What They Found (And What It Means for ML in the Real World)

Section of this node - https://www.hubnx.com/nodes/200b2ec1-37ec-4be6-adc7-57838917d255/sections/c1677a39-c987-48cd-8d2f-c896a09eff0a

So, we went over the machine learning methods used, and how well they performed, but what did we get from this research and was it useful or influential for ML industry? Let's go over some of the insights and discuss them concisely and understandably.

Findings

Insight 1: People Talk About What They’ll Buy

The tweet-rate had an incredibly strong correlation with real-world revenue.
Even without sentiment, just knowing how often people were talking was a solid predictor.

Insight 2: Sentiment Is Lagging, Not Leading

You can’t predict pre-release success based on sentiment - people are mostly hyped or neutral.
But you can predict drop-offs and longer-term reputation based on tone after release.

Insight 3: Public chatter outperformed Wall Street

Their model beat the Hollywood Stock Exchange, which is a real prediction market used by insiders.

This suggests that crowd signals, when aggregated, are better than many expert forecasts - even when using basic models.