Can We Predict a Riot? Disruptive Event Detection Using Twitter
read more
Citations
Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies
Breaking News Detection and Tracking in Twitter.
Monitoring the public opinion about the vaccination topic from tweets analysis
Using AI and Social Media Multimodal Content for Disaster Response and Management: Opportunities, Challenges, and Future Directions
Iktishaf: a Big Data Road-Traffic Event Detection Tool Using Twitter and Spark Machine Learning
References
Latent dirichlet allocation
Latent Dirichlet Allocation
The anatomy of a large-scale hypertextual Web search engine
The Anatomy of a Large-Scale Hypertextual Web Search Engine.
Term Weighting Approaches in Automatic Text Retrieval
Related Papers (5)
Frequently Asked Questions (18)
Q2. What future works have the authors mentioned in the paper "Can we predict a riot? disruptive event detection using twitter" ?
There are many directions for future work. Finally, the detection of rumors in the social media, the analysis of the distinctive characteristics of rumors and the way in which they propagate in the microblogging communities will be addressed in the future. Spammer detection in various online social networking platforms is another interesting task that is reserved for future work. The authors intend to further evaluate the summarization output to not only map onto real events, but to provide qualitatively useful output for decision making.
Q3. What are the classification algorithms used in the experiment?
The classification algorithms used in the experiment were: Naive Bayes [Lewis 1998] a statistical classifier based on the Bayes’ theorem; Logistic Regression [Friedman et al. 1998], a generalized linear model to apply regression to categorical variables; and support vector machines (SVMs) [Joachims 1998] which aims at maximizing (maximum margin) the minimum distance between two classes of data using a hyperplane that separates them.
Q4. What are the discriminative features of the NDCG?
The near-duplicate measure, the favourite ratio and the positive sentiment ratio are the least discriminative features, which suggest that they appear in all different types of posts, not only in disruptive events.
Q5. How can it be used to detect large and small events?
using an online clustering algorithm with a sliding window timeframe, it can be utilised to detect large and small-scale events from social media streams - with particular attention to filtering from large to small-scale events.
Q6. What is the proposed framework for detecting events?
Their proposed framework is based on collecting data over time windows for a given location which supports the automatic detection and summarization of events from social media.
Q7. What is the main challenge of social media?
One challenge is that online posts are often constrained in length (referred to as microblogs), which means that only a small amount of text is available to be analysed to gain insights.
Q8. What is the effect of using supervised classification of each tweet before clustering?
Employing supervised classification of each tweet before clustering (large scale event detection) reduces the computational overhead at the clustering stage as the number of tweets is significantly reduced (containing only event-related tweets).
Q9. How has large-scale event detection been explored?
Large-scale event detection has also been explored through clustering of discrete wavelet signals built from individual words generated by Twitter [Weng and Lee 2011].
Q10. What is the role of social media in disasters?
Research in recent years has uncovered the increasingly important role of utilising data from social networking sites in disaster situations, and shown that information broadcast via social media can enhance situational awareness during a crisis situation [Alsaedi et al.
Q11. What is the effect of the removal of posts that were less than 3 words long?
posts that were less than 3 words long were removed, as were messages where over half the total words were the same word, since these posts were less likely to have useful information.
Q12. What are the main features of their approach?
Their contributions can be summarized as follows:—Using temporal, spatial and textual features, their approach is able to detect small-scale events in agiven place and time better than existing algorithms, to which the authors compare their performance results;
Q13. Why did the authors choose to use an online clustering algorithm?
The decision to use an online clustering algorithm was taken for three main reasons: (i) it supports high dimensional data as it effectively handles the large volume of social media data produced around events; (ii) many clustering algorithms such as K-means require previous knowledge of the number of clusters.
Q14. What are the advantages of using a clustering algorithm?
Thus clustering (small-scale event detection), feature selection and summarization are much faster and suitable for real-time analysis.
Q15. What is the method for evaluating the clustering algorithm?
The authors evaluate the algorithm’s performance on the training data using a range of thresholds, and identify the threshold setting that yields the highest-quality solution according to a given clustering quality metric (here the authors implement the f-measure).
Q16. What is the way to obtain a good performance of the textual feature model?
Using the textual feature model, the authors are still able to obtain a reasonable performance of on average, 40% content about an event, provides situational awareness information about that event.
Q17. How did the researchers find the way to detect a riot?
Their experiments suggest that their framework yields better performance than many leading approaches in real-time event detection, and using a real-world ground truth published by the Metropolitan Police Services (MPS) after the 2011 riots in England, the authors showed their system to detect events far quicker than they were reported to MPS.
Q18. How did the authors propose to combine the summarization of tweets?
More fine-grained summarization was proposed by considering sub-events detection and combining the summaries extracted from each sub-topic (tweet selection, tweet ranking) [Shen et al. 2013; Yajuan et al. 2012; Zubiaga et al. 2012].