AI driven categorization
R
Robert Wood
It would be help to have more of a "set and forget" aspect to the categorization. It seems to require a lot of watching currently. If we could apply LLMs to the categorization it may help to automate this.
Log In
K
Karan Lyons
You don't need something as heavy and random as LLMs for this, traditional ML/stats techniques will be far more efficient and likely work completely fine given how rigid/consistent transaction descriptions are. Any flavor of Bayes classification (decomposing the descriptions to individual words/"words"; including prices, possibly bucketed or distance matched as a score modifier; and accounts) would probably work great.
You could get fancy with word2vec or something similar but I'd be unsurprised if the improvements were marginal.
I don't know how the matching currently works, it may be something approximating this if not it exactly.
h
huulong
Karan Lyons Yes, I was thinking the same, the basic categorization of my bank already covers most cases and I think it's very barebone (it probably just identifies the recipient's name, which of course has its limitations such as thinking that every Patreon payment is for videos if I label one Patreon payment to a video maker to "Videos", even if other creators do other things), but it shows some ability to remember labelling transactions and reapplying them to future ones. So ML classification would already be a big improvement over that (although still limited by the amount of metadata available, I'm thinking specifically of generic payment platforms like PayPal which tend to hide a lot of things, but then that applies to every method).
For now I'll try the Rules system, it looks pretty complete already.
A hybrid method would also be nice: auto-generate Rules but let the human control the final Rules for fine tuning (a bit similar to how Gmail lets you "Create a filter out of this email" but a bit more smarter as it would look at a few different similar transactions to extract the common point - in Gmail I have to do this manually by finding the perfect keywords with trial-and-error and it takes a bit of time - hopefully transactions will be simpler to handle).