Researchers believe this will help tour operators and everyone associated with the tourism industry to predict the demand for resorts during the holiday season. The findings have been published in the Journal of Ambient Intelligence and Humanized Computing.
A similar approach was earlier tested in the study of reviews of services on high-speed trains in India.
Every day, people share their photos, posts, comments, and locations on social networks. NUST MISiS scientists used a machine learning (ML) approach and Big Data analysis to study how to predict a user’s next location based on Twitter data. ML methods allow the computer to study historical records and use them to predict and make decisions when receiving new data.
“We used not only open travel data, but also the data about users’ personalities. First, we extracted all geotagged tweets and categorised them. From a random set of 5,000 user profiles from different European countries (France, Germany, Sweden, Spain, Italy, Switzerland, Poland, Greece, and many others), more than 800,000 tweets were uploaded. When selecting data, the most visited categories were Food, Nightclubs, Stations, Churches, and Beaches. For each category, we prepared a separate data set", Marina Nezhurina, the author of the study and Director of the Institute of Information Business Systems at NUST MISiS, said.
According to Nezhurina, when selecting data, the categories were compared with the characteristics of the user’s personality, since the word choice in tweets mainly depends on their personal values.
“For any forecast model, accuracy is a critical and mandatory parameter. If the forecast model doesn’t provide good accuracy, it cannot be considered reliable. Therefore, we used the ensemble classification method that combines the results of all basic classifiers", Sachin Kumar, co-author of the study and postdoctoral research fellow at NUST MISiS, said.
According to scientists, a more specific forecast can be obtained by collecting parameters such as citizenship, gender, and user age. The next stage will be the analysis and construction of models using an ensemble of machine learning methods.