bermaneh/codeswitching-sentiment-bias-results-v2

该数据集是语言扰动和SHAP分析的实验结果，基于LinCE Spa-Eng数据集，包含3,360个有效行，并增加了每词的SHAP指标。使用的模型是cardiffnlp/twitter-xlm-roberta-base-sentiment。数据集包含原始句子、扰动句子、交换词、翻译、交换方向、SHAP排名、原始情感标签、原始情感分数、扰动情感标签、扰动情感分数、情感变化、标签变化、原始词SHAP值、扰动词SHAP值、SHAP值变化、原始SHAP排名、扰动SHAP排名、SHAP排名变化、原始SHAP集中度、扰动SHAP集中度、SHAP集中度变化、原始SHAP值（JSON字符串）、扰动SHAP值（JSON字符串）等列。

This dataset contains the results of language perturbation and SHAP analysis experiments, based on the LinCE Spa-Eng dataset with 3,360 valid rows enriched with per-word SHAP metrics. The model used is cardiffnlp/twitter-xlm-roberta-base-sentiment. The dataset includes columns such as original_sentence, perturbed_sentence, swapped_word, translation, swap_direction, shap_rank, original_sentiment_label, original_sentiment_score, perturbed_sentiment_label, perturbed_sentiment_score, sentiment_delta, label_changed, original_word_shap, perturbed_word_shap, shap_value_delta, original_shap_rank, perturbed_shap_rank, shap_rank_delta, original_shap_concentration, perturbed_shap_concentration, shap_concentration_delta, original_shap_values (JSON string), perturbed_shap_values (JSON string), etc.