centered index. Each record (e, t, k, v) represents an entity e, its type t, a semantic category k in {property, relation, aspect}, and a value v, while retaining links to the original source passages. At query time, these records serve as retrieval handles, and an LLM synthesizes the retrieved passages into the final answer. This design separates evidence localization from answer synthesis while preserving traceable source evidence. Across Loong and Oolong, EnSI-RAG achieves an average accuracy of 78.24. Relative to the published baseline scores used as references, this is 6.62 points higher, suggesting its effectiveness across these settings. The code is available at https://github.com/RamonMeng/EnSI-RAG.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.