Little Known POI Category Estimation via Syntactical Knowledge Graph Generated via Tweets.

Local events are important for the vitality and dynamism of a city. However, some events or sites are lesser-known or only known by the local community. The information about them is then limited or not available. To address this issue, we propose an approach to automatically generate knowledge graphs from information posted on social networks and data on human mobility. This graph is generated from the tweets using the syntactic relationships between the verbs, nouns, and adjectives. Its exploitation would allow to estimate the categories of lesser-known Points of Interest (POIs) or events. It can also allow to generate sentences and to answer questions that tourists may ask. The specificity of our approach lies in the Graph Convolutional Network (GCN)-based encoding of POIs with a graph of verbs, nouns, and adjectives and the estimation of POI categories using the proposed encoding. Experimental results show that our approach performs better than approaches in the literature.