Recommendation System for Tourist Attractions Based on Wi-Fi Packet Sensor Data

The needs of tourists have become diversified in recent years, and providing tourist information can help satisfy these various needs. Although such information should reflect actual travel behavior, it is not easy to collect data on how tourists visit tourist attractions. Nowadays, it is not uncommon to see tourists looking at their smartphone, searching for information about where to go next. If a recommendation system could suggest the next destination, more tourists would appreciate the information and visit there. Therefore, we envisioned a recommendation system based on records of tourist movements from Wi-Fi packet sensor data. The system would provide tourists with relevant information about the recommended locations. The aim of this research was to investigate the computation method for the recommendations, which is necessary to demonstrate this idea. We also compared the results and characteristics of recommendations made based on cosine similarity. The results showed that, when using Wi-Fi data without cleaning, some tourist spots were strongly recommended because of bias in the data observations. Therefore, a dataset was also used that was restricted to tourists who visited three or more places. As a result of this restriction, the accuracy of the recommendations was improved. Furthermore, we made a recommendation based on a dummy variable that represented whether tourists had visited each location, which enabled recommendations to be generated for locations where it was difficult to observe data.