Towards collaborative filtering recommender systems for tailored health communications.

Area of Science:

• Health Communication
• Computer Science
• Behavioral Science

Background:

• Current computer tailored health communications (CTHC) rely on static patient profiles and rule-based systems.
• There is a need for dynamic CTHC systems that adapt message selection based on individual patient feedback.
• Collaborative filtering offers a promising approach for personalizing CTHC.

Purpose of the Study:

• To investigate the effectiveness of collaborative filtering-based CTHC systems.
• To collect explicit user feedback on health messages within a smoking cessation domain.
• To evaluate the accuracy of predicting user message ratings.

Main Methods:

• Collected explicit feedback (ratings) on four message aspects from 100 subjects in a smoking cessation study.
• Developed and compared various collaborative filtering models for rating prediction.
• Analyzed the correlation between different message rating aspects.

Main Results:

• Most users expressed positive opinions regarding the health messages.
• Ratings across the four message aspects were highly correlated.
• Predicting future ratings using a user's past ratings yielded the highest predictive accuracy.

Conclusions:

• Collaborative filtering models show potential for enhancing CTHC systems.
• User feedback is crucial for tailoring health messages effectively.
• Personalized rating prediction significantly improves the accuracy of CTHC.