From edit distance to augmented space-time-weighted edit distance: Detecting and clustering patterns of human activities in Puget Sound region

Abstract Considering and measuring the similarity of human activities remains challenging. Existing studies of similarity measures based on traditional edit distance (ED), specifically on activity patterns, do not reflect the spatiotemporal characteristics in the measurement model. Additionally, interdependence between activities is ignored in existing multidimensional sequence alignment methods. To address the gap, we initially extend the traditional edit distance to a space-time-weighted edit distance (STW-ED). Specifically, differences in distance and time between activities are considered cost functions in the operation cost calculation (insertion, deletion, and substitution). We advance STW-ED to an augmented space-time-weighted edit distance method (ASTW-ED) that integrates an optimum-trajectory-based multidimensional sequence alignment method (OT-MDSAM) with STW-ED, treating the nonspatiotemporal dimensions as augment factors. In addition, ontology is considered for the similarity measure for nonspatiotemporal dimensions. To show the feasibility of our proposed approach, we conduct an empirical study based on an activity-based travel survey in the Puget Sound Region. Eight clusters (homemakers, regular workers with a colorful life, regular workers with a monotonous life, part-time workers, recreation travelers, senior travelers, no-job travelers, and night owl adventurers) are identified based on ASTW-ED and ontology. To cluster the similarity matrix derived from the introduced methods, the affinity propagation (AP) clustering method is employed because it is free of prior knowledge for clustering and can produce exemplars of the clusters. The empirical study indicates that, relative to existing methods for multidimensional activity similarity measurement and clustering, ASTW-ED performs better in terms of within-group homogeneity and between-group heterogeneity of clusters. In addition, the results reveal that ontology can improve clustering performance if it is considered for nonspatiotemporal dimensions provide better understanding of human behavior for urban governance..