Organizing Tagged Knowledge: Similarity Measures and Semantic Fluency in Structure Mining.

Area of Science:

• Computer Science
• Data Mining
• Knowledge Discovery

Background:

• Structure mining is crucial for system inference but challenging to perform.
• Expert knowledge is often elicited via tags, but their relationships are hard to quantify.
• Existing tag similarity models have limitations in structure recovery.

Purpose of the Study:

• To investigate the assumptions and implications of existing tag similarity measures.
• To develop a novel tag similarity measure inspired by human semantic fluency.
• To improve structure recovery for system analysis and prediction.

Main Methods:

• Analyzed correlation-based and conditional-based tag similarity models.
• Developed a new tag similarity measure emulating human tag recall from memory.
• Empirically tested the proposed measure on structure recovery tasks.

Main Results:

• The new tag similarity measure effectively combines strengths of existing paradigms.
• Demonstrated improved structure recovery compared to traditional methods.
• Validated the approach through a case study on excavator maintenance data.

Conclusions:

• The novel tag similarity measure enhances structure mining capabilities.
• This method offers a more intuitive and effective way to infer system structures from tags.
• It shows promise as a pre-processor for semi-supervised learning in real-world applications.