Data contamination poses a significant challenge to the fairness of LLM evaluations in natural language processing tasks by inadvertently exposing models to test data during training.Current studies mitigate this issue by modifying existing datasets or generating new ones from freshly collected information. However, these methods fall short of ensuring contamination-resilient evaluation, as they fail to fully eliminate pre-existing knowledge from models or preserve the semantic complexity of the original datasets. To address these limitations, we propose CoreEval, a Contamination-resilient Evaluation strategy for automatically updating data with real-world knowledge. This approach begins by extracting entity relationships from the original data and leveraging the GDELT database to retrieve relevant and up-to-date knowledge. The retrieved knowledge is then recontextualized and integrated with the original data, which is refined and restructured to ensure semantic coherence and enhanced task relevance. Ultimately, a robust data reflection mechanism in a Chain-of-Thought manner to iteratively verify and refine labels, ensuring consistency between the updated and original datasets. Extensive experiments on updated datasets validate the robustness of CoreEval, demonstrating its effectiveness in mitigating performance overestimation caused by data contamination.

Entity linking aims to identify mentions from the text and link them to a knowledge base. Further, Multi-lingual Entity Linking (MEL) is a more challenging task, where the language-specific mentions need to be linked to a multi-lingual knowledge base. To tackle the MEL task, we propose a novel model that employs the merit of adversarial learning and few-shot learning to generalize the learning ability across languages. Specifically, we first randomly select a fraction of language-agnostic unlabeled data as the language signal to construct the language discriminator. Based on it, we devise a simple and effective adversarial learning framework with two characteristic branches, including an entity classifier and a language discriminator with adversarial training. Experimental results on two benchmark datasets indicate the excellent performance in few-shot learning and the effectiveness of the proposed adversarial learning framework.

Conversational question answering aims to respond to questions based on relevant contexts and previous question-answer history. Existing studies typically use ground-truth answers in history, leading to the inconsistency between the training and inference phases. However, in real-world scenarios, progress in question answering can only be made using predicted answers. Since not all predicted answers are correct, indiscriminately using all predicted answers for training introduces noise into the model. To tackle these challenges, we propose an automatic answer correctness evaluation method named **Auto-ACE**. Specifically, we first construct an Att-BERT model which employs attention weight to the BERT model, so as to bridge the relation between the current question and the question-answer pair in history. Furthermore, to reduce the interference of the irrelevant information in the predicted answer, A-Scorer, an answer scorer is designed to evaluate the confidence of the predicted answer. We conduct a series of experiments on QuAC and CoQA datasets, and the results demonstrate the effectiveness and practicality of our proposed Auto-ACE framework.

This paper investigates the sentiment analysis task from a novel perspective by incorporating sentiment knowledge and eye movement into a graph architecture, aiming to draw the eye movement-based sentiment relationships for learning the sentiment expression of the context. To be specific, we first explore a linguistic probing eye movement paradigm to extract eye movement features based on the close relationship between linguistic features and the early and late processes of human reading behavior. Furthermore, to derive eye movement features with sentiment concepts, we devise a novel weighting strategy to integrate sentiment scores extracted from affective commonsense knowledge into eye movement features, called sentiment-eye movement weights. Then, the sentiment-eye movement weights are exploited to build the sentiment-eye movement guided graph (SEMGraph) model, so as to model the intricate sentiment relationships in the context. Experimental results on two sentiment analysis datasets with eye movement signals and three sentiment analysis datasets without eye movement signals show that the proposed SEMGraph achieves state-of-the-art performance, and can also be directly generalized to those sentiment analysis datasets without eye movement signals.