Android Package Classification in 60 Minutes
Few-shot learning using Google Vertex AI’s language learning model (LLM).
Recently, we encountered an interesting challenge: from a provided list of Android package descriptions, we needed to determine which ones might be associated with gambling and quick-cash applications. This information would be used as part of feature engineering for fraud analysis.
It goes without saying that fraud is always a never-ending cat-and-mouse game between fraudsters and the organizations trying to protect their customers and assets. To stay ahead, we’ve been increasingly using machine learning models to detect fraudulent activities.
In this article, I’ll demonstrate how we managed to classify around 4,000 Android package descriptions in around an hour from start to finish!
What We Would Have Done Before LLMs
Before LLMs, in order to achieve this task, we would have to:
- Label by hand quite a substantial number package IDs with their respective categories
- Build and train a classifier
- Rinse and repeat if the results are not satisfactory
