A two-stage recommendation optimization algorithm based on item popularity and user features.

Area of Science:

• Computer Science
• Data Science
• Financial Technology

Background:

• Current financial product recommendation systems are often product-centered.
• Existing methods struggle with user cold-start and popularity bias, where users favor
• hot
• products.

Purpose of the Study:

• To propose a novel two-stage recommendation optimization algorithm, CPCF-TSP.
• To enhance financial product recommendation by integrating user features and popularity.
• To mitigate popularity deviation and address the user cold-start problem.

Main Methods:

• Developed a two-stage recommendation optimization algorithm (CPCF-TSP) incorporating item popularity and user features.
• Introduced a popularity weight factor to normalize popularity and modify Pearson's similarity.
• Combined a modified Pearson's similarity function with popularity normalization and user features.
• Integrated a collaborative filtering algorithm within the two-stage procedure for improved precision.

Main Results:

• CPCF-TSP effectively utilizes user demographic characteristics.
• The algorithm mitigates the bias towards recommending only popular financial products.
• It improves modeling performance by combining user features and normalized popularity.
• Demonstrated reduced inaccuracy in calculating recommendation popularity and similarity weights.
• Successfully addressed the user cold-start problem in hybrid recommendation models.

Conclusions:

• CPCF-TSP offers a more precise and balanced approach to financial product recommendation.
• The algorithm is particularly suitable for scenarios with abundant user data and a vast product catalog.
• It enhances recommendation precision by considering user features and mitigating popularity deviations.