TG-ReDial

TG-ReDial 是一个主题引导的会话推荐数据集，用于研究会话/交互式推荐系统。我们发布了一个主题引导的会话推荐数据集，用于研究会话/交互式推荐系统：TG-ReDial v1.0。它拥有以下亮点：

1st：多轮话题引导对话推荐
我们的数据集包含主题线程以强制向推荐场景进行自然语义转换。从非推荐主题开始，主题线程通过多轮交互通过一系列不断发展的主题自然地将用户引导到推荐场景。

2nd：半自动标注数据集
我们利用真实数据来构建推荐的电影、主题线程、用户资料和话语。我们提出了半自动方法和新颖的规则，以迫使人类注释变得更加合理和可控。

3rd：详细的用户背景资料
在我们的数据集中，我们将每个用户与用户个人资料和交互历史结合起来，这有助于对用户的个性和偏好进行建模。

TG-ReDial is a topic-guided conversational recommendation dataset dedicated to researching conversational and interactive recommendation systems. We present TG-ReDial v1.0, a targeted dataset for this research domain. It boasts the following highlights:
1st: Multi-turn topic-guided conversational recommendation
Our dataset incorporates topic threads that facilitate natural semantic transitions into recommendation scenarios. Starting from non-recommendation-related topics, these topic threads naturally guide users into the recommendation scenario through multi-turn interactions and a series of evolving topics.
2nd: Semi-automatically annotated dataset
We leverage real-world data to construct recommended movies, topic threads, user profiles and conversational utterances. We propose semi-automatic annotation methods and novel rules to ensure human annotations are more reasonable and controllable.
3rd: Detailed user background profiles
In this dataset, each user is paired with both their personal profile and interaction history, which supports the modeling of user personalities and preferences.

搜集汇总

背景与挑战

背景概述

TG-ReDial是一个主题引导的会话推荐数据集，专为研究会话/交互式推荐系统设计。其特点包括：通过多轮话题引导自然转换到推荐场景，采用半自动标注方法提高数据合理性，并提供详细的用户背景资料以支持个性化建模。该数据集由中国人民大学于2020年发布，适用于自然语言处理和推荐系统领域的研究。