Rish Bhardwaj

I started this off as an exercise to practise my newly learnt NLP and ML skills in my 2nd year at college. This is an Android Application to manage money transactions (borrowing and lending). I built this because I was using Google Keep to keep a track of how much I owe people, and how much people owe me. We didn't use Splitwise then, and so this was an easy to use application where you enter transaction details as a sentence and the app interprets it automatically. The extraction bit was fairly simple really; but I made slightly difficult by not using NLTK. I took about a bunch of sentences that conveyed transactional information; for ex:

give 100 to rish; (I lose money here (-100))
take 50 from andrew (I get back money here (+50))
I owe 40 to mike (-40)
Monica owes me 60 (+60)

I then created several hundred more sentences by replacing the value with random numbers and different names that I picked up from a github database. Then, I ran a simple SVM to classify whether I gained or lost money in the transaction. This then had to broken down into [person1] [person2] [value]. I was happy with the gained/lost classifier, and so I started building a UI on Android before completing the project; but then sadly I never got to porting the tiny bit of learning in the app; instead I just used regex to get the number in the sentence, and kind of hardcoded if-this-structure-then-do-that.