MM-IMDB 公开数据集

MM-IMDB Datasets是基于IMDb核心资源扩展形成的多模态电影领域数据集，旨在整合文本、图像和结构化元数据等多维度信息，支持跨模态人工智能研究与应用。数据集的资源来源包括IMDb公开的电影元数据（如标题、导演、演员、评分、用户评论等），以及通过爬虫技术或API获取的视觉数据，如电影海报、剧照和预告片关键帧，可能还会整合第三方资源库（如TMDB）。在多智能体系统研究中，MM-IMDB数据集为多重知识融合的超图表示提供了丰富的数据支持。通过将文本、图像以及结构化信息整合到一个统一的表示框架中，可以在超图结构中捕捉到电影元素之间复杂的关系。每一部电影不仅是单一数据源的集合，而是多个智能体（如导演、演员、用户评论、视觉资源等）的交互结果，这为构建复杂的电影推荐系统、情感分析、以及跨模态检索任务提供了更加精准的多层次表达方式。这种超图表示能够有效地融合不同模态的知识，提升模型在多智能体环境下的推理能力和泛化能力。在本项目中，数据的采集采用自动化工具抓取IMDb的结构化数据，并通过图像爬虫获取视觉资源，标注则通过人工或半自动化的方式对文本和图像的关联进行标注，并通过众包平台验证数据的一致性。数据清洗后，按分辨率分级存储，存储总容量大约为200GB到1TB。该数据集的核心内容包括1万多条电影描述、10万多条用户评论、高清海报（5万+）、剧照（50万+）和视频关键帧，以及电影的导演、演员、票房、奖项等结构化信息。MM-IMDB涵盖超过5万部电影，时间跨度从经典电影至2024年新片，支持多年代和多文化背景分析。MM-IMDB Dataset is a multimodal film domain dataset developed by expanding upon the core resources of IMDb. It aims to integrate multi-dimensional information including text, images and structured metadata to support cross-modal artificial intelligence research and applications. The dataset's sources include publicly available film metadata from IMDb (such as titles, directors, actors, ratings, user reviews, etc.), as well as visual data obtained via web crawling technologies or APIs, including film posters, stills, and key frames from trailers. It may also integrate third-party resource repositories such as TMDB. In multi-agent system research, the MM-IMDB Dataset provides robust data support for hypergraph-based representations that fuse diverse knowledge modalities. By integrating text, images and structured information into a unified representation framework, complex relationships between film elements can be captured within the hypergraph structure. Each film is not merely a collection of single-source data, but rather the interactive outcome of multiple agents (such as directors, actors, user reviews, visual resources, etc.). This enables more precise multi-level representations for building sophisticated film recommendation systems, sentiment analysis tasks, and cross-modal retrieval applications. Such hypergraph representations can effectively fuse knowledge across different modalities, enhancing the model's reasoning and generalization capabilities in multi-agent environments. For this project, data collection involves scraping structured data from IMDb using automated tools and acquiring visual resources via image crawlers. Annotations are conducted manually or semi-automatically to label the associations between text and images, and data consistency is verified through crowdsourcing platforms. After data cleaning, the dataset is stored hierarchically by resolution, with a total storage capacity ranging from approximately 200 GB to 1 TB. The core content of this dataset includes over 10,000 film descriptions, more than 100,000 user reviews, high-definition posters (over 50,000), stills (over 500,000), video key frames, as well as structured information about films such as directors, actors, box office revenues and awards. The MM-IMDB Dataset covers more than 50,000 films, spanning from classic films to new releases in 2024, supporting multi-era and multi-cultural background analyses.

搜集汇总

背景与挑战

背景概述

MM-IMDB 公开数据集是基于IMDb扩展的多模态电影领域数据集，整合了文本、图像和结构化元数据等多维度信息，支持跨模态人工智能研究。该数据集涵盖超过5万部电影，时间跨度从经典至2024年，包含电影描述、用户评论、海报和剧照等丰富内容，数据量约200GB到1TB。