Multimodal knowledge graph completion enriches link prediction by combining structured triples with auxiliary modalities such as text and images, but existing solutions often rely on relation-insensitive fusion and are brittle under instance-level modality noise, leading to weak alignment between entity evidence and relational semantics. We propose MMFusion, a unified reasoning framework with three key designs. First, Relation-Guided Entity Transformation (RGET) constructs relation-conditioned entity views within each modality to enhance entity–relation compatibility. Second, Adaptive Modality Gating (AMG) performs instance-level gating across structural, visual, and textual signals to suppress noisy modalities and improve fusion robustness. Finally, Query-Guided Cross-modal Fusion (QCF) builds a unified relation-conditioned query via dual-stream attention between entity and relation modalities, enabling deep cross-modal alignment for link prediction. Extensive experiments on four public multimodal knowledge graph datasets show that MMFusion consistently outperforms strong baselines, especially in sparse and noisy settings.

Biomedical document-level relation extraction plays a crucial role in mining structured knowledge from biomedical texts. However, existing large language model-based methods struggle with information redundancy, noise interference, and insufficient domain-specific knowledge when processing complex documents. To address these challenges, we propose Bio-$\mathbf{R}^{\mathbf{3}}$, a novel framework that enhances document-level relation extraction through entity aware rewriting and domain-adaptive reasoning. Our Bio-R${ }^{3}$follows a “Rewrite-Retrieval-Reason” pipeline: (1) Entity aware rewriting refocuses the document content on target entity pairs, reducing irrelevant context; (2) Retrieval-augmented generation supplements the document with external biomedical evidence to support relation inference; and (3) Chain-of-Thought prompting and few-shot learning improve reasoning accuracy, while Low-Rank Adaptation fine-tuning enhances domain adaptation. Experiments on the CDR and GDA benchmarks demonstrate that Bio-$\mathbf{R}^{\mathbf{3}}$significantly outperforms existing methods, demonstrating its effectiveness in the extraction of biomedical document relations.

With the rapid development of the internet, the news domain is flooded with an increasing amount of multi-modal fake news, which poses a great threat to today’s society. Although some methods can automatically detect fake news recently, the lack of explainability remains a challenge. This paper first pre-defines three discriminative patterns: image tampering, text forgery, and inconsistency between image and text. Revealing these three patterns not only lead to better prediction results but also provide clear and concise explanations. Therefore, we propose Multi-modal Explainability Analysis Network (MEAN) for fake news detection. This model first extracts discriminative patterns using both uni-modal and multi-modal methods, then employs a multi-branch network to obtain fine-grained representations for these patterns, and finally learns the discriminative patterns through logical rules to obtain the predicted labels of news and the explainability analysis of the model. Extensive experiments on two real-world datasets demonstrate the superiority of the proposed MEAN model in detecting fake news while providing explanations for the discriminative patterns.

MOTIVATION: With the advancement of large language models (LLMs), the field of biomedical document-level relation extraction (BioDocRE) has encountered new opportunities. However, LLMs often face challenges such as hallucinated generation, insufficient reasoning capabilities, and a lack of interpretability when performing relation extraction tasks. RESULTS: To address these issues, we propose the SyRACT (Synergistic Retrieval Augmented Generation and Chain of Thought) framework for high precision relation extraction in biomedical documents. This framework is built around three core strategies: (i) reframing the relation extraction task as a question answering problem to better align with the processing logic of LLMs; (ii) leveraging an external database constructed from PubMed to provide LLMs with rich and reliable contextual information, thus mitigating hallucination generation; and (iii) construct a specific Chain of Thought for BioDocRE tasks, thereby enhancing the model's reasoning ability and the interpretability of its output. We validated this approach on three biomedical relation extraction datasets: CDR, GDA, and ADE. Experimental results show that the SyRACT model improves F1 scores by 11.04%, 9.10%, and 41.00% on three datasets, respectively, compared to the DocRE method, which uses standard prompts for LLMs. AVAILABILITY AND IMPLEMENTATION: Our source code and data are available at https://github.com/donggggxin/SyRACT.

Biomedical Named Entity Recognition (NER) serves as a crucial task in biomedical information extraction, aiming to identify and classify entities from unstructured biomedical texts. However, obtaining high-quality annotated biomedical texts is scarce due to their high privacy and the specialized expertise required. Therefore, more researchers are focusing on few-shot NER in the biomedical field. Recent methods mainly fall into three categories: 1) transferring knowledge from high-resource data and fine-tuning with low-resource data, 2) modifying data using various methods to generate new samples, and 3) decompose the NER task into two subtasks with multi-task learning. However, these methods either suffer from domain shift, generate low-quality synthetic data, or encounter error propagation issues. To address these limitations, we have investigated an few-shot NER method, which firstly employs specific prompt templates to guide Large Language Models in generating high-quality new samples. Then, we propose a novel more fine-grained grid-tagging strategy and single-end sensitive multi-task learning framework. This method sets multiple losses to enable the model to learn from multiple task objectives: what constitutes a fully correct, partially correct, and incorrect entity. Additionally, through the new grid-tagging strategy, the model can decode entities based on more tagging clues. To better align with real-world scenarios, we trained in a few-shot scenario and evaluated on the full test set. Extensive experimental results on the NCBI, BC5CDR, BioNLP11EPI, and BioNLP13GE datasets confirm that our method outperforms previous state-of-the-art methods in most scenarios.1

Style transfer has been proven to be effective in mitigating domain shift in clinical settings, enhancing the adaptability of pathology image models. However, existing methods assume that the texture of the entire image is domain-specific and irrelevant to class-specific representations. These methods enhance all regions with a single style, which can result in information loss. In this work, we propose CLAP (Cross-modality Local Augmentation for histoPathology), a data augmentation approach that enables cross-modality local manipulation for pathology images. Specifically, the combination of a text extractor network and a feature mapping network enables the integration of CLIP embeddings for style descriptions into editable latent spaces at a fine-grained level. This approach prevents the loss of regional information in whole-slide images, eliminates the need to painstakingly select directions in latent space, and enhances creativity style selection. Experimental results demonstrate that CLAP improves the regional accuracy of cross-modality editing and achieves state-of-the-art performance by enhancing generalization in histopathology classification tasks.

The difficulty of cross-domain text sentiment classification is that the data distributions in the source domain and the target domain are inconsistent. This paper proposes an attention network based on feature sequences (ANFS) for cross-domain sentiment classification, which focuses on important semantic features by using the attention mechanism. Particularly, ANFS uses a three-layer convolutional neural network (CNN) to perform deep feature extraction on the text, and then uses a bidirectional long short-term memory (BiLSTM) to capture the long-term dependency relationship among the text feature sequences. We first transfer the ANFS model trained on the source domain to the target domain and share the parameters of the convolutional layer; then we use a small amount of labeled target domain data to fine-tune the model of the BiLSTM layer and the attention layer. The experimental results on cross-domain sentiment analysis tasks demonstrate that ANFS can significantly outperform the state-of-the-art methods for cross-domain sentiment classification problems.

Sequence transfer learning is of interest in both academia and industry with the emergence of numerous new text domains from Twitter and other social media tools. In this paper, we put forward the data-sensitive granularity for transfer learning, and then, a novel substructural regularization transfer learning model (STLM) is proposed to preserve target domain features at substructural granularity in the light of the condition of labeled data set size. Our model is underpinned by hidden Markov model and regularization theory, where the substructural representation can be integrated as a penalty after measuring the dissimilarity of substructures between target domain and STLM with relative entropy. STLM can achieve the competing goals of preserving the target domain substructure and utilizing the observations from both the target and source domains simultaneously. The estimation of STLM is very efficient since an analytical solution can be derived as a necessary and sufficient condition. The relative usability of substructures to act as regularization parameters and the time complexity of STLM are also analyzed and discussed. Comprehensive experiments of part-of-speech tagging with both Brown and Twitter corpora fully justify that our model can make improvements on all the combinations of source and target domains.