Detecting and Managing Bad Accounts on a Social Platform

1) Probability of a Bad Account Sending Friend Requests

Context: 1% of accounts are bad. Bad accounts send friend requests at 10× the rate of good accounts.

• If a user receives one friend request, what is the probability it comes from a bad account?
• If a user receives five friend requests, what is the probability at least one is from a bad account?

2) Classification Model Performance

We build a classifier to detect bad accounts. It achieves a true positive rate (TPR) of 95% and a true negative rate (TNR) of 95%.

• If the model predicts an account is bad, what is the probability the account is actually bad?

3) Feature Engineering for Bad Account Detection

What types of data would you use to determine whether an account should be classified as a bad or good account?

4) Assessing the Bad Account Problem

How would you determine whether bad accounts pose a significant issue to the platform? Would you use stratified sampling, random sampling, or another approach?

5) Defining a Bad User

How would you define a “bad user” in the context of a social media platform?

6) Impact of Fraudulent Users

What are the potential impacts of fraudulent or bad users on the platform and its community?

7) Impact of Friend Requests from Bad Accounts

What potential effects might arise from friend requests initiated by bad accounts?

8) Precision–Recall Tradeoff

When building a machine learning model to identify bad accounts, how would you approach the tradeoff between precision and recall? In which situations would you prioritize one over the other?