Designing a generalist scientific agent capable of performing tasks in laboratory settings to assist researchers has become a key goal in recent Artificial Intelligence (AI) research. Unlike everyday tasks, scientific tasks are inherently more delicate and complex, requiring agents to possess a higher level of reasoning ability, structured and temporal understanding of their environment, and a strong emphasis on safety. Existing approaches often fail to address these multifaceted requirements. To tackle these challenges, we present DAVIS. Unlike traditional retrieval-augmented generation (RAG) approaches, DAVIS incorporates structured and temporal memory, which enables model-based planning. Additionally, DAVIS implements an agentic, multi-turn retrieval system, similar to a human’s inner monologue, allowing for a greater degree of reasoning over past experiences. DAVIS demonstrates substantially improved performance on the ScienceWorld benchmark comparing to previous approaches on 8 out of 9 elementary science subjects. In addition, DAVIS’s World Model demonstrates competitive performance on the famous HotpotQA and MusiqueQA dataset for multi-hop question answering. To the best of our knowledge, DAVIS is the first RAG agent to employ an interactive retrieval method in a RAG pipeline.
论文检索
输入标题、作者或关键词，从 12,319 篇学术成果中精准定位
DrKGC: Dynamic Subgraph Retrieval-Augmented LLMs for Knowledge Graph Completion across General and Biomedical Domains
PDF ↗Knowledge graph completion (KGC) aims to predict missing triples in knowledge graphs (KGs) by leveraging existing triples and textual information. Recently, generative large language models (LLMs) have been increasingly employed for graph tasks. However, current approaches typically encode graph context in textual form, which fails to fully exploit the potential of LLMs for perceiving and reasoning about graph structures. To address this limitation, we propose DrKGC (Dynamic Subgraph Retrieval-Augmented LLMs for Knowledge Graph Completion). DrKGC employs a flexible lightweight model training strategy to learn structural embeddings and logical rules within the KG. It then leverages a novel bottom-up graph retrieval method to extract a subgraph for each query guided by the learned rules. Finally, a graph convolutional network (GCN) adapter uses the retrieved subgraph to enhance the structural embeddings, which are then integrated into the prompt for effective LLM fine-tuning. Experimental results on two general domain benchmark datasets and two biomedical datasets demonstrate the superior performance of DrKGC. Furthermore, a realistic case study in the biomedical domain highlights its interpretability and practical utility.
TTPA: Token-level Tool-use Preference Alignment Training Framework with Fine-grained Evaluation
PDF ↗Existing tool-learning methods usually rely on supervised fine-tuning, they often overlook fine-grained optimization of internal tool call details, leading to limitations in preference alignment and error discrimination. To overcome these challenges, we propose **T**oken-level **T**ool-use **P**reference **A**lignment Training Framework (TTPA), a training paradigm for constructing token-level tool-use preference datasets that align LLMs with fine-grained preferences using a novel error-oriented scoring mechanism. TTPA first introduces reversed dataset construction, a method for creating high-quality, multi-turn tool-use datasets by reversing the generation flow. Additionally, we propose _Preference Oriented Tool-use Dataset Construction_ to capture fine-grained preferences by modeling token-level differences during generation. To address biases in scoring, we introduce the _Error-oriented Scoring Mechanism_, which quantifies tool-call errors and can be used as a training signal. Extensive experiments on three diverse benchmark datasets demonstrate that TTPA significantly improves tool-using performance while showing strong generalization ability across models and datasets.
Hallucination remains a key challenge in applying large language models (LLMs) to structured query generation, especially for semi-private or domain-specific languages underrepresented in public training data. In this work, we focus on hallucination detection in these low-resource structured language scenarios, using Splunk Search Processing Language (SPL) as a representative case study. We start from analyzing real-world SPL generation to define hallucination in this context and introduce a comprehensive taxonomy. To enhance detection performance, we propose the Self-Debating framework, which prompts an LLM to generate contrastive explanations from opposing perspectives before rendering a final consistency judgment. We also construct a synthetic benchmark, SynSPL, to support systematic evaluation of hallucination detection in SPL generation. Experimental results show that Self-Debating consistently outperforms LLM-as-a-Judge baselines with zero-shot and chain-of-thought (CoT) prompts in SPL hallucination detection across different LLMs, yielding 5–10% relative gains in hallucination F1 scores on both real and synthetic datasets, and up to 260% improvement for LLaMA-3.1–8B. Besides hallucination detection on SPL, Self-Debating also achieves excellent performance on the FaithBench benchmark for summarization hallucination, demonstrating the strong generalization ability of Self-Debating, with OpenAI o1-mini achieving state-of-the-art performance. All these results consistently demonstrate the strong robustness and wide generalizability of Self-Debating.
Recent years have witnessed a growing interest in personalizing the responses of large language models (LLMs). While existing evaluations primarily focus on whether a response aligns with a user’s preferences, we argue that factuality is an equally important yet often overlooked dimension. In the context of personalization, we define a model as robust if its responses are both factually accurate and align with the user preferences. To assess this, we introduce PERG, a scalable framework for evaluating robustness of LLMs in personalization, along with a new dataset, PERGData. We evaluate fourteen models from five different model families using different prompting methods. Our findings show that current LLMs struggle with robust personalization: even the strongest models (GPT-4.1, LLaMA3-70B) fails to maintain correctness in 5% of previously successful cases without personalization, while smaller models (e.g., 7B scale) can fail more than 20% of the time. Further analysis reveals that robustness is significantly affected by the nature of the query and the type of user preference. To mitigate these failures, we propose Pref-Aligner, a two-stage approach that improves robustness by an average of 25% across models. Our work highlights critical gaps in current evaluation practices and introduces tools and metrics to support more reliable, user-aligned LLM deployments.
MFTCXplain: A Multilingual Benchmark Dataset for Evaluating the Moral Reasoning of LLMs through Multi-hop Hate Speech Explanation
PDF ↗Ensuring the moral reasoning capabilities of Large Language Models (LLMs) is a growing concern as these systems are used in socially sensitive tasks. Nevertheless, current evaluation benchmarks present two major shortcomings: a lack of annotations that justify moral classifications, which limits transparency and interpretability; and a predominant focus on English, which constrains the assessment of moral reasoning across diverse cultural settings. In this paper, we introduce MFTCXplain, a multilingual benchmark dataset for evaluating the moral reasoning of LLMs via multi-hop hate speech explanations using the Moral Foundations Theory. MFTCXplain comprises 3,000 tweets across Portuguese, Italian, Persian, and English, annotated with binary hate speech labels, moral categories, and text span-level rationales. Our results show a misalignment between LLM outputs and human annotations in moral reasoning tasks. While LLMs perform well in hate speech detection (F1 up to 0.836), their ability to predict moral sentiments is notably weak (F1 < 0.35). Furthermore, rationale alignment remains limited mainly in underrepresented languages. Our findings show the limited capacity of current LLMs to internalize and reflect human moral reasoning.
We propose XRAG, a novel benchmark designed to evaluate the generation abilities of LLMs in cross-lingual Retrieval-Augmented Generation (RAG) settings where the user language does not match the retrieval results. XRAG is constructed from recent news articles to ensure that its questions require external know-ledge to be answered. It covers the real-world scenarios of monolingual and multilingual retrieval, and provides relevancy annotations for each retrieved document. Our novel dataset construction pipeline results in questions that require complex reasoning, as evidenced by the significant gap between human and LLM performance. Consequently, XRAG serves as a valuable benchmark for studying LLM reasoning abilities, even before considering the additional cross-lingual complexity. Experimental results on five LLMs uncover two previously unreported challenges in cross-lingual RAG: 1) in the monolingual retrieval setting, all evaluated models struggle with response language correctness; 2) in the multilingual retrieval setting, the main challenge lies in reasoning over retrieved information across languages rather than generation of non-English text.
Understanding and solving complex reasoning tasks is vital for addressing the information needs of a user. Although dense neural models learn contextualised embeddings, they underperform on queries containing negation. To understand this phenomenon, we study negation in traditional neural information retrieval and LLM-based models. We (1) introduce a taxonomy of negation that derives from philosophical, linguistic, and logical definitions; (2) generate two benchmark datasets that can be used to evaluate the performance of neural information retrieval models and to fine-tune models for a more robust performance on negation; and (3) propose a logic-based classification mechanism that can be used to analyze the performance of retrieval models on existing datasets. Our taxonomy produces a balanced data distribution over negation types, providing a better training setup that leads to faster convergence on the NevIR dataset. Moreover, we propose a classification schema that reveals the coverage of negation types in existing datasets, offering insights into the factors that might affect the generalization of fine-tuned models on negation. Our code is publicly available on GitHub, and the datasets are available on HuggingFace.
Time series Forecasting with large language models (LLMs) requires bridging numerical patterns and natural language. Effective forecasting on LLM often relies on extensive pre-processing and fine-tuning. Recent studies show that a frozen LLM can rival specialized forecasters when supplied with a carefully engineered natural-language prompt, but crafting such a prompt for each task is itself onerous and ad-hoc. We introduce FLAIRR-TS, a test-time prompt optimization framework that utilizes an agentic system: a Forecaster-agent generates forecasts using an initial prompt, which is then refined by a refiner agent, informed by past outputs and retrieved analogs. This adaptive prompting generalizes across domains using creative prompt templates and generates high-quality forecasts without intermediate code generation. Experiments on benchmark datasets show FLAIRR-TS improves forecasting over static prompting and retrieval-augmented baselines, approaching the performance of specialized prompts.FLAIRR-TS provides a practical alternative to fine-tuning, achieving strong performance via its agentic approach to adaptive prompt refinement and retrieval.
Context Minimization for Resource-Constrained Text Classification: Optimizing Performance-Efficiency Trade-offs through Linguistic Features
PDF ↗Pretrained language models have transformed text classification, yet their computational demands often render them impractical for resource-constrained settings. We propose a linguistically-grounded framework for context minimization that leverages theme-rheme structure to preserve critical classification signals while reducing input complexity. Our approach integrates positional, syntactic, semantic, and statistical features, guided by functional linguistics, to identify optimal low-context configurations. We present a methodical iterative feature exploration protocol across 6 benchmarks, including our novel CMLA11 dataset. Results demonstrate substantial efficiency gains: 69-75% reduction in GPU memory, 81-87% decrease in training time, and 82-88% faster inference. Despite these resource savings, our configurations maintain near-parity with full-length inputs, with F1 (macro) reductions averaging just 1.39-3.10%. Statistical significance testing confirms minimal practical impact, with some configurations outperforming the baseline. SHAP analysis reveals specific feature subsets contribute most significantly across datasets, and these recurring configurations offer transferable insights, reducing the need for exhaustive feature exploration. Our method also yields remarkable data compression (72.57% average reduction, reaching 92.63% for longer documents). Ablation studies confirm synergistic feature contributions, establishing our context minimization as an effective solution for resource-efficient text classification with minimal performance trade-offs.
Long-form egocentric video understanding provides rich contextual information and unique insights into long-term human behaviors, holding significant potential for applications in embodied intelligence, long-term activity analysis, and personalized assistive technologies. However, existing benchmark datasets primarily focus on single, short (e.g., minutes to tens of minutes) to moderately long videos, leaving a substantial gap in evaluating extensive, ultra-long egocentric video recordings. To address this, we introduce X-LeBench, a novel benchmark dataset meticulously designed to fill this gap by focusing on tasks requiring a comprehensive understanding of extremely long egocentric video recordings. Our X-LeBench develops a life-logging simulation pipeline that produces realistic, coherent daily plans aligned with real-world video data. This approach enables the flexible integration of synthetic daily plans with real-world footage from Ego4D—a massive-scale egocentric video dataset covers a wide range of daily life scenarios—resulting in 432 simulated video life logs spanning from 23 minutes to 16.4 hours. The evaluations of several baseline systems and multimodal large language models (MLLMs) reveal their poor performance across the board, highlighting the inherent challenges of long-form egocentric video understanding, such as temporal localization and reasoning, context aggregation, and memory retention, and underscoring the need for more advanced models.
Can Large Language Models (LLMs) simulate humans in making important decisions? Recent research has unveiled the potential of using LLMs to develop role-playing language agents (RPLAs), mimicking mainly the knowledge and tones of various characters. However, imitative decision-making necessitates a more nuanced understanding of personas. In this paper, we benchmark the ability of LLMs in persona-driven decision-making. Specifically, we investigate whether LLMs can predict characters’ decisions provided by the preceding stories in high-quality novels. Leveraging character analyses written by literary experts, we construct a dataset LIFECHOICE comprising 2,512 characters’ decision points from 470 books. Then, we conduct comprehensive experiments on LIFECHOICE with various LLMs and RPLA methodologies. The results demonstrate that state-of-the-art LLMs exhibit promising capabilities in this task, yet substantial room for improvement remains. Hence, we further propose the CHARMAP method, which adopts persona-based memory retrieval and significantly advances RPLAs on this task.
Concerns about text-to-image (T2I) generative models infringing on privacy, copyright, and safety have led to the development of concept erasure techniques (CETs). The goal of an effective CET is to prohibit the generation of undesired “target” concepts specified by the user, while preserving the ability to synthesize high-quality images of other concepts. In this work, we demonstrate that concept erasure has side effects and CETs can be easily circumvented. For a comprehensive measurement of the robustness of CETs, we present the Side Effect Evaluation (SEE) benchmark that consists of hierarchical and compositional prompts describing objects and their attributes. The dataset and an automated evaluation pipeline quantify side effects of CETs across three aspects: impact on neighboring concepts, evasion of targets, and attribute leakage. Our experiments reveal that CETs can be circumvented by using superclass-subclass hierarchy, semantically similar prompts, and compositional variants of the target. We show that CETs suffer from attribute leakage and a counterintuitive phenomenon of attention concentration or dispersal. We release our benchmark and evaluation tools to aid future work on robust concept erasure.
[MASK]ED - Language Modeling for Explainable Classification and Disentangling of Socially Unacceptable Discourse.
PDF ↗Analyzing Socially Unacceptable Discourse (SUD) online is a critical challenge for regulators and platforms amidst growing concerns over harmful content. While Pre-trained Masked Language Models (PMLMs) have proven effective for many NLP tasks, their performance often degrades in multi-label SUD classification due to overlapping linguistic cues across categories. In this work, we propose an artifact-guided pre-training strategy that injects statistically salient linguistic features, referred to as artifacts, into the masked language modelling objective. By leveraging context-sensitive tokens, we guide an importance-weighted masking scheme during pre-training to enhance generalization across discourse types. We further use these artifact signals to inform a lightweight dataset curation procedure that highlights noisy or ambiguous instances. This supports targeted relabeling and filtering, enabling more explainable and consistent annotation with minimal changes to the original data. Our approach provides consistent improvements in 10 datasets extensively used in SUD classification benchmarks.*Disclaimer: This article contains some extracts of unacceptable and upsetting language.*
Data-efficient training requires strong inductive biases. To the extent that transformer attention matrices encode syntactic relationships, we would predict that knowledge distillation (KD) targeting attention should selectively accelerate syntax acquisition relative to conventional logit-based KD. To test this hypothesis, we train GPT-2 student models on datasets ranging from 10K to 5M sentences using both distillation methods, evaluating them on both syntactic benchmarks and perplexity. Surprisingly, while logit-based KD dramatically improves data-efficiency, attention-based KD provides minimal benefit even for syntactic tasks. This suggests that output distributions provide sufficient supervisory signal for syntax acquisition, indicating that syntactic knowledge may be distributed throughout the network rather than localized in attention patterns.
FactReasoner: A Probabilistic Approach to Long-Form Factuality Assessment for Large Language Models
PDF ↗Large language models (LLMs) have achieved remarkable success in generative tasks, yet they often fall short in ensuring the factual accuracy of their outputs thus limiting their reliability in real-world applications where correctness is critical. In this paper, we present FactReasoner, a novel neuro-symbolic based factuality assessment framework that employs probabilistic reasoning to evaluate the truthfulness of long-form generated responses. FactReasoner decomposes a response into atomic units, retrieves relevant contextual information from external knowledge sources, and models the logical relationships (e.g., entailment, contradiction) between these units and their contexts using probabilistic encodings. It then estimates the posterior probability that each atomic unit is supported by the retrieved evidence. Our experiments on both labeled and unlabeled benchmark datasets demonstrate that FactReasoner often outperforms state-of-the-art prompt-based methods in terms of factual precision and recall.
Human-Inspired Obfuscation for Model Unlearning: Local and Global Strategies with Hyperbolic Representations
PDF ↗Large language models (LLMs) achieve remarkable performance across various domains, largely due to training on massive datasets. However, this also raises growing concerns over the exposure of sensitive and private information, making model unlearning increasingly critical.However, existing methods often struggle to balance effective forgetting with maintaining model utility. In this work, we propose HyperUnlearn, a human-inspired unlearning framework. We construct two types of fuzzy data—local and global—to simulate forgetting, and represent them in hyperbolic and Euclidean spaces, respectively. Unlearning is performed on a model with frozen early layers to isolate forgetting and preserve useful knowledge.Experiments demonstrate that HyperUnlearn effectively forgets sensitive content while maintaining the model’s language understanding, fluency, and benchmark performance, offering a practical trade-off between forgetting and capability preservation.
Pre-training on large, high-quality datasets is essential for improving the reasoning abilities of Large Language Models (LLMs), particularly in specialized fields like mathematics. However, the field of Multimodal LLMs (MLLMs) lacks a comprehensive, open-source dataset for mathematical reasoning. To fill this gap, we present InfiMM-WebMath-40B, a high-quality dataset of interleaved image-text documents. It consists of 24 million web pages, 85 million image URLs, and 40 billion text tokens, all carefully extracted and filtered from CommonCrawl. We outline our data collection and processing pipeline in detail. Models trained on InfiMM-WebMath-40B demonstrate strong performance in both text-only and multimodal settings, setting a new state-of-the-art on multimodal math benchmarks such as MathVerse and We-Math.
Keyword extraction involves identifying the most descriptive words in a document, allowing automatic categorisation and summarisation of large quantities of diverse textual data. Relying on the insight that real-world keyword detection often requires handling of diverse content, we propose a novel supervised keyword extraction approach based on the mixture of experts (MoE) technique. MoE uses a learnable routing sub-network to direct information to specialised experts, allowing them to specialise in distinct regions of the input space. SEKE, a mixture of Specialised Experts for supervised Keyword Extraction, uses DeBERTa as the backbone model and builds on the MoE framework, where experts attend to each token, by integrating it with a bidirectional Long short-term memory (BiLSTM) network, to allow successful extraction even on smaller corpora, where specialisation is harder due to lack of training data. The MoE framework also provides an insight into inner workings of individual experts, enhancing the explainability of the approach. We benchmark SEKE on multiple English datasets, achieving state-of-the-art performance compared to strong supervised and unsupervised baselines. Our analysis reveals that depending on data size and type, experts specialise in distinct syntactic and semantic components, such as punctuation, stopwords, parts-of-speech, or named entities. Code is available at https://github.com/matejMartinc/SEKE_keyword_extraction.
Benchmarking Contextual and Paralinguistic Reasoning in Speech-LLMs: A Case Study with In-the-Wild Data
PDF ↗Recent speech-LLMs have shown impressive performance in tasks like transcription and translation, yet they remain limited in understanding the paralinguistic aspects of speech crucial for social and emotional intelligence. We propose CP-Bench, a benchmark for evaluating speech-LLMs on contextual paralinguistic reasoning the integration of verbal content with non-verbal cues like emotion and prosody. The benchmark includes two curated question answering (QA) datasets requiring both linguistic and empathetic understanding. We evaluate state-of-the-art speech-LLMs from both open and closed-source models and perform a comprehensive analysis across different question types. The top two models were further analyzed under temperature tuning to understand its effect on this task. Our benchmark reveals a key gap in existing evaluations and offers insights into building more context-aware and emotionally intelligent speech-capable LLMs.