Evaluation Strategies for Learning Algorithms of Hierarchies

摘要: Several learning tasks comprise hierarchies. Comparison with a “gold-standard” is often performed to evaluate the quality of learned hierarchy. We assembled various similarity metrics that have been proposed in different disciplines and compared them unified interdisciplinary framework for hierarchical evaluation which based on distinction three fundamental dimensions. Identifying deficiencies measuring structural similarity, we suggest new measures this purpose, either extending existing ones or ideas. Experiments an artificial dataset were compare measures. As shown by our results, vary greatly their properties.