Domain-specific question answering in low-resource languages faces two key challenges: scarcity of annotated datasets and limited domain knowledge in general-purpose language models. In this work, we present a multi-stage finetuning strategy to adapt lightweight language models to the Hindi tourism domain by leveraging both original and synthetic training data. Synthetic question-answer pairs are generated using large LLMs (LLaMA-70B, Phi-14B) and used to augment the limited original dataset. We explore several training methodologies and analyse their impact on domain generalisation. Our results demonstrate that large models can efficiently generate synthetic data, while small models can effectively adapt to it, offering a scalable pathway for low-resource, domain-specific QA.

Recent progress, such as DeepSeek-R1, has shown that the GRPO algorithm, a Reinforcement Learning (RL) approach, can effectively train Chain-of-Thought (CoT) reasoning in Large Language Models (LLMs) and Vision-Language Models (VLMs). In this paper, we analyze three challenges of GRPO: gradient coupling between thoughts and answers, sparse reward signals caused by limited parallel sampling, and unstable advantage estimation. To mitigate these challenges, we propose GRPO-MA, a simple yet theoretically grounded method that leverages multi-answer generation from each thought process, enabling more robust and efficient optimization. Theoretically, we show that the variance of thought advantage decreases as the number of answers per thought increases. Empirically, our gradient analysis confirms this effect, showing that GRPO-MA reduces gradient spikes compared to GRPO. Experiments on math, code, and diverse multimodal tasks demonstrate that GRPO-MA substantially improves performance and training efficiency. Our ablation studies further reveal that increasing the number of answers per thought consistently enhances model performance.

Document Visual Question Answering (DocVQA) models often produce overconfident or ethically misaligned responses, especially under uncertainty. Existing models like LayoutLMv3, UDOP, and DONUT focus on accuracy but lack ethical calibration. We propose HonestVQA, a model-agnostic, self-supervised framework that aligns model confidence with correctness using weighted loss and contrastive learning. We introduce two new metrics Honesty Score (H-Score) and Ethical Confidence Index (ECI)-to evaluate ethical alignment. HonestVQA improves accuracy and F1 by up to 4.3% across SpDocVQA, InfographicsVQA, and SROIE datasets, while reducing overconfidence. It also generalizes well across domains, achieving 78.9% accuracy and 76.1% F1-score.

Intent identification serves as the foundation for generating appropriate responses in personalized question answering (PQA). However, existing benchmarks evaluate only response quality or retrieval performance without directly measuring intent identification capabilities. This gap is critical because without understanding which intents users prioritize, systems cannot generate responses satisfying individual information needs. To address this, we introduce the concept of core intents: intents users prioritize when selecting answers to satisfy their information needs. To evaluate these core intents, we propose IPQA, a benchmark for core Intent identification in Personalized Question Answering. Since users do not explicitly state their prioritized intents, we derive core intents from observable behavior patterns in answer selection, grounded in satisficing theory where users choose answers meeting their acceptance thresholds. We construct a dataset with various domains through systematic filtering, LLM-based annotation, and rigorous quality control combining automated verification with human validation. Experimental evaluations across state-of-the-art language models reveal that current systems struggle with core intent identification in personalized contexts. Models fail to identify core intents from user histories, with performance degrading as question complexity increases. The code and dataset will be made publicly available to facilitate future research in this direction.

Figure 1: Overview of the proposed ToolEQA for Embodied Question Answering (EQA). ToolEQA enables to decompose questions into structured plans, reasoning to select tools, and invoke tools to explore and answer. ToolEQA achieves highest accuracy with fewer reasoning steps. Embodied Question Answering (EQA) requires agents to explore 3D environments to obtain observations and answer questions related to the scene. Existing methods leverage VLMs to directly explore the environment and answer questions without explicit thinking or planning, which limits their reasoning ability and results in excessive or inefficient exploration as well as ineffective responses. In this paper, we introduce T oolEQA, an agent that integrates external tools with multi-step reasoning, where external tools can provide more useful information for completing the task, helping the model derive better exploration directions in the next step of reasoning and thus obtaining additional effective information. This enables ToolEQA to generate more accurate responses with a shorter exploration distance. To enhance the model's ability for tool-usage and multi-step reasoning, we further design a novel EQA data generation pipeline that automatically Based on the pipeline, we collect the EQA-RT dataset that contains about 18K tasks, divided into a training set EQA-RT -Train, and two test sets EQA-RT -Seen (scenes overlapping with the training set) and EQA-RT -Unseen (novel scenes). Experiments on EQA-RT -Seen and EQA-RT -Unseen show that ToolEQA improves the success rate by 9.2 20.2% over state-of-the-art baselines, while outperforming the zero-shot ToolEQA by 10% in success rate.

We introduce and study an online problem arising in question answering systems. In this problem, an agent must sequentially classify user-submitted queries represented by $d$-dimensional embeddings drawn i.i.d. from an unknown distribution. The agent may consult a costly human expert for the correct label, or guess on her own without receiving feedback. The goal is to minimize regret against an oracle with free expert access. When the time horizon $T$ is at least exponential in the embedding dimension $d$, one can learn the geometry of the class regions: in this regime, we propose the Conservative Hull-based Classifier (CHC), which maintains convex hulls of expert-labeled queries and calls the expert as soon as a query lands outside all known hulls. CHC attains $\mathcal{O}(\log^d T)$ regret in $T$ and is minimax optimal for $d=1$. Otherwise, the geometry cannot be reliably learned without additional distributional assumptions. We show that when the queries are drawn from a subgaussian mixture, for $T \le e^d$, a Center-based Classifier (CC) achieves regret proportional to $N\log{N}$ where $N$ is the number of labels. To bridge these regimes, we introduce the Generalized Hull-based Classifier (GHC), a practical extension of CHC that allows for more aggressive guessing via a tunable threshold parameter. Our approach is validated with experiments, notably on real-world question-answering datasets using embeddings derived from state-of-the-art large language models.

Graph Retrieval-Augmented Generation (Graph RAG) effectively builds a knowledge graph (KG) to connect disparate facts across a large document corpus. However, this broad-view approach often lacks the deep structured reasoning needed for complex multi-hop question answering (QA), leading to incomplete evidence and error accumulation. To address these limitations, we propose SubQRAG, a sub-question-driven framework that enhances reasoning depth. SubQRAG decomposes a complex question into an ordered chain of verifiable sub-questions. For each sub-question, it retrieves relevant triples from the graph. When the existing graph is insufficient, the system dynamically expands it by extracting new triples from source documents in real time. All triples used in the reasoning process are aggregated into a "graph memory," forming a structured and traceable evidence path for final answer generation. Experiments on three multi-hop QA benchmarks demonstrate that SubQRAG achieves consistent and significant improvements, especially in Exact Match scores.

This paper presents the VLSP 2025 MLQA-TSR - the multimodal legal question answering on traffic sign regulation shared task at VLSP 2025. VLSP 2025 MLQA-TSR comprises two subtasks: multimodal legal retrieval and multimodal question answering. The goal is to advance research on Vietnamese multimodal legal text processing and to provide a benchmark dataset for building and evaluating intelligent systems in multimodal legal domains, with a focus on traffic sign regulation in Vietnam. The best-reported results on VLSP 2025 MLQA-TSR are an F2 score of 64.55% for multimodal legal retrieval and an accuracy of 86.30% for multimodal question answering.

Question Answering (QA) systems have traditionally relied on structured text data, but the rapid growth of multimedia content (images, audio, video, and structured metadata) has introduced new challenges and opportunities for retrieval-augmented QA. In this survey, we review recent advancements in QA systems that integrate multimedia retrieval pipelines, focusing on architectures that align vision, language, and audio modalities with user queries. We categorize approaches based on retrieval methods, fusion techniques, and answer generation strategies, and analyze benchmark datasets, evaluation protocols, and performance tradeoffs. Furthermore, we highlight key challenges such as cross-modal alignment, latency-accuracy tradeoffs, and semantic grounding, and outline open problems and future research directions for building more robust and context-aware QA systems leveraging multimedia data.

Although recent advancements in diffusion models have significantly enriched the quality of generated images, challenges remain in synthesizing pixel-based human-drawn sketches, a representative example of abstract expression. To combat these challenges, we propose StableSketcher, a novel framework that empowers diffusion models to generate hand-drawn sketches with high prompt fidelity. Within this framework, we fine-tune the variational autoencoder to optimize latent decoding, enabling it to better capture the characteristics of sketches. In parallel, we integrate a new reward function for reinforcement learning based on visual question answering, which improves text-image alignment and semantic consistency. Extensive experiments demonstrate that StableSketcher generates sketches with improved stylistic fidelity, achieving better alignment with prompts compared to the Stable Diffusion baseline. Additionally, we introduce SketchDUO, to the best of our knowledge, the first dataset comprising instance-level sketches paired with captions and question-answer pairs, thereby addressing the limitations of existing datasets that rely on image-label pairs. Our code and dataset will be made publicly available upon acceptance.