I'm tinkering around with a subset of the IMDb dataset, and have been thinking about what specific attributes of a movie impact its user rating. I am looking to survey what methods can be used to find such attributes.

One approach could be to use quantitative attributes and integer-coded qualitative attributes (e.g. for genre, action could be 1, horror could be 2, etc.). To find if a particular attribute is important, one idea off the top of my head is to group movies into buckets according to the value of that attribute (e.g. critic score $0-10, 10-20,\ldots,90-100$ or genre $1,2,\ldots$). Then find the average score of all movies for each bucket. Then find the variation in the means of these buckets. For example, if the variance of the bucket means is small, that means even if we change the value of that attribute, the average movie score doesn't change, and so that attribute isn't very important in determining a movie's rating.

This is all an intuitive idea though, and I'm sure there are caveats. What caveats should I be aware of? What assumptions must hold for the above method to work? Is the above method a good one to even begin with?

Apart from that, I could try machine learning methods (maybe based on bagging and boosting?) with attributes as features and movie scores as target variable. Then use these methods' built-in functionality to find feature importances. Is this a viable method?

What I'm trying to get at is - I have an idea in mind, but I'd still like to know all its pitfalls, and whether there are better alternative methods at hand.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.