What should I use for explainable recommendations?

For explainable recommendations, the best choice depends on how much flexibility vs. interpretability you need. In practice, these are the most common options:

1. Rule-based or heuristic recommenders

• Best for: Maximum transparency
• How they work: “If user bought X and Y, recommend Z.”
• Pros: Easy to explain, easy to debug
• Cons: Usually less accurate and less personalized

2. Content-based filtering

• Best for: Recommendations based on item attributes
• How they work: Recommend items similar to ones the user liked, using features like genre, brand, topic, price, etc.
• Pros: Easy to explain: “We recommended this because it’s similar to what you liked.”
• Cons: Can be limited if item features are weak or sparse

3. Association rule mining

• Best for: Basket/market-basket style recommendations
• How they work: Find patterns like “users who bought A also bought B”
• Pros: Very explainable, intuitive
• Cons: Works best for co-purchase behavior, not deep personalization

4. Case-based / similarity-based recommendations

• Best for: Human-readable similarity explanations
• How they work: Recommend items/users similar to a known case
• Pros: Simple, explainable
• Cons: Can be less scalable or less powerful than ML models

5. Interpretable ML models

• Best for: When you want more predictive power but still need explanations
• Examples: Decision trees, rule lists, GAMs, monotonic models, linear models with meaningful features
• Pros: More accurate than pure rules, still understandable
• Cons: Often not as strong as black-box models for complex patterns

6. Post-hoc explainability on black-box recommenders

• Best for: Strong accuracy with explanation layer
• How it works: Use a model like matrix factorization or deep learning, then generate explanations separately using:
  • feature attribution
  • attention weights
  • counterfactuals
  • nearest-neighbor examples
• Pros: Good balance of performance and explainability
• Cons: Explanations may be approximate, not fully faithful

---

Practical recommendation

If explainability is a top priority, I’d usually suggest:

• Content-based + rule/association explanations for the clearest user-facing reasons
• Interpretable ML if you need better ranking quality
• Post-hoc explanations only if you must use a more complex model

---

If you want a single default choice

Use content-based recommendation with feature-based explanations.

Example:

“We recommended this book because you liked other sci-fi novels by the same author and with similar themes.”

This is usually the easiest to explain to users and stakeholders.

If you tell me your domain (e-commerce, news, movies, jobs, etc.) and whether you need explanations for users or auditors, I can recommend a more specific approach.