Visualizing Functional Regions by Analysis of Geo-textual Data

Abstract

Using tremendous geo-textual data collected from social media applications, we facilitate the analysis of region functions. By extracting semantics from textual properties, we aim at classifying geographical locations in terms of their functional types. Hence, we train a classification model with the Support Vector Machine, and apply it to aggregated word embeddings to predict the function of spots. We highly cooperate with techniques in graph analysis. Firstly, regions are segmented based on a latent graph. Then, we propose an adaptive layout solution to deal with situations of multi-AOI queries. The generated layout and interactive metaphor provide convenience for observation and comparison. Experiments are conducted with the YFCC100M dataset to prove the effectiveness of our system.