抄録
This paper presents a method for automatically generating pictorial maps that visually represent regional characteristics by
analyzing social media photos. The primary challenge is accurately identifying points of interest (POIs) in areas with dense
and diverse content, which traditional methods often fail to capture. The proposed approach improves on existing methods
by using a high-dimensional spatial clustering algorithm that incorporates both semantic and geographical data, enabling
more accurate discovery of POIs in geographical density areas with diverse content. It also selects representative images for
each POI by assessing the semantic similarity within each cluster. To assess the effectiveness of the approach, a pictorial
map prototype was developed using Flickr photo datasets and user evaluations was conducted. The results show that users
rated the representativeness of the POIs and POI photos highly, with an average score of 4 out of 5, demonstrating that the
proposed method produces maps that effectively convey diverse regional characteristics compared to traditional methods
with an average score of 3.4 out of 5.
