Abstract
This paper introduces two comprehensive datasets containing information on digital ads in U.S. federal elections from Meta (including Facebook and Instagram) and Google (including YouTube) for the 2022 midterm general election period. We collected ads published on these platforms utilizing their ad transparency libraries and web scraping techniques and added labels to make them more comparable. The collected data underwent processing to extract audiovisual and textual information through automatic speech recognition (ASR), face recognition, and optical character recognition (OCR). Additionally, we performed several classification tasks to enhance the utility of the dataset. The resulting datasets encompass a rich array of features, including metadata, transcripts, and classifications. These datasets provide valuable resources for researchers, policymakers, and journalists to analyze the digital election advertising landscape, campaign strategies, and public engagement. By offering detailed and structured data, our work facilitates diverse reuse possibilities in fields such as political science, communication studies, and data science, enabling comprehensive analysis and insights into the dynamics of digital political campaigns.