arXiv Science⌕ Search

arXiv subjects

Emine Yilmaz

Publications and source records attributed to Emine Yilmaz.

At least 19 recordsLinked to original sources
arXiv2026

PMRecBench: A Benchmark for Personalization in Recommendation Agents under Multimodal Queries

On Web platforms for shopping and entertainment, recommendation agents under multimodal queries aim to deliver personalized recommendations that satisfy users' visually specified requirements and reflect preferences inferred from user histories. However, existing personalized recommendation benchmarks use text-only queries, while multimodal ones assess query matching without evaluating personalization based on real user histories. We introduce the Personalized Multimodal Recommendation Benchmark (PMRecBench), the first benchmark to evaluate the personalization of recommendation agents under multimodal queries. It pairs 510 quality-controlled image-text queries across clothing and movies with the corresponding users' real histories. PMRecBench features two key designs: (1) queries derived from consistent behavioral patterns in real user records, with reference images and targets linked to actual interactions; and (2) challenging diagnostic pools with candidate items rated highly by the user but violating requirements, and items satisfying requirements but never rated highly, testing requirement understanding and preference following, respectively. Extensive experiments on diagnostic pools reveal a personalization imbalance in recommendation agents: relative to personalization signals alone, adding multimodal queries weakens preference following, even though the additional information about query requirements improves overall ranking performance. Further analysis links this decline in preference following to image interpretation, where agents can prioritize resemblance to reference images over evidence from user histories, and stronger commercial backbones alone fail to resolve this imbalance. This personalization imbalance highlights a central challenge for recommendation agents: improving ranking performance while preserving preference following when responding to multimodal user queries

cs.IR↗
arXiv2026

What Improves During Pretraining? Tracking Knowledge Acquisition with Point-in-Time Models

Language models pretrained on web corpora collected over multiple years learn both general and domain-specific knowledge. How these different kinds of knowledge are learned as pretraining progresses remains unclear. Recent point-in-time models trained on web snapshots from different years provide a way to probe how knowledge prevalence varies over time. Using these models, we propose a temporal variability score (TVS), defined as the coefficient of variation of an item's negative log-likelihood across models. A higher score suggests that the knowledge may be less consistently represented across years of web text. We evaluate six benchmarks covering diverse topics using six models from multiple open-source model families. We find that low- and high-TVS items are learned similarly early in training, but high-TVS items improve faster later on. At the end of the pretraining run, high-TVS items outperform low-TVS items in 34 of 36 model-benchmark settings, with an average advantage of 7.3 points of chance-normalized accuracy. This pattern is most robust on knowledge-intensive tasks, even when comparing items of similar difficulty. Our findings reveal differences in how knowledge is learned over the course of pretraining, with temporal variability providing a diagnostic of these learning dynamics beyond aggregate benchmark scores.

cs.AI↗
arXiv2026

$Δτ$-Bench: Pushing Agents Toward a Moving Capability

In this paper, we introduce $Δτ$-Bench, an executable agent benchmark that models the structural evolution of large tool ecosystems. In $Δτ$-Bench, independently evolving services continually reshape how tasks can be accomplished: familiar tools may disappear or be replaced, while more efficient alternatives emerge amid overlapping functionality and noisy documentation. Against this moving capability frontier, we evaluate whether agents can proactively discover efficient tool-use paths and promptly revise their accumulated procedural knowledge, bringing together demands that existing benchmarks fail to examine. $Δτ$-Bench comprises 1,588 validated multi-turn tasks across 42 domains and six successive ecosystem states generated through capability-preserving transformations. Our evaluation shows that even the strongest evaluated model or harness achieves only 42.3% task success, while initial gains from existing skill-learning methods can not consistently translate into better adaptation as the ecosystem evolves. These findings motivate Steward, an ecosystem manager that jointly manages tools and procedural knowledge through incremental skill maintenance and dynamic provisioning during task execution. Experiments show that Steward sustains higher task success across ecosystem changes with lower skill-maintenance overhead, while improving both recovery from disrupted workflows and adoption of more efficient alternatives.

cs.CL↗
arXiv2026

A Russian, a Brazilian, and a German Walk into a Bar, Ask the Same Question-- And Get Three Different Answers: Profiling LLM Perspective Diversity on Contested Topics

Large Language Models(LLMs) have shown strong performance on tasks with verifiable answers, but many real-world queries involve subjectivity, disagreement, and conflicting evidence. Most existing benchmarks for such queries remain English-centric, overlooking the vast diversity of perspectives held by users worldwide. To systematically study how LLM perspectives vary across languages, we curate PerspectivesQA, a manually verified multilingual dataset of globally contested topics covering 18 languages, and propose ROOTS, a framework that profiles linguistic, perspective, and framing diversity. We evaluate a diverse set of eight models covering both open-source and proprietary families at varying scales, establishing that LLMs hedge their answers in English but commit to positions in lower-resource languages. Larger models maintain a more consistent cross-lingual worldview, while smaller models are more easily swayed. All models also exhibit high sensitivity to minor surface-level input variations, producing notable stance even in larger models.

cs.CL↗
arXiv2026

RouterWorld: Scaling Conversational AI with Latency-Aware LLM Routing

Conversational AI systems built on Large Language Models (LLMs) require routing mechanisms that balance answer quality and response latency at scale. However, existing LLM routing benchmarks assume full observability of all candidate model outcomes, under-emphasize latency as a first-class metric, and fragment evaluation across heterogeneous protocols misaligned with real-world settings. We introduce RouterWorld, a comprehensive benchmark for LLM routing in conversational GenAI. RouterWorld unifies 371k single-turn responses and 76k multi-turn agentic trajectories, adopts natural user interactions with a standardized LLM-as-a-judge protocol, and decouples efficiency from transient pricing via hardware-grounded latency annotations. Crucially, RouterWorld formalizes routing under partial observability through two setups: Cold-Start, which restricts learning to sparse, static logs, and Voyager, which models continuous data streams where routing decisions shape future training distributions. Evaluating 13 routing baselines, we find that most routers over-emphasize accuracy while ignoring latency under partial observability. Code and benchmark data are publicly available.

cs.AI↗
arXiv2026

StepBench: Diagnosing Real-World Procedural Understanding in AI Agents

Evaluating an agent's planning ability remains challenging for open-domain tasks, where verifying procedures through execution is often impractical. Existing evaluations typically assess only the final plan and do not distinguish procedural errors such as missing, incorrect, or misordered actions. We introduce STEPBENCH, a benchmark of 6,885 verified procedures crawled from wikiHow across 81 open-domain categories. From each procedure, we derive 5 diagnostic tasks: four with deterministic answers involving withheld, incorrect, or shuffled steps, and one requiring next-step generation. We additionally evaluate full-plan generation to examine how diagnostic performance correlates with end-to-end plan validity. Using the same procedures across tasks lets us directly compare different planning capabilities. For generation, STEPEVAL identifies critical failures that prevent goal completion. Across 14 configurations of 12 models, we find substantial differences across planning capabilities, with step ordering proving the most difficult: the strongest model correctly orders only 63.0% of procedures, dropping to 33.1% for those with nine or more steps. Incorrect-step detection also predicts judged plan validity across models, suggesting a lower-cost alternative for evaluating generation quality. On STEPBENCH-Val, comprising 214 procedures published in 2026, comprehension scores and rankings remain stable, suggesting limited influence from memorisation.

cs.AI↗
arXiv2026

Hear2Act: Benchmarking When Prosody Should Change What an Assistant Does

Prosodic cues can carry task-relevant information that words omit, but existing benchmarks rarely test whether it changes downstream decisions. We introduce Hear2Act, a 480-scenario protocol with unstated user needs and verifiable task success. Across matched scenarios, we keep the task and user needs fixed, varying whether an unmet need is stated in words or absent from the transcript while the user's reaction is conveyed through prosody, yielding Explicit lexical and Prosody-mediated conditions. We vary assistant access to transcript, audio, and an indication of whether the recommendation is resolved, and evaluate two audio-capable LLMs and five text LLMs on behavior and task outcomes. Our results show an asymmetry: prosodic information is valuable when the need is absent from the transcript, but adds little when that need is stated in words. Yet neither audio-capable LLM benefits much from raw audio: adding audio to the transcript changes task success only from 14.3% to 15.1%. When models infer from audio whether the recommendation remains unresolved and use that inference with the transcript, success rises to 39.6%, close to 40.5% with ground-truth state. This reveals a perception-action gap: raw audio barely helps, while making the information explicit in text enables much better decisions.

cs.CL↗
arXiv2026

PersEvalBench: Can LLM Evaluators Judge Contextually Appropriate Use of Persona Information?

Large language model (LLM) evaluators are increasingly used to assess personalized assistants, yet whether they can reliably judge personalization has rarely been tested. We introduce PersEvalBench, a benchmark testing whether an evaluator can recognize appropriate and inappropriate uses of user history. It contains 403 multi-episode scenarios where user information should persist, transfer, or cease to apply. The benchmark covers nine controlled error subtypes across three families: under-personalization, over-personalization, and misapplied-state errors. Human audits confirm that generated errors match their intended types and that the prescribed uses of user information are reasonable. Across five LLM evaluators, recognition of controlled personalization errors remains limited even with the full dialogue history. On the human-evaluated subset, under-personalization shows the largest gap: humans recognize 94.4% of these errors, compared with 63.9% for the strongest reported evaluator. Evaluators often recognize that a response is problematic but reverse the error direction, mistaking failure to follow an applicable preference (UNDER) for use of an inapplicable preference (OVER). Easier evidence access offers little help: highlighting the relevant history turns raises error recognition only slightly, suggesting that the remaining difficulty lies in judging what role that information should play in the response.

cs.CL↗
arXiv2026

ATOD: An Evaluation Framework and Benchmark for Agentic Task-Oriented Dialogue Systems

Agentic task-oriented dialogue (TOD) requires systems to track concurrent goals, dependencies, and long-horizon state. We examine goal-lifecycle recovery from fixed dialogue trajectories. ATOD contains 1,000 synthetic dialogues annotated for six advanced-TOD properties, and ATOD-Eval defines metrics for dependency-sensitive completion, memory recall, and proactivity. We implement a symbolic-vector memory evaluator for ATOD-Eval. Among five prompt-only predictors and six matched-backbone memory baselines, our evaluator is the only configuration above 90% in both goal detection F1 and conditional status accuracy on medium dialogues. On complex dialogues, no baseline exceeds it on both metrics; it has the highest conditional status accuracy within the memory-based block and the lowest measured per-turn latency. These experiments assess lifecycle tracking rather than interactive agent task success. A cross-family judge swap and a manual audit with 94.0% agreement provide initial checks on measurement reliability. Code and data will be released at https://github.com/amazon-science/ATOD.

cs.CL↗
arXiv2026

See it, Say it, Sorted: Mechanistic Diagnosis and Parameter-Space Mitigation of Emergent Misalignment in LLMs

Safety-aligned LLMs can exhibit emergent misalignment (EM): narrow domain adaptation unexpectedly triggers catastrophic safety failures across unrelated domains. Prior static analyses leave training dynamics unmapped, while existing defenses rely on heuristics that degrade utility. We present a dynamic, second-order geometric study of EM. Tracking training trajectories reveals that directional Hessian curvature concentrates sharply on semantic pivot tokens. Grassmannian projections show that, in most settings, harmful-safe gap widens mainly because safe-gradient overlap declines. Leveraging these insights, we introduce a parameter-level Geometric Mitigation Framework that orthogonally projects empirical harmful gradient subspace out of parameter updates. On Qwen2.5-14B-IT, our defense suppresses free-generation EM by up to 80.0%; across the other three of four open-weight instruction-based model families (3B--20B), where single-layer behavioral EM is already near zero, teacher-forced evaluation shows same harmful subspace controls the conditional support of frozen EM responses. Crucially, these diagnostics unmask the illusion of behavioral safety: the same subspace remains measurable and steerable in models where behavioral EM is near zero. Code: https://github.com/WeiqiaoQUE/mechanistic-emergent-misalignment.

cs.LG↗
arXiv2026

Automatic multimodal UX improvement recommendations from LLM agent user simulations

Evaluating user experience (UX) on live websites through user testing is expensive, subjective, and difficult to scale. LLM agents offer a promising route to automating UX testing by simulating realistic user behaviour. However, existing simulation approaches typically lack multimodality and require time-consuming manual review to extract actionable insights. We formalise UX improvement recommendation from simulation data as a structured natural language generation and ranking problem, and establish an evaluation protocol using expert annotation and LLM-as-a-Judge. We present AMUSER, a multimodal framework which simulates user behaviour and automatically generates prioritised UX improvement recommendations from resulting data. We evaluate AMUSER on commercial websites and show that its recommendations substantially outperform those from text-only simulation (NDCG@3 = 0.758 versus 0.359) at an 89% lower simulation cost. Our results suggest an asymmetric role of multimodality: visual access during simulation improves recommendations through richer traces, while providing visual inputs during recommendation generation can modestly degrade quality. We also discuss practical deployment lessons from applying AMUSER to commercial websites.

cs.CL↗
arXiv2026

SciAtlas: A Computable Atlas of Science for Knowledge-Grounded AI Research

Artificial intelligence is rapidly entering the core workflows of scientific research. Yet reliable scientific reasoning requires access to accumulated scientific knowledge with sufficient breadth, depth, and standardization. Current AI scientists typically assemble scientific knowledge through workflow- and discipline-specific pipelines, which provide incomplete coverage, leave relations implicit, and make knowledge acquisition pathways fragmented. Here we present SciAtlas, a shared, machine-actionable cross-disciplinary scholarly knowledge infrastructure that integrates evidential, conceptual, disciplinary, expertise, and normative layers under a shared schema. SciAtlas further achieves a unified neuro-symbolic retrieval mechanism that grounds heterogeneous research objects, propagates relevance across the scholarly topology, and projects the resulting relevance field into the context required by each scientific workflow. Across three representative workflows, SciAtlas broadens trajectory reconstruction by recovering overlooked research branches, deepens opportunity discovery by uncovering underexplored bottlenecks and cross-domain insights, and strengthens innovation assessment by integrating evidence, expertise, and evaluation signals. Across three representative workflows, SciAtlas broadens trajectory reconstruction by recovering overlooked stages and branches, deepens opportunity discovery by uncovering underexplored bottlenecks and cross-domain connections, and standardizes innovation assessment by integrating evidence, expertise and evaluation signals. Extensive evaluations validate the foundational capabilities underpinning it as reusable knowledge infrastructure for knowledge-intensive scientific research.

cs.AI↗
arXiv2026

Retrieval-Augmented Agentic Rubric Generation for Reliable Medical Response Evaluation

Large Language Models (LLMs) are increasingly used for clinical decision support, where hallucinations and unsafe suggestions may pose direct risks to patient safety. These risks are hard to assess: subtle clinical errors are often missed by generic metrics and LLM judges using general criteria, while expert-authored fine-grained rubrics are expensive and difficult to scale. In this paper, we propose a retrieval-augmented multi-agent framework for automatically generating instance-specific evaluation rubrics. Our approach grounds evaluation in authoritative medical evidence by decomposing retrieved content into atomic facts and synthesizing them with user interaction constraints to form fine-grained evaluation criteria. Evaluated on HealthBench and LLMEval-Med, our framework achieves Clinical Intent Alignment (CIA) scores of 50.20% and 31.90%, significantly outperforming the GPT-4o baseline and showing consistent improvements across English and Chinese medical benchmarks. In discriminative tests on HealthBench, our rubrics achieve a 7.8% point higher win rate than GPT-4o and increase the mean score difference from 4.972 to 8.658. Ablation studies further show that individual components contribute differently across datasets, with interaction-intent modeling providing the most consistent contribution to clinical-criterion coverage. Beyond evaluation, our rubrics guide response refinement, improving response quality by 9.2%. These results suggest that automated, knowledge-grounded rubric generation provides a scalable foundation for evaluating and improving medical LLMs. The code is available at https://anonymous.4open.science/r/Automated-Rubric-Generation-E716/.

cs.CL↗
arXiv2026

BiasTrace: Linking Reasoning Behaviours to Biased Outputs in LLMs

LLMs exhibit social biases that can produce inaccurate and discriminatory inferences, posing risks in high-stakes applications. While prior work has made progress in measuring and mitigating bias, it largely focuses on final outputs of models, with limited understanding of the mechanisms that produce biased outcomes. Recent advances in LLM reasoning offers a new lens for investigating bias, yet the link between reasoning and bias remains poorly understood. Existing approaches focus primarily on final answer correctness or explicitly biased language, overlooking different behaviours in reasoning that can drive biased outcomes. We introduce BiasTrace, an annotation scheme for labelling reasoning behaviours in model-generated traces and linking them to biased outcomes. BiasTrace captures bias-specific behaviours (e.g., unsupported demographic assumptions) as well as general reasoning patterns that may implicitly contribute to bias (e.g. overthinking). We apply BiasTrace to reasoning traces in bias-sensitive contexts, scaled using validated LLM-as-a-judge methods, producing a large annotated dataset. Our analysis shows that biased outputs often stem from subtle reasoning behaviours rather than explicitly biased language, and that reasoning-level annotations improve bias detection. We further show that BiasTrace behaviours can be exploited for inference-time mitigation. These findings underscore the importance of examining a broader range of reasoning patterns to better understand bias in LLMs.

cs.AI↗
arXiv2026

Mitigating Context Interference for Reliable and Efficient Search Agents

Recent research empowers Large Language Models (LLMs) as multi-turn search agents to iteratively retrieve and generate outputs until complex tasks are solved. However, the contexts of multi-turn search agents are lengthy and complex. For example, the retrieved set of documents in each turn would inevitably introduce irrelevant information that distracts LLMs, referring to \textit{context interference}, potentially hindering the reliability and efficiency of search agents. Therefore, we conduct a systematic study on context interference in multi-turn search agents, focusing on investigating i) which parts of the context of search agents will contribute to the context interference, ii) how to refine the contexts of search agents to mitigate the interference, and iii) can incorporating context refinement into search agent training yield further improvements. We reveal that interference primarily arises from the latest retrieved documents. Based on the explored findings, we then introduce a distill-based context refiner to dynamically mitigate context interference for multi-turn search agents. Finally, we validate that incorporating context refinement into RL training pipelines of search agents can significantly enhance both reliability and efficiency. This study highlights the importance of mitigating context interference of search agents, inspiring a novel paradigm of ``refine context and then generate'' for AI agents.

cs.CL↗
arXiv2026

InnoEval: On Research Idea Evaluation as a Knowledge-Grounded, Multi-Perspective Reasoning Problem

The rapid evolution of Large Language Models has catalyzed a surge in scientific idea production, yet this leap has not been accompanied by a matching advance in idea evaluation. The fundamental nature of scientific evaluation needs knowledgeable grounding, collective deliberation, and multi-criteria decision-making. However, existing idea evaluation methods often suffer from narrow knowledge horizons, flattened evaluation dimensions, and the inherent bias in LLM-as-a-Judge. To address these, we regard idea evaluation as a knowledge-grounded, multi-perspective reasoning problem and introduce InnoEval, a deep innovation evaluation framework designed to emulate human-level idea assessment. We apply a heterogeneous deep knowledge search engine that retrieves and grounds dynamic evidence from diverse online sources. We further achieve review consensus with an innovation review board containing reviewers with distinct academic backgrounds, enabling a multi-dimensional decoupled evaluation across multiple metrics. We construct comprehensive datasets derived from authoritative peer-reviewed submissions to benchmark InnoEval. Experiments demonstrate that InnoEval can consistently outperform baselines in point-wise, pair-wise, and group-wise evaluation tasks, exhibiting judgment patterns and consensus highly aligned with human experts.

cs.CL↗
arXiv2026

Beyond Individual Personas: Aligning Synthetic Dialogue to Population-Level Behavior Distributions

Synthetic dialogue corpora are increasingly used as proxies for target dialogue data, yet persona-grounded generators optimize individual conversations rather than corpus composition, yielding locally plausible dialogues with distorted population-level behavior mixes. We introduce GroupPersona, a framework that aligns synthetic dialogue corpora to the behavior distribution of a reference corpus. GroupPersona turns population statistics into generation controls: it separates each dialogue's core behavioral signature from predictable side effects, and uses the resulting behavioral groups to condition user agents on the interaction patterns that define the reference population. We evaluate GroupPersona on four corpora crossing two dialogue sources, assistant-style and Reddit-derived, with two construction variants: structure-preserving and variation-enhanced. GroupPersona lowers Jensen-Shannon divergence between synthetic and reference distributions over 12 behavior attributes from 0.234 to 0.177 relative to the strongest average baseline, a 24.4% reduction, and is best or tied-best on all four corpora while preserving structural alignment. It also achieves the closest calibration to reference-conversation quality scores, reducing mean absolute deviation from the reference-conversation profile to 0.63 versus 0.91 for the next-best baseline.

cs.CL↗
arXiv2026

AgentSearchBench: A Benchmark for AI Agent Search in the Wild

The rapid growth of AI agent ecosystems is transforming how complex tasks are delegated and executed, creating a new challenge of identifying suitable agents for a given task. Unlike traditional tools, agent capabilities are often compositional and execution-dependent, making them difficult to assess from textual descriptions alone. However, existing research and benchmarks typically assume well-specified functionalities, controlled candidate pools, or only executable task queries, leaving realistic agent search scenarios insufficiently studied. We introduce AgentSearchBench, a large-scale benchmark for agent search in the wild, built from nearly 10,000 real-world agents across multiple providers. The benchmark formalizes agent search as retrieval and reranking problems under both executable task queries and high-level task descriptions, and evaluates relevance using execution-grounded performance signals. Experiments reveal a consistent gap between semantic similarity and actual agent performance, exposing the limitations of description-based retrieval and reranking methods. We further show that lightweight behavioral signals, including execution-aware probing, can substantially improve ranking quality, highlighting the importance of incorporating execution signals into agent discovery. Our code is available at https://github.com/Bingo-W/AgentSearchBench.

cs.AI↗