A contextual edit distance for semantic trajectories

Abstract

The understanding of daily human activity is an active research topic. Thanks to GPS and smartphones, human movements can be monitored and analyzed. In addition, by exploiting Linked Open Data and user personal data, semantic labels and annotations can be added to movements. Thus, semantic trajectories can be considered as sequences of timestamped activities where each activity is described by a semantic label. In this context, a major challenge is the comparison of such semantic trajectories, looking to extract and learning similar human mobility behaviors. We propose CED (Contextual Edit Distance), a generic similarity measure for semantic sequences comparison which improve the Edit Distance to take into account the context similarity between elements in the sequence. CED is configurable to any sequence data and business needs. CCS CONCEPTS• Information systems → Similarity measures; Geographic information systems.

Clustering sequences of multi-dimensional sets of semantic elements

“…The extraction of behavior from a dataset is a process usually performed thanks to unsupervised machine learning. Indeed, clustering methods are based on similarity measures like the ones described in previous subsections and are widely used for the discovery of human behavior, in particular in sequences of mobility [21,31,35]. However, the topology created by similarity measures for semantic sequences is hard to apprehend.…”

“…Alternatively, Moreau et al propose the CED similarity measure [31], which extends Edit Distance measures adapting cost computation to typical mobility characteristics, in particular the redundancy of certain elements, repetition [41] and a certain form of cyclicity [40]. CED measure answers the following requirements: (i) edition cost depends on the similarity of nearby elements (the more similar and closer the elements, the lower the cost of operations), (ii) edition of repeated close elements has low cost, and (iii) similar and close elements can be exchanged with a low cost.…”

“…To the best of our knowledge, there is no similarity measure able to compare sequences of multi-dimensional sets of semantic elements. In this paper we propose to reuse the CED [31] measure, pairing it with an extended similarity measure adapted to multidimensional sets of semantic elements.…”

“…Equations 4 to 6, indicate the CED computation as described in [31]. Firstly, CED includes a modification of the cost operation function 𝛾 which generalizes the classical definition of Edit Distance and takes into account the local context of each element in the sequence.…”

“…Yet, the main interest of these studies on semantic sequences, in particular of semantic mobility sequences, is the clustering of the dataset in order to extract coherent and understanding behaviors or patterns over the sequences. [17,21,31,35]. Moreover, a difficult problem in this context is the comparison of semantic sequences, generally concerning complex semantic elements.…”