4 ms·
Why not just build a big platform contains all the data, then write kernels for each type of 'violation', learn a pattern, find suspects, then use results to im
by methou 11y ago
Why not just build a big platform contains all the data, then write kernels for each type of 'violation', learn a pattern, find suspects, then use results to improve the kernel. Human labor + Machine intelligence, Yay! Eventually you'll have all kinds of tax law text covered, and patterns of violations.
- raverbashing 11y agoSure, makes sense Except there are new changes to the tax code with some frequency Not sure if it would be easy to find violations or to filter out 'correct usage' (which seems to have some level of subjectivity as well) Benford's law seems to be a nice proxy for some kinds of frauds as well https://en.wikipedia.org/wiki/Benford%27s_law https://en.wikipedia.org/wiki/Benford%27s_law