Planet caravan: full trip planner

• Ano de defesa: 2022
• Autor(a) principal: Lucas Gabriel da Silva Felix
• Orientador(a): Pedro Olmo Stancioli Vaz de Melo
• Banca de defesa: Jussara Marques de Almeida, Fernanda Sumika Hojo de Souza
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Federal de Minas Gerais
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação
• Departamento: ICX - DEPARTAMENTO DE CIÊNCIA DA COMPUTAÇÃO
• País: Brasil
• Palavras-chave em Português: Sistemas de recomendação; Heurísticas de otimização; Turismo; Recomendação de tours; Pontos de interesse
• Link de acesso: http://hdl.handle.net/1843/45134

One of the services that most benefited from the internet expansion was tourism. The internet allowed people to share information about their trips, assisting other tourists to decide the best destination. However, the high data availability on social networks and platforms specialized in tourism (e.g. Foursquare, TripAdvisor), made it difficult to choose the best places to visit, bringing a new problem known as information overload or analysis paralysis. Thus, considering that is difficult to identify one place to visit, planning a complete trip is considered to be an even harder scenario. To tackle this problem, computational techniques such as Recommender Systems (RS) are been applied. Nevertheless, considering the tourism context, traditional RS techniques do not take into consideration several variables that are important in tourism such as the venues' price, distance, and working hours. Hence, aiming to better model such problem works focus on the task of automating trip planning, known in the literature as Orienteering Problem (OP). The OP consists of the task of identifying in a graph the path that maximizes the users' utility while respecting a distance constraint. This task is considered to be computationally expensive and is usually solved through optimization heuristics. To meet the real-world constraints in a trip works in the literature consider different constraints (e.g. places working hours, visit costs). However, several of these works leave aside important aspects for the real-world cases, such as personalization, multiple days of travel, and automated hotel selection. In this work, we tackle the problem presented above with a methodology composed of an RS, responsible for the personalization of the generated routes, and an optimization heuristic, responsible for meeting the several constraints considered in a travel. We named our methodology as Planet Caravan. To evaluate the proposed methodology, we test different techniques on RS and optimization heuristics, introducing $5$ new datasets collected from TripAdvisor of cities in Brazil and Europe. Also, we introduce an appointment constraint that enables users to generate routes around their schedules. Our results show that the Genetic Algorithm (GA) and GRASP are the most suited technique in our evaluation scenarios. Lastly, we evaluate the proposed methodology with real users through a web application. In this scenario, our results show the potential of our proposal with real users.