Few-shot and LightGCN Learning for Multi-Label Estimation of Lesser-Known Tourist Sites Using Tweets

Landy Rajaonarivo, Tsunenori Mine, Yutaka Arakawa

Abstract: When a tourist wants to visit a site, he/she first asks about the category of the site. The availability of detailed information about the site allows him/her to tailor his/her visit to his/her preferences. This information is therefore essential for Point of Interest (POI) recommendation. However, it is rarely or never available for lesser-known POIs. Lesser-known POIs can be considered as places or events that local people know well, and that may be important to them, but that others do not know about. We propose an approach to estimate the categories of lesser-known POIs based on information from social media. The originality of this approach lies in the extraction of information and links between them, the encoding of the POIs, the representation of the data, and the combination of machine learning techniques such as Few Shot Learning, LightGCN, and Clustering for the estimation of POI categories. The results of the experiments would allow us to confirm that our approach can estimate POI categories and thus discover information about POIs that may be relevant. This approach would be useful for our future work on POI recommendations.