(The first thing I’d probably try would be SVD to see if it isolates a “correctness factor”, since it’s simple and worked famously well on the Netflix dataset.)
I’d like to give people quizzes to identify cognitive biases, perform SVD (or factor analysis) on the results, and see if the first dimension matches up with “liberal / conservative”.
PS—The technique referred to as SVD by the Netflix contestants is actually multiple linear regression. SVD uses principal component analysis, so that the first dimension is the dimension with greatest variance, etc. Simon Funk’s original algorithm uses gradient search to approximate SVD, but most Netflix contestants after Funk sped it up by using a gradient-search method that adjusts all of the basis vectors in parallel, without concentrating the variance in any one of them. Therefore, it’s more like multiple linear regression.
I’d like to give people quizzes to identify cognitive biases, perform SVD (or factor analysis) on the results, and see if the first dimension matches up with “liberal / conservative”.
PS—The technique referred to as SVD by the Netflix contestants is actually multiple linear regression. SVD uses principal component analysis, so that the first dimension is the dimension with greatest variance, etc. Simon Funk’s original algorithm uses gradient search to approximate SVD, but most Netflix contestants after Funk sped it up by using a gradient-search method that adjusts all of the basis vectors in parallel, without concentrating the variance in any one of them. Therefore, it’s more like multiple linear regression.