feat: group movie recommendation engine ("Movie Checker") #17

New Issue

GKaszewski · 2026-07-10T20:16:01Z

GKaszewski commented

2026-07-10 20:16:01 +00:00

When a group of friends wants to watch a movie together but can't decide, the app should suggest the most compatible movies for that specific group based on their viewing history and taste profiles.

Concept

Given N users, find movies that maximize group satisfaction — optimize for minimum unhappiness (nobody hates the pick) rather than maximum average rating.

Approach: Hybrid (local catalog + TMDb discovery)

1. Taste profile per user

Build a preference vector from each user's review history:

Genre weights (derived from ratings — a 5★ Action movie contributes more than a 2★ one)
Director/actor affinity scores
Rating distribution tendencies (some users rate high, some low — normalize)
Preferred decades, languages, runtime ranges
Watch medium preferences (optional signal)

2. Group scoring

For a set of users, score candidate movies by:

Predicted rating per user (collaborative filtering or content-based)
Minimize variance across the group (fairness constraint)
Exclude movies anyone in the group has already watched
Bonus for movies in genres where the group overlaps

3. Candidate sources (hybrid)

Tier 1 — Local catalog (fast, high confidence):

Movies in the DB that none of the group members have watched
Scored using real ratings from other users on the instance
Presented as "From your community"

Tier 2 — TMDb discovery (broader, predicted ratings only):

Query TMDb Discover API with filters derived from group taste overlap (shared top genres, preferred decades, favorite directors/actors)
Score results against each user's preference vector
Movies don't need to exist in the DB — pull metadata on-demand
Presented as "You might all enjoy"

Implementation considerations

New domain model: GroupSession (temporary, contains user IDs + generated recommendations)
New port: RecommendationEngine with recommend_for_group(user_ids) -> Vec<Recommendation>
TMDb Discover API integration (already have TMDb client infra in tmdb-enrichment adapter)
API: POST /api/v1/recommendations/group with { user_ids: [uuid] }
SPA: "Group Pick" flow — select friends from followers, get ranked suggestions with match percentage
Could run as a background computation for larger groups
Consider caching taste profiles and invalidating on new reviews

Open questions

Algorithm choice: collaborative filtering vs content-based vs hybrid scoring
How to handle users with very few reviews (cold start within the group)
Should group sessions persist or be ephemeral?
Integration with watchlists (prioritize movies on someone's watchlist?)

When a group of friends wants to watch a movie together but can't decide, the app should suggest the most compatible movies for that specific group based on their viewing history and taste profiles. ## Concept Given N users, find movies that maximize group satisfaction — optimize for minimum unhappiness (nobody hates the pick) rather than maximum average rating. ## Approach: Hybrid (local catalog + TMDb discovery) ### 1. Taste profile per user Build a preference vector from each user's review history: - Genre weights (derived from ratings — a 5★ Action movie contributes more than a 2★ one) - Director/actor affinity scores - Rating distribution tendencies (some users rate high, some low — normalize) - Preferred decades, languages, runtime ranges - Watch medium preferences (optional signal) ### 2. Group scoring For a set of users, score candidate movies by: - Predicted rating per user (collaborative filtering or content-based) - Minimize variance across the group (fairness constraint) - Exclude movies anyone in the group has already watched - Bonus for movies in genres where the group overlaps ### 3. Candidate sources (hybrid) **Tier 1 — Local catalog** (fast, high confidence): - Movies in the DB that none of the group members have watched - Scored using real ratings from other users on the instance - Presented as "From your community" **Tier 2 — TMDb discovery** (broader, predicted ratings only): - Query TMDb Discover API with filters derived from group taste overlap (shared top genres, preferred decades, favorite directors/actors) - Score results against each user's preference vector - Movies don't need to exist in the DB — pull metadata on-demand - Presented as "You might all enjoy" ## Implementation considerations - New domain model: `GroupSession` (temporary, contains user IDs + generated recommendations) - New port: `RecommendationEngine` with `recommend_for_group(user_ids) -> Vec<Recommendation>` - TMDb Discover API integration (already have TMDb client infra in `tmdb-enrichment` adapter) - API: `POST /api/v1/recommendations/group` with `{ user_ids: [uuid] }` - SPA: "Group Pick" flow — select friends from followers, get ranked suggestions with match percentage - Could run as a background computation for larger groups - Consider caching taste profiles and invalidating on new reviews ## Open questions - Algorithm choice: collaborative filtering vs content-based vs hybrid scoring - How to handle users with very few reviews (cold start within the group) - Should group sessions persist or be ephemeral? - Integration with watchlists (prioritize movies on someone's watchlist?)

GKaszewski added the backend big feature labels 2026-07-10 20:17:43 +00:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: GKaszewski/movies-diary#17