CyberSec Research

Digital health analytics face critical challenges nowadays. The sophisticated analysis of patient-generated health content, which contains complex emotional and medical contexts, requires scarce domain expertise, while traditional ML approaches are constrained by data shortage and privacy limitations in healthcare settings. Online Health Communities (OHCs) exemplify these challenges with mixed-sentiment posts, clinical terminology, and implicit emotional expressions that demand specialised knowledge for accurate Sentiment Analysis (SA). To address these challenges, this study explores how Large Language Models (LLMs) can integrate expert knowledge through in-context learning for SA, providing a scalable solution for sophisticated health data analysis. Specifically, we develop a structured codebook that systematically encodes expert interpretation guidelines, enabling LLMs to apply domain-specific knowledge through targeted prompting rather than extensive training. Six GPT models validated alongside DeepSeek and LLaMA 3.1 are compared with pre-trained language models (BioBERT variants) and lexicon-based methods, using 400 expert-annotated posts from two OHCs. LLMs achieve superior performance while demonstrating expert-level agreement. This high agreement, with no statistically significant difference from inter-expert agreement levels, suggests knowledge integration beyond surface-level pattern recognition. The consistent performance across diverse LLM models, supported by in-context learning, offers a promising solution for digital health analytics. This approach addresses the critical challenge of expert knowledge shortage in digital health research, enabling real-time, expert-quality analysis for patient monitoring, intervention assessment, and evidence-based health strategies.

Beyond simple text generation, Large Language Models (LLMs) have evolved into agentic systems capable of planning and interacting with external tools to solve complex tasks. This evolution involves fine-tuning LLMs on agent-specific tasks to enhance their proficiency. However, safety concerns are frequently overlooked during this fine-tuning process. In this work, we show that aligned LLMs can become unintentionally misaligned, leading to a higher likelihood of executing harmful tasks and a reduced tendency to refuse them when fine-tuned to execute agentic tasks. To address these safety challenges, we propose Prefix INjection Guard (PING), a simple yet effective method that prepends automatically generated natural language prefixes to agent responses, guiding them to refuse harmful requests while preserving performance on benign tasks. Specifically, we introduce an iterative approach that alternates between (1) generating candidate prefixes and (2) selecting those that optimize both task performance and refusal behavior. Experimental results demonstrate that PING significantly enhances the safety of fine-tuned LLM agents without sacrificing their effectiveness. PING consistently outperforms existing prompting approaches across diverse benchmarks in both web navigation and code generation tasks. Our analysis of internal hidden states via linear probes reveals that prefix tokens are crucial for behavior modification, explaining the performance gains. WARNING: This paper contains contents that are unethical or offensive in nature.

Reinforcement Learning with Verifiable Rewards (RLVR) has recently emerged as a key paradigm for post-training Large Language Models (LLMs), particularly for complex reasoning tasks. However, vanilla RLVR training has been shown to improve Pass@1 performance at the expense of policy entropy, leading to reduced generation diversity and limiting the Pass@k performance, which typically represents the upper bound of LLM reasoning capability. In this paper, we systematically analyze the policy's generation diversity from the perspective of training problems and find that augmenting and updating training problems helps mitigate entropy collapse during training. Based on these observations, we propose an online Self-play with Variational problem Synthesis (SvS) strategy for RLVR training, which uses the policy's correct solutions to synthesize variational problems while ensuring their reference answers remain identical to the originals. This self-improving strategy effectively maintains policy entropy during training and substantially improves Pass@k compared with standard RLVR, sustaining prolonged improvements and achieving absolute gains of 18.3% and 22.8% in Pass@32 performance on the competition-level AIME24 and AIME25 benchmarks. Experiments on 12 reasoning benchmarks across varying model sizes from 3B to 32B consistently demonstrate the generalizability and robustness of SvS.

Recent advances in large language models (LLMs) have significantly improved the performance of dialog systems, yet current approaches often fail to provide accurate guidance of topic due to their inability to discern user confusion in related concepts. To address this, we introduce the Ask-Good-Question (AGQ) framework, which features an improved Concept-Enhanced Item Response Theory (CEIRT) model to better identify users' knowledge levels. Our contributions include applying the CEIRT model along with LLMs to directly generate guiding questions based on the inspiring text, greatly improving information retrieval efficiency during the question & answer process. Through comparisons with other baseline methods, our approach outperforms by significantly enhencing the users' information retrieval experiences.

Long-context modeling is critical for a wide range of real-world tasks, including long-context question answering, summarization, and complex reasoning tasks. Recent studies have explored fine-tuning Large Language Models (LLMs) with synthetic data to enhance their long-context capabilities. However, the effectiveness of such approaches is often limited by the low diversity and factual inconsistencies in the generated data. To address these challenges, we propose LongMab-PO, a novel framework that leverages a Multi-Armed Bandit (MAB) rollout strategy to identify the most informative chunks from the given long context for sampling high-quality and diverse responses and constructing preference data pairs for Direct Preference Optimization (DPO) training. Specifically, we treat context chunks as arms of MAB, select chunks based on their expected reward scores to input into LLMs to generate responses, and iteratively update these scores based on reward feedback. This exploration and exploitation process enables the model to focus on the most relevant context segments, thereby generating and collecting high-quality and diverse responses. Finally, we collect these generated responses from the rollout process and apply the DPO method to further optimize the LLM. Experimental results show that LongMab-PO significantly improves the diversity and quality of preference data pairs, achieving state-of-the-art performance on long-context reasoning benchmarks. All code and data will be released on https://github.com/NEUIR/LongMab-PO.

We investigate to what extent Multimodal Large Language Models (MLLMs) can accurately identify the orientation of input images rotated 0{\deg}, 90{\deg}, 180{\deg}, and 270{\deg}. This task demands robust visual reasoning capabilities to detect rotational cues and contextualize spatial relationships within images, regardless of their orientation. To evaluate MLLMs on these abilities, we introduce RotBench -- a 350-image manually-filtered benchmark comprising lifestyle, portrait, and landscape images. Despite the relatively simple nature of this task, we show that several state-of-the-art open and proprietary MLLMs, including GPT-5, o3, and Gemini-2.5-Pro, do not reliably identify rotation in input images. Providing models with auxiliary information -- including captions, depth maps, and more -- or using chain-of-thought prompting offers only small and inconsistent improvements. Our results indicate that most models are able to reliably identify right-side-up (0{\deg}) images, while certain models are able to identify upside-down (180{\deg}) images. None can reliably distinguish between 90{\deg} and 270{\deg}. Simultaneously showing the image rotated in different orientations leads to moderate performance gains for reasoning models, while a modified setup using voting improves the performance of weaker models. We further show that fine-tuning does not improve models' ability to distinguish 90{\deg} and 270{\deg} rotations, despite substantially improving the identification of 180{\deg} images. Together, these results reveal a significant gap between MLLMs' spatial reasoning capabilities and human perception in identifying rotation.

In the hospitality industry, understanding the factors that drive customer review ratings is critical for improving guest satisfaction and business performance. This work proposes ReviewGraph for Review Rating Prediction (RRP), a novel framework that transforms textual customer reviews into knowledge graphs by extracting (subject, predicate, object) triples and associating sentiment scores. Using graph embeddings (Node2Vec) and sentiment features, the framework predicts review rating scores through machine learning classifiers. We compare ReviewGraph performance with traditional NLP baselines (such as Bag of Words, TF-IDF, and Word2Vec) and large language models (LLMs), evaluating them in the HotelRec dataset. In comparison to the state of the art literature, our proposed model performs similar to their best performing model but with lower computational cost (without ensemble). While ReviewGraph achieves comparable predictive performance to LLMs and outperforms baselines on agreement-based metrics such as Cohen's Kappa, it offers additional advantages in interpretability, visual exploration, and potential integration into Retrieval-Augmented Generation (RAG) systems. This work highlights the potential of graph-based representations for enhancing review analytics and lays the groundwork for future research integrating advanced graph neural networks and fine-tuned LLM-based extraction methods. We will share ReviewGraph output and platform open-sourced on our GitHub page https://github.com/aaronlifenghan/ReviewGraph

In the digital era, user interactions with various resources such as databases, data warehouses, websites, and knowledge graphs (KGs) are increasingly mediated through digital platforms. These interactions leave behind digital traces, systematically captured in the form of logs. Logs, when effectively exploited, provide high value across industry and academia, supporting critical services (e.g., recovery and security), user-centric applications (e.g., recommender systems), and quality-of-service improvements (e.g., performance optimization). Despite their importance, research on log usage remains fragmented across domains, and no comprehensive study currently consolidates existing efforts. This paper presents a systematic survey of log usage, focusing on Database (DB), Data Warehouse (DW), Web, and KG logs. More than 300 publications were analyzed to address three central questions: (1) do different types of logs share common structural and functional characteristics? (2) are there standard pipelines for their usage? (3) which constraints and non-functional requirements (NFRs) guide their exploitation?. The survey reveals a limited number of end-to-end approaches, the absence of standardization across log usage pipelines, and the existence of shared structural elements among different types of logs. By consolidating existing knowledge, identifying gaps, and highlighting opportunities, this survey provides researchers and practitioners with a comprehensive overview of log usage and sheds light on promising directions for future research, particularly regarding the exploitation and democratization of KG logs.

Large Language Models (LLMs) require sophisticated prompting, yet current practices face challenges in structure, data integration, format sensitivity, and tooling. Existing methods lack comprehensive solutions for organizing complex prompts involving diverse data types (documents, tables, images) or managing presentation variations systematically. To address these gaps, we introduce POML (Prompt Orchestration Markup Language). POML employs component-based markup for logical structure (roles, tasks, examples), specialized tags for seamless data integration, and a CSS-like styling system to decouple content from presentation, reducing formatting sensitivity. It includes templating for dynamic prompts and a comprehensive developer toolkit (IDE support, SDKs) to improve version control and collaboration. We validate POML through two case studies demonstrating its impact on complex application integration (PomLink) and accuracy performance (TableQA), as well as a user study assessing its effectiveness in real-world development scenarios.

Recently, multimodal large language models (MLLMs) have achieved significant advancements across various domains, and corresponding evaluation benchmarks have been continuously refined and improved. In this process, benchmarks in the scientific domain have played an important role in assessing the reasoning capabilities of MLLMs. However, existing benchmarks still face three key challenges: 1) Insufficient evaluation of models' reasoning abilities in multilingual scenarios; 2) Inadequate assessment of MLLMs' comprehensive modality coverage; 3) Lack of fine-grained annotation of scientific knowledge points. To address these gaps, we propose MME-SCI, a comprehensive and challenging benchmark. We carefully collected 1,019 high-quality question-answer pairs, which involve 3 distinct evaluation modes. These pairs cover four subjects, namely mathematics, physics, chemistry, and biology, and support five languages: Chinese, English, French, Spanish, and Japanese. We conducted extensive experiments on 16 open-source models and 4 closed-source models, and the results demonstrate that MME-SCI is widely challenging for existing MLLMs. For instance, under the Image-only evaluation mode, o4-mini achieved accuracy of only 52.11%, 24.73%, 36.57%, and 29.80% in mathematics, physics, chemistry, and biology, respectively, indicating a significantly higher difficulty level compared to existing benchmarks. More importantly, using MME-SCI's multilingual and fine-grained knowledge attributes, we analyzed existing models' performance in depth and identified their weaknesses in specific domains. The Data and Evaluation Code are available at https://github.com/JCruan519/MME-SCI.

LLMs have recently been used to generate Python programs representing generalized plans in PDDL planning, i.e., plans that generalize across the tasks of a given PDDL domain. Previous work proposed a framework consisting of three steps: the LLM first generates a summary and then a strategy for the domain, both in natural language, and then implements that strategy as a Python program, that gets debugged on example planning tasks. In that work, only one strategy is generated and passed directly to the program generation. If the strategy is incorrect, its implementation will therefore result in an incorrect generalized plan. Here, we introduce an approach that generates the strategy in the form of pseudocode and enables automatic debugging of the pseudocode, hence allowing us to identify and fix errors prior to the generation of the generalized plan itself. Additionally, we extend the Python debugging phase with a reflection step prompting the LLM to pinpoint the reason for the observed plan failure. Finally, we take inspiration from LLM code generation to produce several program variants and pick the best one. Running experiments on 17 benchmark domains, we show that these extensions substantially improve (and never deteriorate) the quality of the generalized plans. In 12 of the domains, our best Python programs solve all tasks that can be generated with the respective instance generator.

This study explores the integration of Building Information Modeling (BIM) with Natural Language Processing (NLP) to automate the extraction of requirements from unstructured French Building Technical Specification (BTS) documents within the construction industry. Employing Named Entity Recognition (NER) and Relation Extraction (RE) techniques, the study leverages the transformer-based model CamemBERT and applies transfer learning with the French language model Fr\_core\_news\_lg, both pre-trained on a large French corpus in the general domain. To benchmark these models, additional approaches ranging from rule-based to deep learning-based methods are developed. For RE, four different supervised models, including Random Forest, are implemented using a custom feature vector. A hand-crafted annotated dataset is used to compare the effectiveness of NER approaches and RE models. Results indicate that CamemBERT and Fr\_core\_news\_lg exhibited superior performance in NER, achieving F1-scores over 90\%, while Random Forest proved most effective in RE, with an F1 score above 80\%. The outcomes are intended to be represented as a knowledge graph in future work to further enhance automatic verification systems.

Evaluating Natural Language Generation (NLG) is crucial for the practical adoption of AI, but has been a longstanding research challenge. While human evaluation is considered the de-facto standard, it is expensive and lacks scalability. Practical applications have driven the development of various automatic evaluation metrics (AEM), designed to compare the model output with human-written references, generating a score which approximates human judgment. Over time, AEMs have evolved from simple lexical comparisons, to semantic similarity models and, more recently, to LLM-based evaluators. However, it seems that no single metric has emerged as a definitive solution, resulting in studies using different ones without fully considering the implications. This paper aims to show this by conducting a thorough examination of the methodologies of existing metrics, their documented strengths and limitations, validation methods, and correlations with human judgment. We identify several key challenges: metrics often capture only specific aspects of text quality, their effectiveness varies by task and dataset, validation practices remain unstructured, and correlations with human judgment are inconsistent. Importantly, we find that these challenges persist in the most recent type of metric, LLM-as-a-Judge, as well as in the evaluation of Retrieval Augmented Generation (RAG), an increasingly relevant task in academia and industry. Our findings challenge the quest for the 'perfect metric'. We propose selecting metrics based on task-specific needs and leveraging complementary evaluations and advocate that new metrics should focus on enhanced validation methodologies.

Controlling the length of text produced by large language models (LLMs) remains challenging: models frequently overshoot or undershoot explicit length instructions because they cannot reliably keep an internal token count. We present a prompt-based, one-shot strategy that compels an off-the-shelf LLM to generate exactly a desired number of tokens - words (English) or characters (Chinese) - without any fine-tuning or iterative sampling. The prompt appends countdown markers and explicit counting rules so that the model "writes while counting." We evaluate on four settings: open-ended generation (1-1000 tokens), XSUM summarization, MT-Bench-LI instruction following, and the LIFEBENCH equal-length track. On MT-Bench-LI, strict length compliance with GPT-4.1 leaps from below 30% under naive prompts to above 95% with our countdown prompt, surpassing the popular draft-then-revise baseline, while judged answer quality is preserved. These results show that precise length control can be achieved through prompt engineering alone, offering a lightweight alternative to training- or decoding-based methods.

How do large language models understand moral dimensions compared to humans? This first large-scale Bayesian evaluation of market-leading language models provides the answer. In contrast to prior work using deterministic ground truth (majority or inclusion rules), we model annotator disagreements to capture both aleatoric uncertainty (inherent human disagreement) and epistemic uncertainty (model domain sensitivity). We evaluate top language models (Claude Sonnet 4, DeepSeek-V3, Llama 4 Maverick) across 250K+ annotations from ~700 annotators on 100K+ texts spanning social media, news, and forums. Our GPU-optimized Bayesian framework processed 1M+ model queries, revealing that AI models typically rank among the top 25\% of human annotators, achieving much better-than-average balanced accuracy. Importantly, we find that AI produces far fewer false negatives than humans, highlighting their more sensitive moral detection capabilities.

While document summarization with LLMs has enhanced access to textual information, concerns about the factual accuracy of these summaries persist, especially in the medical domain. Tracing evidence from which summaries are derived enables users to assess their accuracy, thereby alleviating this concern. In this paper, we introduce TracSum, a novel benchmark for traceable, aspect-based summarization, in which generated summaries are paired with sentence-level citations, enabling users to trace back to the original context. First, we annotate 500 medical abstracts for seven key medical aspects, yielding 3.5K summary-citation pairs. We then propose a fine-grained evaluation framework for this new task, designed to assess the completeness and consistency of generated content using four metrics. Finally, we introduce a summarization pipeline, Track-Then-Sum, which serves as a baseline method for comparison. In experiments, we evaluate both this baseline and a set of LLMs on TracSum, and conduct a human evaluation to assess the evaluation results. The findings demonstrate that TracSum can serve as an effective benchmark for traceable, aspect-based summarization tasks. We also observe that explicitly performing sentence-level tracking prior to summarization enhances generation accuracy, while incorporating the full context further improves completeness.

The role of large language models (LLMs) in education is increasing, yet little attention has been paid to whether LLM-generated text resembles child language. This study evaluates how LLMs replicate child-like language by comparing LLM-generated texts to a collection of German children's descriptions of picture stories. We generated two LLM-based corpora using the same picture stories and two prompt types: zero-shot and few-shot prompts specifying a general age from the children corpus. We conducted a comparative analysis across psycholinguistic text properties, including word frequency, lexical richness, sentence and word length, part-of-speech tags, and semantic similarity with word embeddings. The results show that LLM-generated texts are longer but less lexically rich, rely more on high-frequency words, and under-represent nouns. Semantic vector space analysis revealed low similarity, highlighting differences between the two corpora on the level of corpus semantics. Few-shot prompt increased similarities between children and LLM text to a minor extent, but still failed to replicate lexical and semantic patterns. The findings contribute to our understanding of how LLMs approximate child language through multimodal prompting (text + image) and give insights into their use in psycholinguistic research and education while raising important questions about the appropriateness of LLM-generated language in child-directed educational tools.

Large Language Models have shown growing ability to generate fluent and coherent texts that are highly similar to the writing style of humans. Current detectors for Machine-Generated Text (MGT) perform well when they are trained and tested in the same domain but generalize poorly to unseen domains, due to domain shift between data from different sources. In this work, we propose MGT-Prism, an MGT detection method from the perspective of the frequency domain for better domain generalization. Our key insight stems from analyzing text representations in the frequency domain, where we observe consistent spectral patterns across diverse domains, while significant discrepancies in magnitude emerge between MGT and human-written texts (HWTs). The observation initiates the design of a low frequency domain filtering module for filtering out the document-level features that are sensitive to domain shift, and a dynamic spectrum alignment strategy to extract the task-specific and domain-invariant features for improving the detector's performance in domain generalization. Extensive experiments demonstrate that MGT-Prism outperforms state-of-the-art baselines by an average of 0.90% in accuracy and 0.92% in F1 score on 11 test datasets across three domain-generalization scenarios.