YELPNLG

YELPNLG数据集是由加州大学圣克鲁兹分校自然语言与对话系统实验室创建的，包含30万条丰富的平行意义表示和高度风格化的参考文本。该数据集通过自动从自由可用的用户评论中提取数据构建而成，特别关注餐厅属性的多样性。创建过程中，研究团队利用了依赖解析和丰富的词汇、句法及情感信息，确保了数据集在语义内容、情感和语言多样性上的丰富性。YELPNLG数据集的应用领域主要集中在神经自然语言生成（NNLG），旨在解决模型输出简单、重复以及新任务数据获取困难的问题。

The YELPNLG dataset was created by the Natural Language and Dialogue Systems Lab at the University of California, Santa Cruz, and contains 300,000 rich parallel meaning representations and highly stylized reference texts. Constructed by automatically extracting data from freely available user reviews, this dataset places special emphasis on the diversity of restaurant attributes. During its development, the research team utilized dependency parsing and rich lexical, syntactic, and affective information to ensure the dataset's richness in terms of semantic content, sentiment, and linguistic diversity. The YELPNLG dataset is primarily applied in the field of neural natural language generation (NNLG), aiming to address issues such as simplistic and repetitive model outputs as well as the difficulty of acquiring data for new tasks.

搜集汇总

背景与挑战

背景概述

YelpNLG是一个用于自然语言生成的餐厅评论语料库，基于Yelp评论构建，包含约30万条语义表示到自然语言参考的实例，支持语义和风格（如情感、长度）的联合控制，并提供训练/开发/测试分割及领域属性词典（如氛围、美食）。