2024-10-26 –, CLASS #1
Language: English
Discover how to harness machine learning to automatically tag personal transaction purposes, eliminating the need for manual categorization. This talk will explore the workaround by utilizing an AI architecture namely Graph Neural Networks, to model the interactions between senders and recipients, as well as other information such as the transfer amount and the transfer notes.
Having a well-recorded of personal transaction data is good. This payment is for bills purpose, that one payment is for shopping. Wouldn't it be helpful if we can do automatic tagging based on our behavior? 🤔
This talk will cover topic about how to do this with machine learning approach. Money transfers data form a mathematical structure called graph, containing the interactions from the senders to the recipients. We use the DuckDB library🦆 to handle the big volume of the data, as well as NetworkX to work with graph data. Finally, we model the interactions with Graph Neural Networks architecture in PyTorch🔥 to infer the purpose of the transaction.
The talk format will roughly be:
- 👋🏼 1-2 mins: Introduction
- 💡 2-3 mins: Problem motivation
- 📊 7-8 mins: Datasets - what differentiates the purpose of one transaction from another?
- 🔍 7-8 mins: Workaround and brief intro to Graph Neural Networks
- 🚀 6-8 mins: Model performance
- 🗣️ 3-5 mins: Q&A / awkward silence
This talk is aimed at:
- 👤 Intermediate audiences, specifically data scientists and machine learning engineers or researchers.
- 💪🏼 Beginners or those keenly interested are welcome as well, with at least a basic knowledge of machine learning and deep learning.
Tags
Machine learning, AI, data analytics, financial data.
Rimba Erlangga is a Data Scientist at Flip, a leading money transfer company in Indonesia. With 2+ years of experience in data analytics, he specializes in analyzing lending and marketing data, as well as building and designing machine learning systems. He is also currently pursuing his master's degree in Artificial Intelligence at Universitas Gadjah Mada.