We may have heard personally from someone or from anecdotal reports of people often saying that they never follow the recommendations given by an automated system. Similarly, we can all make the claim that most recommender systems are not intelligent enough and do not provide us with enough information.

Ago Luberg’s doctoral thesis ‘Consolidation of Crowd-Sourced Geo-Tagged Data for Parameterised Travel Recommendations’ explores the different aspects of building an automated tourism recommender system, focusing specifically on visually attractive and intriguing locations.

Thus, the basic idea of the thesis runs counter to the hitherto widespread automated tourism or travel recommender systems which primarily serve commercial purposes, such as offering the most affordable hotels, restaurants, travel tickets, and so on.

The main focus of the research, however, is on the collection and processing of data on tourism sites. The research involved designing and developing several recommender systems – Sightsplanner, Sightsmap, and a recommender system for the Visit Estonia website –, as well as collecting, processing, and consolidating data for the developed systems. The overall objective was to investigate how the so-called weaknesses of different automated systems could be alleviated in order to build better systems.

Notably, different systems consider different aspects. For example, an abundance of photos taken at a specific site indicates the visual appeal of that site; the number of logged visits to the corresponding Wikipedia pages on such sites indicates how well those sites are known; meanwhile, check-ins to systems such as Foursquare/Swarm indicate the actual number of visits.

Thus the research focused on the collection of information from various sources; for example, names, locations, descriptions, photo titles, different types of popularities, etc. One question of particular interest to the author was the problem of detecting duplicate objects in
databases.

What are the results?

The first major result of Ago Luberg’s research was showing how probabilistic and fuzzy logic can be used to calculate a suitability score for tourism sites by employing uncertain categories, ontologies, and reasoner-based algorithms. This was based on easily expressible user preferences and an existing dataset on tourism sites.

Secondly, the author designed a machine learning-based system for detecting duplicate sites from different databases. For example, the developed algorithm, after being trained on a dataset on eateries located in Tallinn, managed to detect duplicates in a dataset for Riga, which also contained museums, art galleries, etc., with an accuracy of 98%. For comparison, with manually configured parameters, the accuracy was significantly lower: only 85%.

Thirdly, the author designed an algorithm for identifying the name and category of tourism sites, using user-entered descriptions of sites as the basis. Moreover, the extracted data enables information collected from different databases to be merged. For example, based on photo titles from Panoramio, the algorithm was able to find about 56% of tourism sites described on Wikipedia in the UK and France.

* Ago Luberg received his doctorate for his doctoral thesis, which is titled ‘Consolidation of Crowd-Sourced
Geo-Tagged Data for Parameterised Travel Recommendations’, on the 103rd anniversary of the Tallinn University of Technology, on September 17. At the anniversary ceremony, a total of 66 doctoral degrees were awarded, with the welcoming speech for the doctors held by Siret Malmberg, who completed an industrial doctorate at the Department of Materials and Environmental Technology. An academic speech titled ‘Green Transition, University, and Games/Gambling!?’ was held by Professor and Vice-Dean for Research Argo Rosin at the School of Engineering. Additionally, Mente et Manu medals of merit were awarded.