The project which is implemented by the program “Research, create, Innovate” is a partnership between two research organizations: Laboratory of Artificial Intelligence & Information Analysis, Department of Informatics AUTh- ‘’ATHENA’’ Research Center/ΙΕΚ of Xanthi and the companies LINK TECHNOLOGIES SA, 3PI INFORMATIONS SYSTEMS SA., DOTSOFT SA, MLS.
The tourism experience is a multidimensional process to find proper answers to a series of questions: Where should I go? Which sights are worth visiting? Where should I stay? Where to eat? How am I going to have fun? How can I get there? Responses should aim to increase visitor satisfaction by taking into account his or her preferences, keeping them private, mining useful and suitable information from an enormous amount of subjective experiences recorded in social media or collected from interconnected devices at the rising of the Internet of Things era. Subjective experiences, as recorded on videos, photos, music clips, tags, texts, ratings, geo-location traces, user profiles constitute indeed big data in terms of volume, veracity, velocity, but also added value for the tourism industry as well as individual visitors.
The vast amount of information creates a scientific and technological challenge: to develop innovative tools and applications which provide personalized and targeting tourism information to visitors by exploiting above-mentioned multiple information sources. The basic premise of the project “PROMOTE” is that such a personalized and targeted tourist information emerges as a solution of a optimization problem subject to proper constraints limitations within a big data framework maintaining protection of personal data.
Such a solution will be sought at the server side by leveraging a) hypergraphs, which can capture the correlations between the multiple heterogeneous sources, for tourism recommendation; and b) dynamic collaborative filtering models (e.g., Kalman filtering, dynamic factorization of hypergraphs) , which take into account the time evolution of each low-dimensional latent component, representing it as multidimensional Brownian motion. A connection between these two driving forces can be established through graphical models and approximate/variational inference. At the client side, it is necessary to develop mobile phone applications (e.g., i-phone/android apps) for applications for tablets that adhere to human-centered interaction within the truly multilingual tourism framework. An open problem in such a human-centered interaction is keyword spotting (i.e., the detection of landmarks, proper names) in a noisy multi-lingual environment. In the project “PROMOTE”, a solution to this problem will be sought by employing deep neural networks. Deep neural networks are also suitable for text/speech generation to describe a landmark or attraction in language, missing from a Wikipedia entry. System integration also address the minimization of the amount of data to be transferred between clients and serve, e.g., by enabling feature extraction from video, images, music clips at the client applications. At both server and client sides, the challenges to be addressed are important and timely, since they are research objects of international concern.