2:00 PM - 2:15 PM
[HTT14-02] Comparative study of visitors’ perceptions of places by analyzing visual content of photographs in Shinjuku, Tokyo
Keywords:Flickr, photo image, machine learning, GIS
Photographs taken by tourists have the potential to play an important role in understanding visitors’ behaviors and perceptions in tourist destinations. The authors had analyzed geocoded photographs taken in central Tokyo to compare the behavior patterns of foreign visitors. However, our previous study could not identify the contents of photographs. To resolve this issue, the present study employed a machine learning method to analyze the visual contents of photographs. The data obtained were geocoded photos in 2018, taken within 1 km from Shinjuku. We analyzed 1,940 photos taken by visitors from four major countries (e.g., USA, UK, Taiwan and Singapore) to Shinjuku who posted their pictures on Flickr. The images of photographs were analyzed by using Google Cloud’s Vision API that enables us to assign labels to images and automatically classify them into predefined categories. Based on the labels assigned to the photographs, we manually classified them into 8 categories. By employing the locational information of Flickr data, each photograph was geocoded as an input to GIS. The spatial distribution of photographs by category was compared between visitors’ home countries to capture the variation in perception and behavior of visitors. The results indicated that the visitors’ perceptions of Shinjuku varied with home countries of visitors. While all visitors showed a common perception of Shinjuku characterized by urbanism, some differences were observed among the other categories. Analysis of the spatial distribution of photos revealed that the photos categorized as urbanism widely distributed in and around Shinjuku station, whereas the photos categorized as culture concentrated at the east or south side of the station.