PELD

该数据集名为PELD，是一个结合了说话者个性特征和语句情感标注的情感对话数据集，它由电视剧《老友记》的对话剧本构建而成。PELD数据集中情感分布存在不平衡，中性情感占据了大多数（44.6%），同时还包括了一些较少出现的情感，如恐惧和厌恶。平均而言，每条语句的长度为9.32个词。该数据集被划分为训练集、验证集和测试集，比例分别为8:1:1。其规模包含6,527个三元组，任务旨在进行受个性影响下的情感生成。

This dataset, named PELD, is an emotional conversation dataset integrating speaker personality traits and utterance-level emotion annotations, constructed from the dialogue scripts of the TV series *Friends*. The emotion distribution of the PELD dataset is imbalanced: neutral emotion accounts for the largest share (44.6%), while rare emotions such as fear and disgust are also included. On average, each utterance is 9.32 words in length. This dataset is split into training, validation and test sets with a ratio of 8:1:1 respectively. It contains a total of 6,527 triples, and the task of this dataset targets emotion generation conditioned on speaker personality traits.

搜集汇总

背景与挑战

背景概述

PELD是一个基于文本的情感对话数据集，融合了MELD和EmoryNLP的情感对话数据以及FriendsPersona的人格特质注释，专注于六个主要角色的平均人格特质。该数据集旨在通过人格影响的情感转换方法，自动选择响应情感，适用于情感计算和对话生成研究。