We can build a model from the dataset to estimate how many stars a rating may entail. However, the data points available within a review are only:

• funny
• useful
• cool

These would not appear to be good indicators for a rating number. We can use a model, such as:

model <- lm(stars ~ funny + useful + cool, data=reviews)
summary(model)  

This produces the statistics of the model:

As expected, we don't have enough information to work with:

• Over four million degrees of freedom, just about one per review
• P values are very small—the probability that we have estimated correctly is non-existent
• 3.7 intercept (close to the halfway point of the range)
• Such low affect rates (under one times each factor) meaning we aren't moving far from the intercept