Automatic Extraction of User-Centric Aspects for Tourist Spot Recommender Systems Using Reviews in Japanese

In tourist reviews, various pieces of information are described to confirm the characteristics of tourist spots. This paper proposes a method to extract tourist spot aspects using Japanese tourist reviews automatically. The aspects of tourist spots are concepts that serve as criteria for searching the features of spots, such as “history” or “nature”. Utilizing these aspects helps tourists easily find tourist spots with their characteristics. Thus, the aspects of whole spots should be readable and understandable for users. Using the Natural Language API, the proposed method extracts entities from tourist reviews on Jalan.net, a travel website. After transforming them into word vectors, the proposed method clusters them to create clusters of entities that reflect the features of the spots. Furthermore, the proposed method seeks synonyms for each cluster to extract aspects. In experiments, this paper quantitatively evaluated the automatically extracted aspects for 30 tourist spots in Okayama Prefecture in Japan. The experimental results revealed that the proposed method extracted a total of 446 aspects, of which 78.9% were deemed appropriate as aspects. In addition, the extracted aspects had been abstracted interpretably for user-centric expression to find tourist spots.