AgentTrek

# AgentTrek Data Collection
AgentTrek dataset is the training dataset for the Web agent AgentTrek-1.0-32B. It consists of a total of 52,594 dialogue turns, specifically designed to train a language model for performing web-based tasks, such as browsing and web shopping. The dialogues in this dataset simulate interactions where the agent assists users in tasks like searching for information, comparing products, making purchasing decisions, and navigating websites.

## Dataset Features:
- Total Dialogue Turns: 52,594
- Task Scope: The dialogues span a variety of web-related tasks, including browsing, online shopping, product comparison, and information retrieval.
- Annotation: Each dialogue turn includes both the user input and the system output, reflecting the real-time interaction between the user and the agent in the context of web agent tasks.

## Dataset Details
- Project Page: Visit the [project page](https://agenttrek.github.io/).
- Repository: Access the [GitHub repository](https://github.com/xlang-ai/AgentTrek/).
- Paper: Read the [research paper](https://arxiv.org/abs/2412.09605).

# AgentTrek 数据集收集
AgentTrek 数据集是网页智能体（Web Agent）AgentTrek-1.0-32B 的训练数据集。该数据集总计包含52,594轮对话，专为训练可执行网页浏览、网络购物等网页端任务的语言模型而设计。数据集中的对话模拟了智能体协助用户完成信息检索、产品对比、购买决策以及网页导航等任务时的交互过程。

## 数据集特征
- 总对话轮次：52,594
- 任务覆盖范围：对话覆盖各类网页相关任务，涵盖网页浏览、在线购物、产品对比与信息检索等场景。
- 标注规范：每一轮对话均包含用户输入与系统输出，完整还原了网页智能体任务场景下用户与智能体的实时交互过程。

## 数据集详情
- 项目主页：访问[项目主页](https://agenttrek.github.io/)
- 代码仓库：访问[GitHub 仓库](https://github.com/xlang-ai/AgentTrek/)
- 研究论文：查阅[相关研究论文](https://arxiv.org/abs/2412.09605).