Problem

The AI system has no knowledge of user preferences and cannot personalize the user experience.

Solution

Elicit user preferences.

Use when

• Other methods for overcoming the cold start problem have failed.

How

Collaborate with an AI/ML practitioner to identify what information the system needs from the user to learn their preferences for personalization.

Trigger an elicitation session to solicit user preferences through selection and/or feedback interactions. For example, ask the user to select favorite movies, cuisines, genres, and so on. Or ask them to indicate their preferences through feedback interactions, such as ratings or like/dislike.

Keep the elicitation session short. Elicit only the minimum amount of information needed to create a meaningful user experience.

Consider, if possible, making the elicitation session optional.

Allow the user to proceed without fully completing the session.

Consider providing the user the option to express more preferences if they wish.

Make the experience of providing preferences fun for the user.

Collect and maintain user data in a privacy-aware way.

User benefits

Enables:

• Efficient personalization
• Self-knowledge
• Learning about various types of system content (See also G1-E: Show a set of system outputs).
• Learning what information the system uses for personalization (See also G11-D: Map system input attributes to system outputs).

Common pitfalls

• The experience of providing preferences is burdensome for the user.
• The trade-off between providing personal information and deriving benefit from the system is not evident and/or raises privacy concerns.
• Keep in mind that eliciting user preferences is cumbersome and distracts from the user’s goals. Use sparingly.