Interpreting the Rhetoric of Visual Advertisements

Area of Science:

• Computer Vision
• Artificial Intelligence
• Marketing Science

Background:

• Visual media possess significant persuasive power, yet traditional computer vision models overlook this aspect.
• Understanding the persuasive messages in visual advertisements is crucial for marketing and communication research.

Purpose of the Study:

• To develop a comprehensive dataset and novel techniques for analyzing the persuasive elements within visual advertisements.
• To enable progress in understanding and decoding the messages conveyed by image and video ads.

Main Methods:

• Creation of a large-scale dataset (64,832 image ads, 3,477 video ads) annotated with ad characteristics.
• Development of multimodal methods utilizing both visuals and slogans for ad analysis.
• Spatial and temporal analysis to identify poignant content in ads.
• Examination of creative story construction, including climax prediction in videos and metaphorical object transformations in images.

Main Results:

• The study provides a rich dataset for advancing research in visual advertising analysis.
• Developed methods demonstrate effectiveness in analyzing ad content using multimodal cues.
• Insights into creative strategies, such as metaphorical depictions and narrative structures, are gained.

Conclusions:

• This work bridges the gap between computer vision and the persuasive nature of visual advertisements.
• The proposed dataset and methods offer a foundation for future research in computational advertising and marketing.
• Enables a deeper understanding of how visual ads construct messages and influence viewers.