Multimodal Large Language Models (MLLMs) have garnered significant attention for their strong visual-semantic understanding. Most existing chart benchmarks evaluate MLLMs' ability to parse information from charts to answer questions. However, they overlook the inherent output biases of MLLMs, where models rely on their parametric memory to answer questions rather than genuinely understanding the chart content. To address this limitation, we introduce a novel Chart Hypothetical Question Answering (HQA) task, which imposes assumptions on the same question to compel models to engage in counterfactual reasoning based on the chart content. Furthermore, we introduce HAI, a human-AI interactive data synthesis approach that leverages the efficient text-editing capabilities of LLMs alongside human expert knowledge to generate diverse and high-quality HQA data at a low cost. Using HAI, we construct Chart-HQA, a challenging benchmark synthesized from publicly available data sources. Evaluation results on 18 MLLMs of varying model sizes reveal that current models face significant generalization challenges and exhibit imbalanced reasoning performance on the HQA task.

Visual Relation Extraction (VRE) is a powerful means of discovering relationships between entities within visually-rich documents. Existing methods often focus on manipulating entity features to find pairwise relations, yet neglect the more fundamental structural information that links disparate entity pairs together. The absence of global structure information may make the model struggle to learn long-range relations and easily predict conflicted results. To alleviate such limitations, we propose a GlObal Structure knowledge-guided relation Extraction (GOSE) framework. GOSE initiates by generating preliminary relation predictions on entity pairs extracted from a scanned image of the document. Subsequently, global structural knowledge is captured from the preceding iterative predictions, which are then incorporated into the representations of the entities. This "generate-capture-incorporate" cycle is repeated multiple times, allowing entity representations and global structure knowledge to be mutually reinforced. Extensive experiments validate that GOSE not only outperforms existing methods in the standard fine-tuning setting but also reveals superior cross-lingual learning capabilities; indeed, even yields stronger data-efficient performance in the low-resource setting. The code for GOSE will be available at https://github.com/chenxn2020/GOSE.

Knowledge graph embedding (KGE) aims to map entities and relations of a knowledge graph (KG) into a low-dimensional and dense vector space via contrasting the positive and negative triples. In the training process of KGEs, negative sampling is essential to find high-quality negative triples since KGs only contain positive triples. Most existing negative sampling methods assume that non-existent triples with high scores are high-quality negative triples. However, negative triples sampled by these methods are likely to contain noise. Specifically, they ignore that non-existent triples with high scores might also be true facts due to the incompleteness of KGs, which are usually called false negative triples. To alleviate the above issue, we propose an easily pluggable denoising mixup method called DeMix, which generates high-quality triples by refining sampled negative triples in a self-supervised manner. Given a sampled unlabeled triple, DeMix firstly classifies it into a marginal pseudo-negative triple or a negative triple based on the judgment of the KGE model itself. Secondly, it selects an appropriate mixup partner for the current triple to synthesize a partially positive or a harder negative triple. Experimental results on the knowledge graph completion task show that the proposed DeMix is superior to other negative sampling techniques, ensuring corresponding KGEs a faster convergence and better link prediction results.

This paper presents our systems for the three Subtasks of SemEval Task4: Reading Comprehension of Abstract Meaning (ReCAM). We explain the algorithms used to learn our models and the process of tuning the algorithms and selecting the best model. Inspired by the similarity of the ReCAM task and the language pre-training, we propose a simple yet effective technology, namely, negative augmentation with language model. Evaluation results demonstrate the effectiveness of our proposed approach. Our models achieve the 4th rank on both official test sets of Subtask 1 and Subtask 2 with an accuracy of 87.9% and an accuracy of 92.8%, respectively. We further conduct comprehensive model analysis and observe interesting error cases, which may promote future researches.