ASIE: Modeling the Bridge Between Offline TV Ads and Online Social Conversations

TL;DR

Mass media campaigns against tobacco use don't just happen on TV—they spark massive digital "aftershocks" on platforms like Twitter. This paper introduces the ASIE (TV Advertisements Social Influence Estimation) framework, which uses psychology-driven probabilistic modeling to predict how offline TV ratings translate into online social media activity.

Background: Beyond the TV Screen

Smoking remains a leading cause of preventable death, and anti-tobacco campaigns like "CDC Tips" have used TV ads for decades to drive behavior change. However, in the age of "Social TV," the conversation has shifted. A single TV ad doesn't just reach the person watching; it triggers a tweet, which reaches a follower, creating a hybrid diffusion process.

The authors identify a critical gap: Existing diffusion models assume information only moves from person to person (internal), ignoring the massive "shocks" from mass media (external).

The Core Insight: Cognitive, Affective, and Conative

To bridge this gap, the authors didn't just look at data—they looked at psychology. They adapted the "Hierarchy of Effects" theory to define how an individual moves from seeing an ad to posting a tweet:

1. Cognitive Stage: Did the user actually see the ad or the tweet? (Awareness)
2. Affective Stage: Did the message leave an impression? (Memory & Emotion)
3. Conative Stage: Did the user decide to act? (Tweeting)

Methodology: The ASIE Framework

The model calculates the probability of each stage using specific variables. For TV ads, the Cognitive probability is tied to Nielsen ratings. For social media, it's tied to the Jaccard similarity (link strength) between users.

A key innovation is the use of the Poisson Binomial Distribution to aggregate these influences. Since a user might be exposed to 10 different TV airings and 5 friend-retweets, the model must calculate the likelihood of being "activated" by any combination of these $m+n$ exposures.

(Note: Refer to Figure 1 and Table II in the paper for the 'Time Window' logic and Probability definitions.)

Experimental Results

The authors tested ASIE against two major datasets:

• CDC Tips: A government campaign with high TV correlation (Pearson: 0.64).
• Legacy Truth: A campaign that successfully utilized viral marketing and celebrity tweets.

Key Findings:

• Superior Accuracy: ASIE consistently achieved lower Mean Absolute Error (MAE) and Mean Square Error (MSE) compared to traditional Regression or KNN models.
• The Time Factor: Information influence isn't instant. The model performed better when using longer "time windows" (up to 72 hours), proving that the Affective (memory) stage has a significant decay period—people often tweet hours after seeing an ad.
• TV vs. Social: In the "CDC Tips" campaign, TV ads were the primary driver. In "Legacy Truth," social peer influence played a much larger role, showing that the model can adapt to different marketing strategies.

(Note: Refer to Figures 5 and 6 for the MAE and MSE score drops across different time windows.)

Critical Insight & Conclusion

The true value of this work lies in its hybrid nature. By acknowledging that we live in a multi-screen world, the ASIE framework provides a template for public health organizations to measure the "Earned Audience" of their expensive TV placements.

Limitations: The model currently assumes a global "conversion rate" for TV ratings to individual views. Future work could benefit from more granular, localized TV data or sentiment analysis of the tweets themselves to refine the "Affective" stage modeling.

Takeaway for Researchers

If you are building an information diffusion model today, ignoring external mass-media triggers is no longer an option. Hybrid models like ASIE are the new baseline for understanding influence in a connected society.