This paper introduces a system based on fine-tuned pretrained language models, which is constructed for SemEval 2026 Task 9: Multilingual Polarization Type Classification. The task aims to perform multi-label polarization classification on texts covering 22 languages, identifying five types of polarization: political, racial/ethnic, religious, gender/sexual, and others. The main challenges of the task lie in handling uneven data distribution across languages, extreme class imbalance, and the complexity of cross-lingual semantic understanding. To address these challenges, a training framework integrating hybrid augmentation and multi-strategy regularization is proposed. Based on XLM-RoBERTa-large, the framework combines feature-space Mixup augmentation, an asymmetric loss function, adversarial training, and exponential moving average. Multi-label decisions are made through dynamic threshold optimization. Experimental results show that the proposed method achieves a macro-F1 score of 0.48 on the validation set, effectively improving classification performance and generalization capability in multilingual and imbalanced scenarios.

This paper describes the system developed by the YNU-HPCC team for SemEval-2026 Task 1 (Humor Generation). The task aims to generate humorous texts from given news headlines or from two unrelated words. The core challenge lies in enabling Large Language Models (LLMs) to understand human humor and align with specific humorous styles. We investigated two approaches: fine-tuning with Proximal Policy Optimization (PPO) and in-context learning with LLMs. We also employed Qwen-Max to evaluate the quality of the generated texts. In the PPO experiments, we constructed a hybrid reward model to align with humor. For our final submission based on LLMs, we used multiple advanced LLMs, along with customized few-shot prompts and a small set of gold samples, to effectively guide the models in generating jokes that resonate with human humor. Experimental results show that our system achieves competitive performance, ranking 4th in the English track, 2nd in the Chinese track, and 2nd in the Spanish track.

This paper presents the system developed by the YNU-HPCC team for SemEval-2026 Task 4: Narrative Story Similarity and Narrative Representation Learning. The task challenges computational systems to identify narrative similarity across three orthogonal dimensions: abstract theme, course of action, and outcomes. The primary scientific difficulty lies in distinguishing the underlying structural fabula from surface-level lexical overlaps, particularly when facing long-context narratives with subtle plot twists. To address this, our approach employs a hybrid architecture that strategically decouples retrieval and ranking tasks. For Track A, we introduce a dynamic routing mechanism where an instruction-tuned E5-Mistral-7B model handles clear cases, while ambiguous hard samples are routed to a Gemini-3-Flash reasoner. For Track B, we leverage the global semantic modeling capabilities of Gemini-Embedding-001 via a structure-preserving chunking strategy, enhanced by All-But-The-Top (ABTT) during inference. Extensive experiments on the official test set show that this divide-and-conquer strategy effectively balances local instruction following with global open-domain generalization. Our system performs competitively, ranking 5th in Track A and 2nd in Track B among all participating teams.

We present the SemEval-2026 shared task on Dimensional Aspect-Based Sentiment Analysis (DimABSA), which improves traditional ABSA by modeling sentiment along valence–arousal (VA) dimensions rather than using categorical polarity labels. To extend ABSA beyond consumer reviews to public-issue discourse (e.g., political, energy, and climate issues), we introduce an additional task, Dimensional Stance Analysis (DimStance), which treats stance targets as aspects and reformulates stance detection as regression in the VA space. The task consists of two tracks: Track A (DimABSA) and Track B (DimStance). Track A includes three subtasks: (1) dimensional aspect sentiment regression, (2) dimensional aspect sentiment triplet extraction, and (3) dimensional aspect sentiment quadruplet extraction, while Track B includes only the regression subtask for stance targets. We also introduce a continuous F1 (cF1) metric to jointly evaluate structured extraction and VA regression.The task attracted more than 400 participants, resulting in 112 final submissions and 42 system description papers. We report baseline results, discuss top-performing systems, and analyze key design choices to provide insights into dimensional sentiment analysis at the aspect and stance-target levels. All resources are available on our GitHub repository.

In political interviews, politicians frequently employ evasion strategies to avoid direct answers, making it challenging to evaluate response clarity in Natural Language Processing. This paper presents the YNU-HPCC system for SemEval-2026 task 6: clarity classification in political interviews. To address the limitation where traditional models capture only surface-level semantics, this paper proposes two reasoning-enhanced frameworks. First, we introduce Hierarchical Taxonomy Prompting. This method guides LLMs to follow a strict top-down classification logic. Specifically, the model determines the clarity level before identifying specific evasion techniques. Furthermore, it explicitly articulates the reasoning process. Second, to balance reasoning capability with resource constraints, we employ Chain-of-Thought Distillation. We utilize DeepSeek V3.1 as a teacher model to generate comprehensive reasoning chains, which are then used to SFT the smaller student models. Experimental results demonstrate the effectiveness of our approach: The system achieved 6th place in Task 1 and 5th place in Task 2 among all participating teams, highlighting the importance of reasoning processes in detecting complex linguistic evasion.

This paper addresses continuous affect modeling in SemEval-2026 Task 2 through two task-specific architectures tailored to static state estimation and dynamic change prediction. To mitigate semantic ambiguity and annotation subjectivity in Subtask 1, a hard-prompt-based regression model is developed and enhanced with unsupervised contrastive learning (SimCSE) and supervised contrastive calibration (SCL) grounded in an external affect lexicon. This design improves the structural consistency and scale stability of textual representations in the Valence–Arousal (V/A) space. For Subtask 2a, which involves irregular time intervals and historical dependencies, a Time-Aware LSTM architecture is introduced to integrate current affective states with temporally enriched historical trajectories. Experimental results show that the YNU-HPCC system ranks 2nd in both subtasks. In Subtask 1, the Valence and Arousal scores are 0.677 and 0.528, respectively; in Subtask 2a, they are 0.692 and 0.647.

This paper addresses SemEval-2026 Task 11, which focused on mitigating content effects in syllogistic reasoning. Logical validity is often conflated with semantic plausibility in large language models.Prior methods rely on standard fine-tuning or prompting, without explicit bias control.A rule- and template-based symbolic data augmentation framework is proposed for fine-tuning the \texttt{Qwen2-1.5B-Instruct} model and instruction-tuning the \texttt{XLM-RoBERTa-large} model. Logic-preserving synthetic data are generated through lexical rules. The system is ranked 1st in Task 1 with a perfect overall score of 100, and 6th in Task 3 with a score of 56.97. Code is publicly available at: \url{https://github.com/YNU-HPCC/semeval-2026-task11}.

This paper describes the YNU-HPCC system for SemEval-2026 Task 12, Abductive EventReasoning (AER). Given multi-document retrieved evidence with distractors, the task requires selecting all direct-cause options for a target event and outputting an answer set. The main challenges are sparse and dispersed evidence in long documents and a boundary-sensitive set-level evaluation. This paper proposes a two-stage framework. Stage 1 trains a DeBERTa-v3-base student with retrieval-guided evidence modeling: documents are split into overlapping windows, BM25 ranks and filters candidate windows, and Top-K pooling aggregates window-level scores into option probabilities. Stage 2 distills soft targets from a Qwen-14B teacher with temperature scaling and high-confidence filtering to reduce pseudo-label noise and improve generalization. The system achieves an official dev score of 0.9712(micro-F1 0.9746, macro-F1 0.9745) and improves the test score from 0.46 to 0.73, ranking 84th out of 221 submissions.

As Large Language Models (LLMs) become prevalent in software development, distinguishing machine-generated from human-written code is increasingly important. This paper describes the system developed by the YNU-HPCC team for SemEval-2026 Task 13, which evaluates detection under cross-language, multi-generator, and hybrid settings. Three modeling paradigms are systematically examined: encoder-based fine-tuning, feature-based machine learning, and task-specific robustness strategies. For Subtask A (Binary Detection), frozen pre-trained encoders and shallow stylometric features exhibit stronger cross-domain robustness than full fine-tuning, with indentation entropy identified as a key discriminative signal. For Subtask B (Multi-Class Attribution), a hierarchical two-stage framework is adopted to decouple human–machine discrimination from fine-grained generator attribution, alleviating severe class imbalance. For Subtask C (Hybrid Detection), a token-level splicing augmentation strategy combined with Supervised Contrastive Learning and Focal Loss is employed to model intra-sample stylistic variation. According to the official leaderboard, our system ranked 12th out of 81 teams in Subtask A, 14th out of 34 in Subtask B, and 8th out of 32 in Subtask C.

This paper introduces our approach to SemEval 2026 Task 5, which evaluates the rationality of word-sense scores in ambiguous stories through narrative comprehension. This task requires models to assess the consistency between a given word-sense definition and the meaning of an ambiguous target word in a short narrative context, and to infer a rationality score on a 1-5 scale. We experimented and compared multiple methods. These methods include multi-head ensembles that simulate the behavior of individual annotators, ordinal classification and regression methods that treat scores as ordered categories, and direct regression using mean squared error (MSE) or L1 loss to predict human-average consensus scores. Additionally, we investigated instructional fine-tuning with low-rank adaptation (LoRA) on large language models (LLMs) such as Qwen3-4B-Instruct and Phi-4-mini. Our experimental results show that the direct MSE regression method performs best. This study indicates that directly optimizing to approach human consensus scores is effective for this task, while methods that model individual annotator differences are less applicable.

Human visual reasoning typically follows a coarse-to-fine attention process, starting from global scene understanding and gradually focusing on question-relevant regions. However, multimodal large language models may deviate from this pattern due to attention drift and the underutilization of visual evidence, which can lead to hallucinations. To mitigate these issues, this study proposes a Dual-Indicator Guided Contrastive Alignment (DICA), which tracks two information-theoretic indicators during inference: Visual Attention Entropy (VAE), which reflects the concentration of visual attention, and Output Image Correlation (OIC), which measures the dependence of generated outputs on the visual input. An abnormal increase in VAE or a decrease in OIC corresponds to different failure modes, which trigger targeted contrastive alignment to restore visual grounding. Experimental results across multiple benchmarks demonstrate that DICA consistently outperforms existing approaches and substantially reduces hallucinations, highlighting the effectiveness of indicator-driven intervention in improving multimodal inference reliability. The code is publicly available at https://github.com/BGWH123/DICA/.

Aspect-Based Sentiment Analysis (ABSA) focuses on extracting sentiment at a fine-grained aspect level and has been widely applied across real-world domains. However, existing ABSA research relies on coarse-grained categorical labels (e.g., positive, negative), which limits its ability to capture nuanced affective states. To address this limitation, we adopt a dimensional approach that represents sentiment with continuous valence–arousal (VA) scores, enabling fine-grained analysis at both the aspect and sentiment levels. To this end, we introduce DimABSA, the first multilingual, dimensional ABSA resource annotated with both traditional ABSA elements (aspect terms, aspect categories, and opinion terms) and newly introduced VA scores. This resource contains 76,958 aspect instances across 42,590 sentences, spanning six languages and four domains. We further introduce three subtasks that combine VA scores with different ABSA elements, providing a bridge from traditional ABSA to dimensional ABSA. Given that these subtasks involve both categorical and continuous outputs, we propose a new unified metric, continuous F1 (cF1), which incorporates VA prediction error into standard F1. We provide a comprehensive benchmark using both prompted and fine-tuned large language models across all subtasks. Our results show that DimABSA is a challenging benchmark and provides a foundation for advancing multilingual dimensional ABSA. We publicly released the DimABSA dataset, which was used for Track A of SemEval-2026 Task 3, attracting over 300 participants.

Chain-of-Thought (CoT) has become a vital technique for enhancing the performance of Large Language Models (LLMs), attracting increasing attention from researchers. One stream of approaches focuses on the iterative enhancement of LLMs by continuously verifying and refining their reasoning outputs for desired quality. Despite its impressive results, this paradigm faces two critical issues: (1) Simple verification methods: The current paradigm relies solely on a single verification method. (2) Wrong Information Ignorance: Traditional paradigms directly ignore wrong information during reasoning and refine the logic paths from scratch each time. To address these challenges, we propose Wrong-of-Thought (WoT), which includes two core modules: (1) Multi-Perspective Verification: A multi-perspective verification method for accurately refining the reasoning process and result, and (2) Wrong Information Utilization: Utilizing wrong information to alert LLMs and reduce the probability of LLMs making same mistakes. Experiments on 8 popular datasets and 5 LLMs demonstrate that WoT surpasses all previous baselines. In addition, WoT exhibits powerful capabilities in difficult computation tasks.

Aspect-based sentiment analysis (ABSA) identifies sentiment information related to specific aspects and provides deeper market insights to businesses and organizations. With the emergence of large language models (LMs), recent studies have proposed using fixed examples for instruction tuning to reformulate ABSA as a generation task. However, the performance is sensitive to the selection of in-context examples; several retrieval methods are based on surface similarity and are independent of the LM generative objective. This study proposes an instruction learning method with retrieval-based example ranking for ABSA tasks. For each target sample, an LM was applied as a scorer to estimate the likelihood of the output given the input and a candidate example as the prompt, and training examples were labeled as positive or negative by ranking the scores. An alternating training schema is proposed to train both the retriever and LM. Instructional prompts can be constructed using high-quality examples. The LM is used for both scoring and inference, improving the generation efficiency without incurring additional computational costs or training difficulties. Extensive experiments on three ABSA subtasks verified the effectiveness of the proposed method, demonstrating its superiority over various strong baseline models. Code and data are released at https://github.com/zgMin/IT-RER-ABSA.

Applying neural models injected with in-domain user and product information to learn review representations of unseen or anonymous users incurs an obvious obstacle in content-based recommender systems. For the generalization of the in-domain classifier, most existing models train an extra plain-text model for the unseen domain. Without incorporating historical user and product information, such a schema makes unseen and anonymous users dissociate from the recommender system. To simultaneously learn the review representation of both existing and unseen users, this study proposed a switch knowledge distillation for domain generalization. A generalization-switch (GSwitch) model was initially applied to inject user and product information by flexibly encoding both domain-invariant and domain-specific features. By turning the status ON or OFF, the model introduced a switch knowledge distillation to learn a robust review representation that performed well for either existing or anonymous unseen users. The empirical experiments were conducted on IMDB, Yelp-2013, and Yelp-2014 by masking out users in test data as unseen and anonymous users. The comparative results indicate that the proposed method enhances the generalization capability of several existing baseline models. For reproducibility, the code for this paper is available at: https://github.com/yoyo-yun/DG_RRR.

This paper describes a system built in the SemEval-2022 competition. As participants in Task 4: Patronizing and Condescending Language Detection, we implemented the text sentiment classification system for two subtasks in English. Both subtasks involve determining emotions; subtask 1 requires us to determine whether the text belongs to the PCL category (single-label classification), and subtask 2 requires us to determine to which PCL category the text belongs (multi-label classification). Our system is based on the bidirectional encoder representations from transformers (BERT) model. For the single-label classification, our system applies a BertForSequenceClassification model to classify the input text. For the multi-label classification, we use the fine-tuned BERT model to extract the sentiment score of the text and a fully connected layer to classify the text into the PCL categories. Our system achieved relatively good results on the competition’s official leaderboard.

This paper will present the methods we use as the YNU-HPCC team in the SemEval-2022 Task 2, Multilingual Idiomaticity Detection and Sentence Embedding. We are involved in two subtasks, including four settings. In subtask B of sentence representation, we used novel approaches with ideas of contrastive learning to optimize model, where method of CoSENT was used in the pre-train setting, and triplet loss and multiple negatives ranking loss functions in fine-tune setting. We had achieved very competitive results on the final released test datasets. However, for subtask A of idiomaticity detection, we simply did a few explorations and experiments based on the xlm-RoBERTa model. Sentence concatenated with additional MWE as inputs did well in a one-shot setting. Sentences containing context had a poor performance on final released test data in zero-shot setting even if we attempted to extract effective information from CLS tokens of hidden layers.

This paper describes the system submitted by our team (YNU-HPCC) to SemEval-2022 Task 8: Multilingual news article similarity. This task requires participants to develop a system which could evaluate the similarity between multilingual news article pairs. We propose an approach that relies on Transformers to compute the similarity between pairs of news. We tried different models namely BERT, ALBERT, ELECTRA, RoBERTa, M-BERT and Compared their results. At last, we chose M-BERT as our System, which has achieved the best Pearson Correlation Coefficient score of 0.738.

In this paper, we (a YNU-HPCC team) describe the system we built in the SemEval-2022 competition. As participants in Task 6 (titled “iSarcasmEval: Intended Sarcasm Detection In English and Arabic”), we implement the sentiment system for all three subtasks in English and Arabic. All subtasks involve the detection of sarcasm (binary and multilabel classification) and the determination of the sarcastic text location (sentence pair classification). Our system primarily applies the sequence classification model of a bidirectional encoder representation from a transformer (BERT). The BERT is used to extract sentence information from both directions for downstream classification tasks. A single basic model is used for single-sentence and sentence-pair binary classification tasks. For the multilabel task, the Label-Powerset method and binary cross-entropy loss function with weights are used. Our system exhibits competitive performance, obtaining 12/43 (21/32), 11/22, and 3/16 (8/13) rankings in the three official rankings for English (Arabic).

This paper describes our system used in the SemEval-2022 Task5 Multimedia Automatic Misogyny Identification (MAMI). This task is to use the provided text-image pairs to classify emotions. In this paper, We propose a multi-label emotion classification model based on pre-trained LXMERT. We use Faster-RCNN to extract visual representation and utilize LXMERT’s cross-attention for multi-modal alignment. Then we use the Bilinear-interaction layer to fuse these features. Our experimental results surpass the F₁ score of baseline. For Sub-task A, our F₁ score is 0.662 and Sub-task B’s F₁ score is 0.633. The code of this study is available on GitHub.