Skip to main content Skip to navigation

Using AI to Summarize US Presidential Campaign TV Advertisement Videos, 1952-2012

Project Overview

The document explores the application of generative AI in the context of education, particularly through the development of a sophisticated AI-based analysis pipeline designed to automate the transcription and summarization of US presidential campaign TV advertisements. This innovative approach tackles significant challenges such as the scarcity of data and the prohibitive costs associated with manual coding. The findings reveal that the AI-generated transcripts and summaries not only match but often exceed the quality of those produced by human researchers, thereby improving the accessibility and usability of an extensive dataset comprising nearly 10,000 ads from 1952 to 2012. This dataset holds substantial potential for academic research, demonstrating how generative AI can enhance educational tools and resources by providing efficient, high-quality data processing capabilities.

Key Applications

Automated analysis pipeline for TV advertisements

Context: Research on political communication and advertising; target audience includes political scientists, researchers, and students.

Implementation: Developed a parallelized AI-based analysis pipeline that automates the transcription and summarization of political campaign ads.

Outcomes: High-quality transcripts and summaries that enhance accessibility for researchers; validation shows AI performance matches or exceeds human quality.

Challenges: Initial lack of available LLMs capable of video summarization; ensuring descriptive completeness and minimizing bias in summaries.

Implementation Barriers

Technical

The absence of LLMs able to summarize videos at the time of writing limited initial capabilities.

Proposed Solutions: Developed a custom pipeline using available AI tools to generate high-quality summaries.

Data Availability

Manual procurement and annotation of campaign ads are time-consuming, leading to reliance on smaller datasets.

Proposed Solutions: Created a comprehensive automated dataset encompassing 9,707 ads from 1952 to 2012.

Project Team

Adam Breuer

Researcher

Bryce J. Dietrich

Researcher

Michael H. Crespin

Researcher

Matthew Butler

Researcher

J. A. Pyrse

Researcher

Kosuke Imai

Researcher

Contact Information

For information about the paper, please contact the authors.

Authors: Adam Breuer, Bryce J. Dietrich, Michael H. Crespin, Matthew Butler, J. A. Pyrse, Kosuke Imai

Source Publication: View Original PaperLink opens in a new window

Project Contact: Dr. Jianhua Yang

LLM Model Version: gpt-4o-mini-2024-07-18

Analysis Provider: Openai

Let us know you agree to cookies