Service interruption on Monday 11 July from 12:30 to 13:00: all the sites of the CCSD (HAL, Epiciences, SciencesConf, AureHAL) will be inaccessible (network hardware connection).
Skip to Main content Skip to Navigation

Fouille de séquences de mobilité sémantique : sur l'élaboration de mesures pour la comparaison, l'analyse et la découverte de comportements

Abstract : "Tell me what you have done, and I will tell you who you are". This aphorism, inspired from Foundation by by Isaac Asimov, questions the predictability and current understanding of humans based on their past actions. Are we what we do? This question has become a major issue in many fields such as individual profiling or recommendation systems that look for a revealing indicator of future behaviour or psychology in the past actions of users. In this thesis, we anchor the previous reflection in the framework of human mobility and propose the implementation of a complete methodology (i.e., data pipeline) for the analysis and discovery of behaviors from a set of semantic mobility sequences. This methodology is based on an extensive review of the literature on the properties of human mobility; however, it provides a generic framework for the study of any semantic sequence. An unsupervised learning process (i.e., clustering) is in charge of extracting behaviours and a post-process explicability phase is ensured in order to translate the clusters into intelligible behaviours. As a consequence, we have retained a set of complementary visual and statistical indicators to inform the different aspects of the sequences while taking care to remain sufficiently concise in order to avoid a cognitive overload. This explanation is essential for practical and ethical reasons, but also to include the user in the discovery process. Also, as the sequences involved are complex due to their temporal character and their possible semantic multi-dimensionality (locations, activities, mode of travel, etc.), we propose two new measures for the comparison of such sequences named Contextual Edit Distance and Fuzzy Temporal Hamming distance. These are respectively inspired by the edit distance and the Hamming distance, and feed the previous clustering process. These new measures are based on ontologies and fuzzy logic in order to overcome the semantic, temporal and structural shortcomings of the original distances. These contributions have been applied on different real datasets from the mobility domain -- physical (urban mobility) and virtual (database mining) and have allowed to significantly improve the process of interpretation and behaviour discovery. Finally, with the aim of reusability and sharing, a web application, SIMBA, completes our achievements in order to allow different experts to appropriate our contributions through an interactive tool for data mining and exploratory analysis. The work of this thesis is in collaboration with two ANR and regional projects: Mobi'kids, which aims to understand and characterise the forms of autonomy and conditions of evolution of the daily mobility of young children. And Smartloire, which aims to offer a set of digital tools for tourism professionals and policy makers for recommending itineraries and analysing tourist tracks in the Centre-Val de Loire region.
Complete list of metadata
Contributor : Clément Moreau Connect in order to contact the contributor
Submitted on : Sunday, March 13, 2022 - 7:25:37 PM
Last modification on : Wednesday, March 23, 2022 - 3:38:31 AM


Phd_thesis_Clement (1).pdf
Files produced by the author(s)


  • HAL Id : tel-03607421, version 1


Clément Moreau. Fouille de séquences de mobilité sémantique : sur l'élaboration de mesures pour la comparaison, l'analyse et la découverte de comportements. Intelligence artificielle [cs.AI]. Université de Tours, 2021. Français. ⟨tel-03607421⟩



Record views


Files downloads