Planning for tourism routes using social networks

Abstract

Traveling recommendation systems have become very popular applications for organizing and planning tourist trips. Among other challenges, these applications are faced with the task of maintaining updated information about popular tourist destinations, as well as providing useful tourist guides that meet the users preferences. In this work we present the PlanTour, a system that creates personalized tourist plans using the human-generated information gathered from the minube¹ traveling social network. The system follows an automated planning approach to generate a multiple-day plan with the most relevant points of interest of the city/region being visited. Particularly, the system collects information of users and points of interest from minube, groups these points with clustering techniques to split the problem into per-day sub-problems. Then, it uses an off-the-shelf domain-independent automated planner that finds good quality tourist plans. Unlike other tourist recommender systems, the PlanTour planner is able to organize relevant points of interest taking into account user’s expected drives, and user scores from a real social network. The paper also highlights how to use human provided recommendations to guide the search for solutions of combinatorial tasks. The resulting intelligent system opens new possibilities of combining human-generated knowledge with efficient automated techniques when solving hard computational tasks. From an engineering perspective we advocate for the use of declarative representations of problem solving tasks that have been shown to improve modeling and maintenance of intelligent systems.

Introduction

Tourism is an important social, cultural and economic phenomenon that includes the movement of millions of people around the world with a big impact on the economy of many countries. Therefore, the generation of tourism-related tools can have a huge impact in society. Traveling recommendation systems have become very popular applications for organizing and planning tourist trips (Berka, Plößnig, 2004, Castillo, Armengol, Onaindía, Sebastiá, González-Boticario, Rodríguez, et al., 2008, Moreno, Valls, Isern, Marin, Borràs, 2013, Vansteenwegen, Souffriau, Berghe, Oudheusden, 2011). One of the main bottlenecks of this type of systems consists of the initial population and later maintenance of the information about Points Of Interest (POIs), user ratings, and connection with geographic systems. However, in recent years we have seen the emergence of new social network platforms where users can easily and are willing to update that kind of information (e.g. TripAdvisor² or minube³). Also, the extensive use of tourist mobile applications allows users to request real time information about the schedules, guides or plans that fulfill their preferences (Rodriguez-Sanchez, Martinez-Romo, Borromeo, & Hernandez-Tamames, 2013). Data might come from different services, so middle layers should be developed, as wrappers and crawlers that obtain and integrate available data.

From a crowdsourcing perspective (Manikonda, Chakraborti, De, Talamadupula, & Kambhampati, 2014), users of traveling social networks do not receive an explicit call for supplying relevant tourist information or composing a plan. Instead, they are encouraged to share their experience of past trips and give recommendations to everyone. Therefore, users are helping to acquire personalized relevant information using a collaborative filtering mechanism (Lucas et al., 2012). Collaborative filtering provides a subset of recommendations on what to visit, where to sleep or where to eat. Additionally, the network structure facilitates the acquisition of personalized information related to user’s contacts, which can greatly help on weighting the recommendations by closeness to the user.

As in other application areas, once the data collection and maintenance of information has a reasonable solution, people look at applications that build on top of that data. One such type of added-value applications on the tourism sector is based on automatically generating tourist plans. Currently, there are some platforms that provide related services. For instance, Tripomatic⁴ is a powerful tool for travel planning, but requires the user to select places to visit and manually set up the plan. As another example, CityTripPlanner⁵ is able to automatically generate tourist plans, but it does not suggest places to eat in a reasonable way according to user pace, hunger and restaurant timetables.

In this paper we present PlanTour, a new tool that uses an automated planning approach to generate tourist plans. This planning system was built for the ondroad project, a framework for the management and planning of digital contents and services provided for bus travelers of the ALSA⁶ company. Within this project, PlanTour is the sub-system in charge of building the tourist plans for users visiting a particular city or region.

In terms of planning applications, the main contributions of this work can be summarized as:

• The automatic composition of the initial state and goals using information from a social network. This partially tackles the problem of the slow start of similar systems, which have to wait until a sufficient amount of data is collected to run properly.

• The modeling of user drives as part of the planning process, to obtain more realistic plans in terms of deliberative reasoning. Specifically, suggested restaurants are smoothly integrated in tourist plans just when it is expected that the user is hungry based on her preferences.

• Modeling the problem of recommending tourist POIs as an oversubscription planning task, given that the available alternatives (visiting POIs) are many more than the ones a tourist can carry out with her time and budget.

• Successful application of compiling away soft goals using the approach by Keyder and Geffner (2009) to solve the oversubscription planning task.

• The domain-dependent problem decomposition using a clustering algorithm. Resulting sub-problems match the problem of finding a plan for a single day where candidate POIs are geographically close.

The following sections describe the problem formulation, the architecture with all its components, the representation of the domain, the experimental results, the related work, and finally, conclusions and future work.