GeoBurst+

Abstract

The real-time discovery of local events (e.g., protests, disasters) has been widely recognized as a fundamental socioeconomic task. Recent studies have demonstrated that the geo-tagged tweet stream serves as an unprecedentedly valuable source for local event detection. Nevertheless, how to effectively extract local events from massive geo-tagged tweet streams in real time remains challenging. To bridge the gap, we propose a method for effective and real-time local event detection from geo-tagged tweet streams. Our method, named G eo B urst+ , first leverages a novel cross-modal authority measure to identify several pivots in the query window. Such pivots reveal different geo-topical activities and naturally attract similar tweets to form candidate events. G eo B urst+ further summarizes the continuous stream and compares the candidates against the historical summaries to pinpoint truly interesting local events. Better still, as the query window shifts, G eo B urst+ is capable of updating the event list with little time cost, thus achieving continuous monitoring of the stream. We used crowdsourcing to evaluate G eo B urst+ on two million-scale datasets and found it significantly more effective than existing methods while being orders of magnitude faster.

Image Analysis Enhanced Event Detection from Geo-Tagged Tweet Streams

“…The following two methods are chosen as the baselines: (1) Geoburst [28], a widely cited event detection algorithm that considers temporal, spatial and semantic information. Although improved versions exist (Geoburst+ [26], TrioVec [27]), we do not use them as baselines in this work as they are supervised approaches, while both Geoburst and our method use unsupervised approaches;…”

“…A common type of solution to the above problem takes the clustering based approach [3,10,12,[25][26][27][28], which generates a list of event candidates by clustering the tweets according to their semantic, spatial and temporal information, and then removes non-event clusters via supervised or unsupervised methods. In this work, we focus on how image analysis can be used to enhance the second step.…”

“…Note also that since the ground truth of the three datasets are not given, it is difficult to calculate the true recall. Therefore, we adopt a similar approach as in [26,27] and calculate the pseudo recall = N true /N total , where N true is the number of true events detected by a method, and N total is the number of true Insert t into the current node; 10 Embedding…”

“…This type of detection method takes a two-step approach [3,4,10,12,22,23,[25][26][27][28]. First, tweets are clustered based on their temporal, spatial, semantic, frequency and user information.…”

“…Finally, these candidates are ranked according to historical activities, and the top K events are returned. In terms of the improved versions: (1) Geoburst+ [26] adopts a supervised approach, and builds a candidate classification module, which learns the latent embeddings of tweets and keywords; then together with the activity timeline, the module extracts spatial unusualness and temporal burstiness to characterise each candidate event;…”