Toward Interpretable Evaluation Measures for Time Series Segmentation