Explanation-guided learning (EGL) has shown promise in aligning model predictions with interpretable reasoning, particularly in computer vision tasks. However, most approaches rely on external annotations or heuristic-based segmentation to supervise model explanations, which can be noisy, imprecise and difficult to scale. In this work, we provide both empirical and theoretical evidence that low-quality supervision signals can degrade model performance rather than improve it. In response, we propose ALIGN, a novel framework that jointly trains a classifier and a masker in an iterative manner. The masker learns to produce soft, task-relevant masks that highlight informative regions, while the classifier is optimized for both prediction accuracy and alignment between its saliency maps and the learned masks. By leveraging high-quality masks as guidance, ALIGN improves both interpretability and generalizability, showing its superiority across various settings. Experiments on the two domain generalization benchmarks, VLCS and Terra Incognita, show that ALIGN consistently outperforms six strong baselines in both in-distribution and out-of-distribution settings. Besides, ALIGN also yields superior explanation quality concerning sufficiency and comprehensiveness, highlighting its effectiveness in producing accurate and interpretable models.

Jinhui Chen, Shizhu He, Xingchang Yang, Huanxuan Liao, Yequan Wang, Xiangwen Liao, Wenhao Teng, Kang Liu, Jun Zhao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Huanxuan Liao, Shizhu He, Yupu Hao, Yequan Wang, Wenhao Teng, Xiangwen Liao, Jun Zhao, Kang Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Deep neural networks (DNNs) have achieved remarkable performance in many tasks, yet they often behave as opaque black boxes. Explanation-guided learning (EGL) methods steer DNNs using human-provided explanations or supervision on model attributions. These approaches improve interpretability but typically assume benign inputs and incur heavy annotation costs. In contrast, both predictions and saliency maps of DNNs could dramatically alter facing imperceptible perturbations or unseen patterns. Adversarial training (AT) can substantially improve robustness, but it does not guarantee that model decisions rely on semantically meaningful features. In response, we propose Explanation-Guided Adversarial Training (EGAT), a unified framework that integrates the strength of AT and EGL to simultaneously improve prediction performance, robustness, and explanation quality. EGAT generates adversarial examples on the fly while imposing explanation-based constraints on the model. By jointly optimizing classification performance, adversarial robustness, and attributional stability, EGAT is not only more resistant to unexpected cases, including adversarial attacks and out-of-distribution (OOD) scenarios, but also offer human-interpretable justifications for the decisions. We further formalize EGAT within the Probably Approximately Correct learning framework, demonstrating theoretically that it yields more stable predictions under unexpected situations compared to standard AT. Empirical evaluations on OOD benchmark datasets show that EGAT consistently outperforms competitive baselines in both clean accuracy and adversarial accuracy (+37%) while producing more semantically meaningful explanations, and requiring only a limited increase (+16%) in training time.

The microscopic cascade prediction task has wide applications in downstream areas like ''rumor detection''. Its goal is to forecast the diffusion routines of information cascade within networks. Existing works typically formulate it as a classification task, which fails to well align with the Social Homophily assumption, as it just use the features of ''infected'' users while neglecting those of ''uninfected'' users in representation learning. Moreover, these methods focus primarily on social relationships, thereby dismissing other vital dimensions like users' historical behavior and the underlying preferences behind it. To address these challenges, we introduce the MSR (Multifaceted Self-Retrieval) framework. During encoding, in addition to the existing social graph, we construct a preference graph to represent ''behavioral preferences'' and further propose a modified multi-channel GRAU for multi-view analysis of cascade phenomenon. For decoding, our approach diverges from classification-based methods by reformulating the task as an information retrieval problem that predicts the target user with similarity measures. Empirical evaluations on public datasets demonstrate that this framework significantly outperforms baselines on Hits@κ and MAP@κ, affirming its enhanced ability.

Fine-grained emotion recognition aims to identify the emotional type in queries through reasoning and decision-making processes, playing a crucial role in various systems. Recent methods use In-Context Learning (ICL), enhancing the representation of queries in the reasoning process through semantically similar examples, while further improving emotion recognition by explaining the reasoning mechanisms. However, these methods enhance the reasoning process but overlook the decision-making process. This paper investigates decision-making in fine-grained emotion recognition through prototype theory. We show that ICL relies on similarity matching between query representations and emotional prototypes within the model, where emotion-accurate representations are critical. However, semantically similar examples often introduce emotional discrepancies, hindering accurate representations and causing errors. To address this, we propose Emotion In-Context Learning (EICL), which introduces emotionally similar examples and uses a dynamic soft-label strategy to improve query representations in the emotion reasoning process. A two-stage exclusion strategy is then employed to assess similarity from multiple angles, further optimizing the decision-making process. Extensive experiments show that EICL significantly outperforms ICL on multiple datasets.

The increasing volume of radiology examinations has created an urgent need for automated report generation that is comparable to those written by radiologists. The major challenge is achieving precise semantic alignment between images and text, ensuring that generated reports accurately capture and describe the visual findings in medical images. Current approaches often struggle with this alignment, compromising diagnostic accuracy and clinical utility. To address these challenges, we present Adaptive Semantic Alignment Method (ASAM), a novel framework that enhances cross-modal semantic alignment through two innovations. First, we introduce a gated disease knowledge base by memory matrix that provides structured medical context to guide the mapping between image and report modalities. Second, we develop a cross-modal pre-trained visual encoder that enriches feature representation through improved understanding of medical imaging characteristics. Extensive experiments demonstrate that ASAM achieves state-of-the-art performance on two public chest X-ray datasets in BLEU-n metrics.

Zhou Yang, Zhaochun Ren, Wang Yufeng, Haizhou Sun, Chao Chen, Xiaofei Zhu, Xiangwen Liao. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Medical image segmentation is crucial for diagnosis and treatment planning. While recent advancements in deep learning, particularly UNet variants, have improved segmentation performance, they often result in increased model complexity, which limits their real-time applicability on resource-constrained devices in clinical settings. To address this challenge, we present the Tiny Efficient U-Net (TE-UNet), a novel lightweight model balancing efficiency and accuracy. TE-UNet uses a U-shaped encoder-decoder framework with a gated linear attention mechanism to process low-level and high-level features, preserving details and reducing complexity. It employs depth-wise separable convolutions for higher-level processing, enhancing efficiency without losing performance. Additionally, skip connections improve multi-scale feature extraction and information flow. Experiments on two public datasets across different modalities demonstrate that TE-UNet outperforms ten state-of-the-art methods, maintaining a parameter size under 40KB and low computational cost. TE-UNet makes real-time segmentation more accessible for various clinical applications.

Empathetic response generation endeavors to empower dialogue systems to perceive speakers’ emotions and generate empathetic responses accordingly. Psychological research demonstrates that emotion, as an essential factor in empathy, encompasses trait emotions, which are static and context-independent, and state emotions, which are dynamic and context-dependent. However, previous studies treat them in isolation, leading to insufficient emotional perception of the context, and subsequently, less effective empathetic expression. To address this problem, we propose Combining Trait and State emotions for Empathetic Response Model (CTSM). Specifically, to sufficiently perceive emotions in dialogue, we first construct and encode trait and state emotion embeddings, and then we further enhance emotional perception capability through an emotion guidance module that guides emotion representation. In addition, we propose a cross-contrastive learning decoder to enhance the model’s empathetic expression capability by aligning trait and state emotions between generated responses and contexts. Both automatic and manual evaluation results demonstrate that CTSM outperforms state-of-the-art baselines and can generate more empathetic responses. Our code is available at https://github.com/wangyufeng-empty/CTSM

To foster trust in machine learning models, explanations must be faithful and stable for consistent insights. Existing relevant works rely on the $\ell_p$ distance for stability assessment, which diverges from human perception. Besides, existing adversarial training (AT) associated with intensive computations may lead to an arms race. To address these challenges, we introduce a novel metric to assess the stability of top-$k$ salient features. We introduce R2ET which trains for stable explanation by efficient and effective regularizer, and analyze R2ET by multi-objective optimization to prove numerical and statistical stability of explanations. Moreover, theoretical connections between R2ET and certified robustness justify R2ET's stability in all attacks. Extensive experiments across various data modalities and model architectures show that R2ET achieves superior stability against stealthy attacks, and generalizes effectively across different explanation methods. The code can be found at https://github.com/ccha005/R2ET.

In recent years, online sentiment texts are generated by users in various domains and in different languages. Binary polarity classification (positive or negative) on business sentiment texts can help both companies and customers to evaluate products or services. Sometimes, the polarity of sentiment texts can be modified, making the polarity classification difficult. In sentiment analysis, such modification of polarity is termed as polarity shifting , which shifts the polarity of a sentiment clue (emotion, evaluation, etc.). It is well known that detection of polarity shifting can help improve sentiment analysis in texts. However, to detect polarity shifting in corpora is challenging: (1) polarity shifting is normally sparse in texts, making human annotation difficult; (2) corpora with dense polarity shifting are few; we may need polarity shifting patterns from various corpora. In this article, an approach is presented to extract polarity shifting patterns from any text corpus. For the first time, we proposed to select texts rich in polarity shifting by the idea of natural annotation , which is used to replace human annotation. With a sequence mining algorithm, the selected texts are used to generate polarity shifting pattern candidates, and then we rank them by C-value before human annotation. The approach is tested on different corpora and different languages. The results show that our approach can capture various types of polarity shifting patterns, and some patterns are unique to specific corpora. Therefore, for better performance, it is reasonable to construct polarity shifting patterns directly from the given corpus.

Delai Qiu, Yuanzhe Zhang, Xinwei Feng, Xiangwen Liao, Wenbin Jiang, Yajuan Lyu, Kang Liu, Jun Zhao. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

How to quantify influences between users, seeing that social network users influence each other in their temporal behaviors? Previous work has directly defined an independent model parameter to capture the interpersonal influence between each pair of users. To do so, these models need a parameter for each pair of users, which results in high-dimensional models becoming easily trapped into the overfitting problem. However, such models do not consider how influences depend on each other if influences are sent from the same user or if influences are received by the same user. Therefore, we propose a model that defines parameters for every user with a latent influence vector and a susceptibility vector, opposite to define influences on user pairs. Such low-dimensional representations naturally cause the interpersonal influences involving the same user to be coupled with each other, thus reducing the model’s complexity. Additionally, the model can easily consider the temporal information and sentimental polarities of users’ messages. Finally, we conduct extensive experiments on two real-world Microblog datasets, showing that our model with such representations achieves best performance on three prediction tasks, compared to the state-of-the-art and pair-wise baselines.

User influence is a very important factor for microblog user recommendation in mobile social network. However, most existing user influence analysis works ignore user’s temporal features and fail to filter the marketing users with low influence, which limits the performance of recommendation methods. In this paper, a Tensor Factorization based User Cluster (TFUC) model is proposed. We firstly identify latent influential users by neural network clustering. Then, we construct a features tensor according to latent influential user’s opinion, activity, and network centrality information. Furthermore, user influences are predicted by the latent factors resulting from the temporal restrained CP decomposition. Finally, we recommend microblog users considering both user influence and content similarity. Our experimental results show that the proposed model significantly improves recommendation performance. Meanwhile, the mean average precision of TFUC outperforms the baselines with 3.4% at least.

Whereas it is well known that social network users influence each other, a fundamental problem in influence maximization, opinion formation and viral marketing is that users' influences are difficult to quantify. Previous work has directly defined an independent model parameter to capture the interpersonal influence between each pair of users. However, such models do not consider how influences depend on each other if they originate from the same user or if they act on the same user. To do so, these models need a parameter for each pair of users, which results in high-dimensional models becoming easily trapped into the overfitting problem. Given these problems, another way of defining the parameters is needed to consider the dependencies. Thus we propose a model that defines parameters for every user with a latent influence vector and a susceptibility vector. Such low-dimensional and distributed representations naturally cause the interpersonal influences involving the same user to be coupled with each other, thus reducing the model's complexity. Additionally, the model can easily consider the sentimental polarities of users' messages and how sentiment affects users' influences. In this study, we conduct extensive experiments on real Microblog data, showing that our model with distributed representations achieves better accuracy than the state-of-the-art and pair-wise models, and that learning influences on sentiments benefit performance.

A single sign-on (SSO) system allow single authentication for multiple services. It is a potential solution to the implications of security, credentials management, et al. Recently, several works have used the threshold-based secret sharing scheme to create a distributed SSO service. All these works setup the threshold parameters first in the system initiation. But in some real-world applications, the threshold value should be dynamically changed in the authentication phase. In this paper, we present a novel threshold-based distributed single sign-on scheme with a dynamically changed threshold value(DctSSO). In DctSSO, two different degree secret polynomials are constructed. Each authentication server has two kinds of secret keys: keys for initiation shares and keys for authentication shares. Through the simply XOR operation, authentication shares keys can be delivered securely. DctSSO is not only as good as Threspassport on the aspects of security, portability, intrusion and fault tolerance, scalability, reliability, and availability, but also it offers two significant advantages over ThresPassport : it has the dynamically, securely and availably changed threshold value in the authentication phase, and it can prevent conspiracy-impersonation attacks.