Skip to content

Apr 1, 2018

Understanding Effective and Emotional Components in Advertisements

DenseNet-121 based modeling of arousal, valence, and ad effectiveness with CVPR workshop acceptance.

Overview

Defined emotional components of ads using arousal and valence signals, then modeled ad effectiveness through learned visual predictors.

What I built

  • Emotion score mapping pipeline based on sentiment-tagged corpus.
  • Effective score proxy using user-response heuristics.
  • Fine-tuned DenseNet-121 models for arousal, valence, and effectiveness prediction.

Results

Accepted as a poster at the CVPR Understanding Visual Advertisements Workshop (Salt Lake City, 2018).