BGAT-CCRF: A novel end-to-end model for knowledge graph noise correction.

Area of Science:

• Artificial Intelligence
• Data Science
• Knowledge Representation

Background:

• Knowledge graphs (KGs) are crucial for data representation but often contain noise.
• Existing KG noise correction methods struggle with triples containing multiple errors.

Purpose of the Study:

• To develop an advanced model for effective knowledge graph noise correction.
• To address the limitations of current methods in repairing multi-error triples.

Main Methods:

• Proposed a novel end-to-end model, BGAT-CCRF, combining a balanced-based graph attention model (BGAT) and a constrained conditional random field (CCRF).
• BGAT learns node features and correlations within triple neighborhoods.
• CCRF utilizes constraints to select multiple candidates for simultaneous multi-noise correction.

Main Results:

• The BGAT-CCRF model demonstrated superior performance in KG noise correction experiments.
• Achieved a 3.58% improvement in the fMRR metric on the FB15K dataset compared to state-of-the-art models.
• Successfully repaired multiple noises in triples simultaneously by selecting candidates from a restricted domain.

Conclusions:

• BGAT-CCRF offers a significant advancement in knowledge graph noise correction, particularly for complex noisy triples.
• The model's ability to handle multiple errors simultaneously enhances its practical applicability.
• This research facilitates the more robust implementation of knowledge graphs in real-world scenarios.