Probato

Problem

Producers know their intent and internal context. Customers only see the output. This interpretation gap makes it easy for an organization to miss where customer aversion begins.

Probato focuses on the origin layer of controversy: which content stimulus activates which point of moral aversion. It does not claim to predict the full diffusion process.

Why qualitative data is difficult

Short reactions combine irony, ridicule, stance, emotion, and cultural context. Similar negative expressions can reflect concern about harm, betrayal, unfairness, or resistance to authority.

Sentiment alone cannot explain why customers react or what an organization should reconsider.

Customer psychology structure

I decomposed the problem into content stimulus, reaction representation, and review unit. The representation separates stance from semantic topic, then uses Moral Foundations as coordinates for the direction of aversion.

The structure does not automate moral judgment. It compares which psychological frames are active in customer reactions.

Technical implementation

The workflow connects multi-source collection, data schemas, LLM-assisted labeling, a 6-dimensional Moral Foundations vector, a 768-dimensional semantic embedding, clustering, and a prototype risk score.

Case evidence

For a 2026 Starbucks Korea controversy, I connected news, comments, community discussions, and official sources at aggregate level. Platform and time patterns showed how fairness, betrayal, and degradation signals changed beyond simple reaction volume.

Chart summary and data

The first two sparse intervals on May 18 show high fairness and harm outputs. After the first marked event, all four signals fall. From late May 18 through May 22, fairness is usually highest at about 0.46-0.54. Betrayal is generally about 0.39-0.52, degradation about 0.31-0.43, and harm about 0.20-0.24. Vertical lines mark when the copy appeared, the apology, and the leadership dismissal. These are descriptive three-hour means on a 0-1 model scale, not causal effect estimates.

My role

I handled the full workflow: problem framing, research design, collection pipelines, schemas, modeling, analysis, interpretation, and public-safe presentation.

I used AI tools for code and classification work. Model outputs were treated as analytical features to evaluate, and raw or restricted source text was excluded from this portfolio.

Limitations

This work is a research and decision-support prototype. It is not a validated controversy prediction model or a production commercial service.

Moral Foundations labels are LLM-generated features, not human ground truth. Platform sampling, contextual interpretation, and model-dependent stance labels can all affect the findings.