Quality Evaluation of Triples in Knowledge Graph by Incorporating Internal With External Consistency

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Evaluating knowledge quality (KQ) in multisource knowledge graphs (KGs) is crucial for applications like knowledge fusion and construction.
• Existing external consistency (EC)-based methods rely on high-quality KGs or statistical analysis, which are often unavailable or compromised by incorrect data.

Purpose of the Study:

• To develop a robust method for evaluating KQ in multisource KGs that mitigates the impact of incorrect knowledge.
• To enhance the accuracy of knowledge quality assessment by incorporating internal structural patterns.

Main Methods:

• Introduced internal consistency (IC) evaluation, assessing triple conformity to the KG's semantic patterns using KG embeddings and logic inference.
• Integrated IC with EC to identify and reduce the influence of erroneous triples.
• Validated the proposed method on multiple datasets.

Main Results:

• The integrated IC-EC approach significantly reduced incorrect knowledge evaluations.
• The method effectively improved the overall quality evaluation of triples in multisource KGs.
• Demonstrated superior performance in mitigating the interference of incorrect knowledge compared to traditional EC methods.

Conclusions:

• Combining internal and external consistency provides a more reliable approach to knowledge quality evaluation in multisource KGs.
• The proposed method offers a practical solution for improving the accuracy and robustness of knowledge graph quality assessment.