City Event Identification from Instagram Data using Word Embedding and Topic Model Visualization

Abstract

In this study, we explore unsupervised methods and models for identifying city events from Instagram data. We combine the use of event keywords with expansions based on word embeddings to identify semantically related terms and seed events, then retrieve event-relevant posts. We apply LDA topic modelling method to the set of relevant posts to a given class of event to discover clusters of targeted events. The system helps us to shed light on the presence or absence of certain type of city events on Instagram posts in the region.