Generic User Behavior: A User Behavior Similarity-Based Recommendation Method.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Recommender systems (RS) are crucial for Big Data, but existing methods struggle with incomplete user preference modeling and ignore behavioral process consistency, leading to biased results.
• Current RS often rely on single interaction types (e.g., purchase) and overlook the sequential nature of user behavior, limiting recommendation accuracy.

Purpose of the Study:

• To propose a novel recommendation method addressing limitations in current recommender systems.
• To enhance user preference modeling by incorporating user behavior similarity and process consistency.

Main Methods:

• Developed a recommendation method based on an Entity Interaction Knowledge Graph (EIKG).
• Extracted fact triples to construct the EIKG, embedded entities and relations, and employed link prediction for recommendations.
• Incorporated collaborative filtering principles with a novel approach to user behavior similarity.

Main Results:

• The proposed EIKG-based method demonstrated superior performance compared to state-of-the-art methods on two public datasets (Scholat and Lizhi).
• Experimental results showed significant improvements across most evaluation metrics, validating the method's effectiveness.

Conclusions:

• The Entity Interaction Knowledge Graph (EIKG) approach effectively models complete user preferences and behavioral consistency for improved recommendations.
• This method offers a promising advancement in recommender system research, particularly for Big Data applications.