Centre de documentation
scientifique

Consolidation of crowd-sourced geo-ragged data for parameterized travel recommendations / Ago Luberg (2021)  

Public Titre : Consolidation of crowd-sourced geo-ragged data for parameterized travel recommendations Type de document : Thèse/HDR Auteurs : Ago Luberg, Auteur ; Tanel Tammet, Directeur de thèse Editeur : Tallinn [Estonia] : Tallinn University of Technology Année de publication : 2021 Importance : 159 p. Format : 21 x 30 cm Note générale : bibliographie Dissertation accepted for the defence of the degree of Doctor of Philosophy in Computer Science Langues : Anglais (eng) Descripteur : [Vedettes matières IGN] Géomatique web [Termes IGN] apprentissage automatique [Termes IGN] base de données [Termes IGN] conception orientée utilisateur [Termes IGN] données localisées des bénévoles [Termes IGN] extraction de données [Termes IGN] géoréférencement [Termes IGN] point d'intérêt [Termes IGN] Riga [Termes IGN] site wiki [Termes IGN] système de recommandation [Termes IGN] Tallinn [Termes IGN] taxinomie [Termes IGN] tourisme Résumé : (auteur) The research covered in this thesis is focused on different aspects of the task of creating automated recommendations for tourism, focusing mostly on places of interest like beautiful views, architectural landmarks, charming areas etc. A significant amount of work has been spent on designing and developing actual recommender systems - Sightsplanner, Sightsmap and the automated recommender of Visit Estonia - and their data harvesting methods in order to create a platform for showing the feasibility of the new methods designed and experimented with. The main results of our research are split between three subfields: • Knowledge engineering: we have shown how to formalize fuzzy and uncertain POI categories along with suitable ontologies and reasoner-based algorithms for object matching and score calculation in a real-life context of actual POI-s, available data and easily expressable user preferences. • Machine learning: we have designed a learnable detection system for detecting duplicate POIs from different databases, usable for cross- category, cross-language and cross-city datasets. • We show that learning on Tallinn eateries improved the algorithm parameters to such a degree that on Riga data containing also museums and galleries it gave us 98% accuracy versus 85% accuracy achieved by tuning the algorithm parameters manually. • Knowledge extraction: we have designed an algorithm for high-quality keyword extraction from short crowd-sourced POI descriptions in different languages, able to find a suitable name and to add suitable types to the POI. Our clusterization algorithm is able to merge the POIs based on the extracted data: on the Panoramio and Wikipedia data about U.K. and French locations it was able to find 56% of Wikipedia objects from the textual titles/annotations of Panoramio pictures in the area. Note de contenu : 1- Introduction 2- Related work 3- Involvement in recommender projects 4- Data acquisition and information extraction 5- Data deduplication (using machine learning) 6- Location category and name detection 7- Data storage and object score calculation 8- Conclusions 9- Future work Numéro de notice : 28600 Affiliation des auteurs : non IGN Thématique : GEOMATIQUE Nature : Thèse étrangère Note de thèse : PhD Thesis : Computer Science : Tallinn University of Technology : 2021 DOI : 10.23658/taltech.23/2021 En ligne : https://doi.org/10.23658/taltech.23/2021 Format de la ressource électronique : URL Permalink : https://documentation.ensg.eu/index.php?lvl=notice_display&id=99407