Swipe Left: Implement a Recommendation Filter for a Dating-App Style Feed

Question

## Problem

You are building the content feed for a swipe-based app. Each user has a set of `seen_ids` (profiles already swiped) and a preference vector. Given a pool of candidate profiles with feature vectors, return the top `k` unseen candidates ranked by cosine similarity to the user's preference vector. Exclude already-seen profiles.

```python
def recommend(
    user_pref: list[float],
    candidates: dict[str, list[float]],  # id -> feature vector
    seen_ids: set[str],
    k: int
) -> list[str]:
    pass
```

## Example

```
user_pref = [1.0, 0.0, 1.0]
candidates = {
  "p1": [1.0, 0.0, 1.0],
  "p2": [0.0, 1.0, 0.0],
  "p3": [0.8, 0.1, 0.9],
  "p4": [1.0, 0.0, 0.5],
}
seen_ids = {"p2"}
k = 2

Output: ["p1", "p3"]
# cosine sim: p1=1.0, p3=~0.99, p4=~0.95; p2 excluded
```

## Follow-ups

1. Cosine similarity is O(d) per candidate — how do you scale to millions of profiles?
2. How would you incorporate mutual-match signals (both users swiped right) into ranking?
3. How do you prevent a popularity bias where the same top profiles dominate all feeds?

Swipe Left: Implement a Recommendation Filter for a Dating-App Style Feed

Question Details

Full Details

Topics

More from Palantir