Real-Time Detection of Violence and Extremism from Social Media

Lead Research Organisation: Aston University
Department Name: Sch of Engineering and Applied Science


The explosive use of social media tools in recent years has turned them into a double-edged sword. On one hand, social media is viewed as a positive factor in Middle East revolutions. On the other hand, violence events such as the UK riots occurred in August this year appeared to be driven by the use of social media. The proposed project represents a timely development of intelligent systems for addressing the emerging defense and security issues arising from social media.

The proposed research describes a new approach to detecting trends of violent radicalization and extremism from social media. In particular, it proposes a Bayesian modeling approach which detects violence contents from social media without the use of any labelled data. Words indicating violence, anger, hate, racism, etc. are naturally incorporated as prior knowledge into the model learning process. Efficient online parameter updating and parallel data processing procedures will be investigated. This proposal falls into the area of "Extracting meaningful information" listed in the original call for proposals. It particularly aims to tackle the technical challenge of real-time processing of large-scale social media data for early detection of violent extremism from text. The results of the research are potentially very important to society as they aim to enable the deployment of the forces of law to prevent violent events.

Planned Impact

This research will have a great impact on enhancing the national security capability by enabling real-time detection of potential threats such as violence and extremism from social media. Online social networking and micro-blogging service have enabled users to share and diffuse information to a tremendous number of audiences within very short time. While such a new form of service has proven to have major social-economic benefits, it is also at the same time under the risk of being used for violent and criminal activities.

This project therefore aims to develop efficient computational tools for detecting violent radicalization and extremism from social media, which will ultimately help improving the national security capability with the online monitoring function offered by the system. Specifically, the tools seek to detect and extract topics relating to violent and criminal activity from large-scale social media data in real-time, and constantly track any events that are identified suspicious. Owing to the fast-evolving nature of social media, such a system will be potentially very important for the forces of law to respond to and deal with the potential security risks timely which will consequently help improve public security.

Although targeting the defense and security sector, the developed tools and algorithms of the project can be applied to other public and private sectors, and be easily transferred into commercial use. For public sectors, it is possible to use the tools to analyse citizen's opinions on hot issues such as welfare reforms, Eurozone crisis, new policies introduced by the government, etc. In the ongoing "Occupy London" movement, our tools can be used to monitor what people are tweeting about and their sentiment and debates towards their most concerned issues.

For private sectors, the tools can be used to detect and monitor the trends of topics being discussed about products or services, allowing business organizations to gauge insights into the potential changes in the market opinions and hence to adapt their business strategies accordingly. As the social media data published by individuals have strong correlations with their behaviors, deploying the tools for individual data analysis can help to address social issues such as preventing adolescent suicide by recognising the suicidal thoughts expressed in social media.

We plan to disseminate our research findings in several ways. First, information about the scientific work will be published in a regularly updated project website. A dedicated online forum will be provided for engaging with interested parties including the public as well as potential researchers and collaborators (academic and non academic). Research results will also be disseminated at public engagement events such as the networking events with potential business partners. Such events help in building connections with industry and mapping the developed technologies with industrial R&D needs.


10 25 50

Related Projects

Project Reference Relationship Related To Start End Award Value
EP/J020427/1 30/05/2012 13/01/2013 £113,921
EP/J020427/2 Transfer EP/J020427/1 14/01/2013 13/06/2013 £40,939
Description Social streams have proven to be the most up-to-date and inclusive information on current events. In this project we have proposed a novel probabilistic modelling framework, called violence detection model (VDM), which enables the identification of text containing violent content and extraction of violence-related topics over social media data. The proposed VDM model does not require any labeled corpora for training, instead, it only needs the incorporation of word prior knowledge which captures whether a word indicates violence or not. We have proposed a novel approach of deriving word prior knowledge from external knowledge sources such as DBpedia. We have also developed a fast learning procedure to training the VDM model from social stream data.
Exploitation Route The proposed Violence Detection Model has been integrated into a joint demonstrator developed together with Edinburgh, Glasgow, Loughborough and Sheffield for real-time detection and tracking of security-related events from Twitter.
Sectors Communities and Social Services/Policy,Digital/Communication/Information Technologies (including Software),Government, Democracy and Justice,Security and Diplomacy

Description The outcome generated from the grant has resulted in another R&D project funded by Innovate UK in collaboration with Folding Space, a company specialising in document management systems. The model extended has been used in topic extraction from NHS patient records.
Sector Digital/Communication/Information Technologies (including Software),Healthcare
Impact Types Economic

Title Weakly-supervised Bayesian models for security-related topics detection from social media 
Description A weakly-supervised Bayesian model has been proposed which can automatically identify security-related topics from social media without the use of labelled data. Also, fast learning and parallel model learning methods have been implemented. 
Type Of Material Computer model/algorithm 
Provided To Others? No  
Impact A further EPSRC funding has been attracted to integrate our proposed model into a joint demo system for real-time event detection, tracking, interpretation and monitoring in social media. 
Description Key Technologies for Information Security Research 
Organisation Chinese Academy of Sciences
Department Institute of Information Engineering
Country China 
Sector Academic/University 
PI Contribution The International Partnership Programme was funded by the Chinese Academy of Sciences (CAS). It invited 5 to 8 international researchers to join a project funded by the CAS. I am the only UK research invited to join the international collaboration team hosted in the Institute of Information Engineering of the CAS. The project involves the collaboration with Chinese researchers as well as international renowned experts from top universities in the US including Princeton University, University of Southern California etc. I will mainly contribute towards the areas of social media analysis and sentiment analysis. CAS will fund my visit to the Institute of Information Engineering for at least one month each year.
Start Year 2013
Description Talk in British Science Festival in September 2014 
Form Of Engagement Activity A talk or presentation
Part Of Official Scheme? No
Geographic Reach National
Primary Audience Public/other audiences
Results and Impact The talk has attracted more than 70 participants. It has attracted much interest in the area of social media analysis.

After the talk, there have been several people asking about accessing the information about my research.
Year(s) Of Engagement Activity 2014