PELD

"PELD 是一个基于文本的具有说话者个性特征的情感对话数据集。PELD 中的对话融合了 MELD 和 EmoryNLP 中的情感对话，以及来自 FriendsPersona 的个性特征注释。PELD 中的个性特征是取自 FriendsPersona 中 711 条不同对话中的个性标注。参考标注，一个角色在不同的对话中可能会表现出不同方面的个性。我们只保留六个主要角色的个性特征，因为它们的标注是最频繁的。对于每个主要角色，为了简化，我们通过 $P_n = \frac{1}{K}\sum_{i=1}^K{P_i}$ 来平均他们在所有对话中的注释人格特征，其中 $K$ 是注释的数量。”

PELD is a text-based emotional dialogue dataset with speaker personality traits. The dialogues in PELD integrate emotional dialogues from MELD and EmoryNLP, as well as personality trait annotations from FriendsPersona. The personality traits in PELD are sourced from the personality annotations of 711 distinct dialogues in FriendsPersona. According to reference annotations, a single character may exhibit different aspects of their personality across various dialogues. We only retain the personality traits of the six main characters, as their annotations are the most frequent. For each main character, to simplify the process, we average their annotated personality traits across all dialogues using the formula $P_n = frac{1}{K}sum_{i=1}^K{P_i}$, where $K$ represents the total number of annotations.

搜集汇总

背景与挑战

背景概述

PELD是一个结合情感对话和个性特征注释的文本数据集，重点关注六个主要角色的平均个性特征，适用于情感分析和自然语言处理研究。