Skip to content

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

"algo-rec-hybrid"

Verified

"Design hybrid recommendation systems combining multiple strategies for improved accuracy. Use this skill when the user needs to overcome single-method limitations, combine collaborative and content-based filtering, or build a production recommendation pipeline — even if they say 'combine recommendation approaches', 'best recommendation architecture', or 'cold start plus personalization'.".

Hybrid Recommendation System

Overview

Hybrid recommendation combines multiple strategies (CF, content-based, knowledge-based) to overcome individual method limitations. Common architectures: weighted, switching, cascade, feature augmentation, and meta-level. Complexity varies by architecture.

When to Use

Trigger conditions:

  • Building a production recommendation system that must handle cold start AND personalization
  • Single methods have known weaknesses for your use case
  • Need to balance accuracy, diversity, and coverage

When NOT to use:

  • When you have a single clean data source (start with the matching single method first)
  • When system simplicity is more important than marginal accuracy gains

Algorithm

IRON LAW: Hybrid Adds Value ONLY With Complementary Strengths
Combining two systems with the SAME weakness amplifies the weakness.
CF fails on cold start + content-based fails on cold start = hybrid
STILL fails on cold start. Choose components that cover each other's gaps.

Phase 1: Input Validation

Identify available data: interaction history (for CF), item features (for content-based), contextual signals (time, device, location). Map data to method capabilities. Gate: At least two complementary data sources available.

Phase 2: Core Algorithm

Weighted hybrid: Score = α × CF_score + β × CB_score. Tune weights via cross-validation.

Switching hybrid: Use CF when sufficient data exists; switch to content-based for cold start items/users.

Cascade hybrid: First stage filters (e.g., content-based), second stage ranks (e.g., CF) within filtered set.

Feature augmentation: Use one method's output as input features for another (e.g., CF embeddings as content features).

Phase 3: Verification

A/B test hybrid vs individual components. Measure: accuracy (NDCG, precision@K), coverage (% of catalog recommended), diversity (intra-list diversity). Gate: Hybrid outperforms best individual component on primary metric.

Phase 4: Output

Return recommendations with source attribution for explainability.

Output Format

{
  "recommendations": [{"item_id": "789", "score": 0.91, "sources": {"cf": 0.85, "content": 0.95}, "method": "weighted"}],
  "metadata": {"architecture": "weighted", "weights": {"cf": 0.6, "content": 0.4}, "coverage": 0.78}
}

Examples

Sample I/O

Input: New user with 2 interactions + rich item feature catalog Expected: Switching hybrid: content-based recommendations (insufficient CF data), transitioning to CF as interactions accumulate

Edge Cases

InputExpectedWhy
Completely new user + new itemFall back to popularityNo data for either method
Methods disagree stronglyDepends on architectureWeighted averages; cascade defers to second stage
One component returns emptyOther component takes overGraceful degradation

Gotchas

  • Complexity cost: Each added component increases latency, maintenance, and debugging difficulty. Start simple, add complexity only when justified by metrics.
  • Weight tuning: Static weights degrade over time. Retune periodically or use learned weights (e.g., a meta-model that predicts which component performs best per context).
  • Evaluation is harder: You must evaluate the hybrid AND each component individually to understand contribution and detect regressions.
  • Feature leakage: In feature augmentation, ensure the augmenting model's predictions don't leak test-set information during training.
  • Diminishing returns: Going from one method to two gives the biggest lift. Adding a third rarely justifies the complexity.

References

  • For architecture selection decision guide, see references/architecture-selection.md
  • For A/B testing recommendation systems, see references/ab-testing-recs.md
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall