Multilingual Sentiment Transfer Datasets

本研究介绍了针对八种印度语言（包括印地语、马拉地语、泰卢固语等）的多语言情感转移数据集。每个数据集包含1000条正面和1000条负面风格平行的句子，源自餐厅评论，旨在保持内容一致性同时改变情感倾向。数据集通过专业语言专家手动翻译英语数据集创建，确保目标语言的自然性和风格准确性。此数据集适用于多语言文本风格转移研究，特别是情感转移任务，有助于解决跨语言情感分析和生成的问题。

This study presents a multilingual sentiment transfer dataset for eight Indian languages, including Hindi, Marathi, Telugu, and other similar languages. For each language, the dataset contains 1,000 pairs of parallel sentences with positive and negative sentiment styles respectively, which are sourced from restaurant reviews. The core design goal of the dataset is to retain consistent content while flipping the original sentiment polarity. The dataset was developed by professional linguists manually translating an English source dataset, ensuring the naturalness and stylistic accuracy of the target language versions. This dataset is suitable for multilingual text style transfer research, particularly sentiment transfer tasks, and can assist in addressing key challenges in cross-lingual sentiment analysis and generation.