Fine-grained tourism demand prediction: challenges and novel solutions

• Autor(a) principal: Amir Hassan Khatibi Moghadam
• Data de Publicação: 2021
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/39112
• topic: Computação - Teses; Análise de redes sociais - Teses; Turismo - Predição - Teses; Redes sociais on-line - Teses; Computação - Teses.; Análise de redes sociais - Teses.; Turismo - Predição - Teses; Redes sociais on-line - Teses

Forecasting is of the utmost importance for the Tourism Industry. The development of models to predict visitation demand to specific places is essential to formulate adequate tourism development plans and policies. It is also essential to reduce negative impacts and costs. Usually, cities and countries invest a huge amount of money for planning and preparation in order to welcome (and profit from) tourists. The success of many businesses depends largely or totally on the state of tourism demand. Estimation of tourism demand can be helpful to business planners in reducing the risk of decisions regarding the future since tourism products are, generally speaking, perishable (gone if not used). However, there are a set of challenges to overcome, for instance most of prior studies in this domain focus on forecasting for a whole country and not for fine-grained areas within a country (e.g., specific tourist attractions) mainly because of lack of official census and available data. In other words, only a limited number of works and baselines are available which deal with the hard problem of fine-grained (per attraction) tourism demand prediction. The other challenge is the high uncertainty of tourism demand due to interference of factors like exchange rate, fuel price, climate changes, local and global financial crises and even epidemics and pandemics over cyclic and/or trending behavior of visitations in where could cause dramatic deviations in tourism demand forecasts, if they are not properly considered. On the other hand, with the rapid popularity and growth of social media applications, each year more people interact within online resources to plan and comment on their trips. Motivated by such observation, we here suggest that accessible data in online social networks or travel websites, in addition to environmental data, can be used to support the inference of visitation count for either indoor or outdoor tourist attractions. In addition, we argue that in the context of fine-grained tourism prediction, three specific key requirements should be fulfilled: (i) recency – forecasting models should consider the impact of recent events; (ii) seasonality – tourism behavior is inherently seasonal; and (iii) model specialization – individual attractions may have very specific idiosyncratic patterns of visitations that should be taken into account. We argue that these three key requirements should be considered explicitly and in conjunction to advance the state-of-the-art in tourism prediction models. Our solution to the challenges in fine-grained tourism prediction is a novel architecture using in jointly social media data and environmental features, adaptive to different scenarios of Tourism demand, while we also propose conjunctive inclusion of three main tourism requirements - recency, seasonality and model specialization in the prediction models not only to be able to capture the seasonal aspects of tourism demand but also follow the recent trends due to local/global changes. In our experiments, we analyze visitation counts, environmental features and social media data related to 27 museums and galleries in the U.K. as well as 76 national parks in the U.S. Our experimental results reveal high accuracy levels for predicting tourism demand while we quantify the effect of each type of these features. We also show that the explicit incorporation of Tourism requirements as features into the models can improve the rate of highly accurate predictions by more than 320% against the current state-of-the-art. Moreover, they also help to solve very difficult prediction cases, previously unsolvable by the current models. We also provide in depth analysis regarding the performance of the models in the (simulated) scenarios in which it is impossible to fulfill all three requirements – for instance, when we do not have enough historical data for an attraction to capture seasonality. Finally, another contribution of our paper is a quantification of the impact of each of the three factors in the learned models. Our results show that the most important ones are indeed model specialization and seasonality but recency is very effective when there is not enough historical data about a specific attraction. keywords: Tourism demand prediction, fine-grained prediction, time-series analysis, social media data, environmental data