Conjunctive query embedding-based few-shot item recommendation.

Area of Science:

• Artificial Intelligence
• Computer Science
• Information Retrieval

Background:

• The user cold-start problem is a significant challenge in recommender systems, particularly with sparse data.
• Existing solutions often require extensive data or complex meta-learning, while knowledge graph-based methods struggle with noise and generalization in sparse settings.

Purpose of the Study:

• To propose a novel recommender system, CQER, that effectively addresses the user cold-start problem.
• To model user intent using conjunctive query embeddings derived from knowledge graphs, improving accuracy in sparse data scenarios.

Main Methods:

• Developed Conjunctive Query Embedding-based Recommender system (CQER).
• Modeled user intent as logical queries composed of relation-level paths.
• Encoded intent via conjunctive query embeddings to capture uncertainty in sparse settings, supporting K-shot recommendation (K=1 or K=5).

Main Results:

• CQER consistently outperforms existing recommendation approaches across multiple datasets.
• The model demonstrates accurate predictions for users with limited interaction history.
• Achieved high interpretability in recommendation inference.

Conclusions:

• CQER offers an effective solution for the user cold-start problem in item recommendation.
• The conjunctive query embedding approach enhances model performance and interpretability in sparse data environments.
• The proposed method provides a robust alternative to traditional meta-learning and knowledge graph propagation techniques.