Imagery as Inquiry Multimodal Dataset

Imagery as Inquiry Multimodal Dataset是由加州大学圣地亚哥分校创建的一个多模态数据集，旨在通过图像表达用户偏好，涵盖从风景到艺术描绘的广泛视觉表达。用户通过图像请求书籍或音乐推荐，这些推荐通过社区的点赞得到认可。数据集支持两种推荐任务：标题生成和多选选择。该数据集的创建涉及从Reddit社区提取包含图像和文本查询的帖子，经过手动检查确保数据的有效性。数据集的应用领域是多模态会话推荐系统，旨在解决如何通过图像更好地理解和提供个性化推荐的问题。This multimodal dataset, named Imagery as Inquiry Multimodal Dataset, was developed by the University of California, San Diego. It aims to express user preferences through imagery, covering a wide range of visual representations spanning from landscapes to artistic depictions. Users submit requests for book or music recommendations via images, and these recommendations are validated by community upvotes. The dataset supports two recommendation tasks: caption generation and multiple-choice selection. Its construction involved extracting posts containing images and text queries from the Reddit community, followed by manual review to ensure data validity. The dataset is applied in multimodal conversational recommendation systems, with the objective of solving the problem of how to better understand user preferences and provide personalized recommendations through imagery.

搜集汇总

背景与挑战

背景概述

Imagery as Inquiry Multimodal Dataset是一个由加州大学圣地亚哥分校创建的多模态数据集，旨在通过图像表达用户偏好，支持书籍或音乐推荐请求，并通过社区点赞验证推荐质量。它支持标题生成和多选选择两种任务，数据来源于Reddit社区帖子，经过手动检查，适用于多模态会话推荐系统，以提升基于图像的个性化推荐能力。