Why should you care?

AIPredict fraud and recs both need recency, velocity, and category affinity from event logs.

See it live — copy this example

Use a .NET console or Web API project with Microsoft.ML. Run dotnet run after pasting.

var sessions = events.GroupBy(e => e.UserId)
    .Select(g => new BehaviorRow {
        UserId = g.Key,
        SessionCount7d = g.Count(e => e.Day >= today.AddDays(-7)),
        AvgCartValue = g.Where(e => e.Type == "AddToCart").Average(e => e.Amount),
        TopCategory = g.GroupBy(e => e.Category).OrderByDescending(x => x.Count()).First().Key
    });
IDataView view = ml.Data.LoadFromEnumerable(sessions);

What happened?

• Roll raw clicks into per-user features before recommendation or fraud pipelines consume them.
• Follow the steps below — typing the code yourself is the fastest way to learn.

Practice next

1. Aggregate 7-day events per user.
2. Compute velocity and top category.
3. Load into IDataView.
4. Add night-session ratio.
5. Join aggregates to MF training rows.

Remember

Events → aggregates. Features for ML. Watch time windows.

AIPredict behavior features

Cart velocity joins fraud and rec training.

Outcome: Models see recency, not only static profiles.