Twitter User Geolocation Based on GCN

The existing text-based social user geolocation methods ignore the association between words, resulting in inadequate text features for a higher user geolocation accuracy. To address this issue, this paper proposes a Twitter user geolocation model based on GCN (graph convolutional neural network). We consider the Twitter user geolocation problem as a text classification task and use the k-d tree algorithm to cluster the user's latitude and longitude coordinates and map the user to different geographic grids. In the proposed model, location-enhanced text features from user tweets are extracted by building a location-enhanced social text feature extraction model. Based on the co-occurrence of words in the social text and the mention relationship between users and words in tweets, the model constructs a heterogeneous graph to make full use of the local correlation and global correlation between words to mine user features and word features. After that, the constructed heterogeneous graph is used as the initial feature and input into the graph convolutional neural network, the node features are updated through the neural network layer, and the output dimension of the last layer network of GCN is set to the total number of categories of user locations. Finally, according to the vector representation of the obtained user node, the Softmax function is used to identify the location category of the user node, and the cross-entropy loss function is used to calculate the loss. Experiments on two public tweet datasets show that our proposed model is superior to existing state-of-the-art text-based location inference methods.