Skip to content

charlieviettq/awesome-agent-skill / .claude/skills/algo-rec-cf

"algo-rec-cf"

Verified

"Implement collaborative filtering for recommendations based on user behavior patterns. Use this skill when the user needs to build a recommendation engine from user-item interaction data, find similar users or items, or predict ratings — even if they say 'users who bought this also bought', 'similar users', or 'recommend based on behavior'.".

Collaborative Filtering

Overview

Collaborative filtering recommends items based on collective user behavior patterns. User-based CF finds similar users; item-based CF finds similar items. Computes in O(U² × I) for user-based or O(I² × U) for item-based where U=users, I=items.

When to Use

Trigger conditions:

  • Building recommendations from user-item interaction data (ratings, clicks, purchases)
  • Finding "users like you also liked" or "frequently bought together" patterns

When NOT to use:

  • When you have no interaction data (cold start — use content-based filtering)
  • When item features matter more than behavior patterns (use content-based)

Algorithm

IRON LAW: CF Requires SUFFICIENT Interaction Data
With sparse matrices (< 1% fill rate), similarity computation is
unreliable. Minimum viable: each user has rated 5+ items, each item
has 5+ ratings. Below this, fallback to content-based or popularity.

Phase 1: Input Validation

Load user-item interaction matrix. Check sparsity level and filter users/items below minimum interaction threshold. Gate: Matrix sparsity < 99%, minimum interaction thresholds met.

Phase 2: Core Algorithm

User-based CF:

  1. Compute pairwise user similarity (cosine or Pearson correlation)
  2. For target user, find top-K most similar users
  3. Predict rating: weighted average of similar users' ratings

Item-based CF:

  1. Compute pairwise item similarity from co-rating patterns
  2. For target item, find top-K most similar items
  3. Predict: weighted average of user's ratings on similar items

Phase 3: Verification

Hold out 20% of interactions for testing. Compute RMSE, MAE, or precision@K / recall@K. Gate: RMSE below baseline (global mean predictor).

Phase 4: Output

Return top-N recommendations with predicted scores.

Output Format

{
  "recommendations": [{"item_id": "123", "predicted_score": 4.2, "similar_items_used": 5}],
  "metadata": {"method": "item-based", "similarity": "cosine", "k_neighbors": 20, "sparsity": 0.97}
}

Examples

Sample I/O

Input: 5 users × 5 items rating matrix, target: user1, item5 Expected: Predicted rating based on weighted similarity of user1's rated items similar to item5

Edge Cases

InputExpectedWhy
New user, no ratingsCannot recommendCold start — fallback to popularity
Item rated by all usersLow differentiationHigh popularity ≠ personalized match
Single shared itemUnreliable similarityNeed multiple co-ratings for stable similarity

Gotchas

  • Scalability: User-based CF with millions of users is O(U²). Use approximate nearest neighbors (LSH) or switch to item-based CF (item catalog is usually smaller).
  • Popularity bias: Popular items have more co-ratings, inflating their similarity scores. Normalize by inverse popularity.
  • Implicit vs explicit feedback: Clicks/views (implicit) need different treatment than ratings (explicit). Use confidence weighting for implicit data.
  • Similarity metric matters: Cosine similarity ignores rating scale differences; Pearson correlation accounts for user rating biases. Choose based on data characteristics.
  • Gray sheep: Users with unusual taste patterns have no similar peers. CF fails for them — consider hybrid approaches.

References

  • For matrix factorization as a scalable alternative, see references/matrix-factorization.md
  • For implicit feedback handling, see references/implicit-feedback.md
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall