The Council for Scientific and Industrial Research (CSIR) on Monday detailed work researchers are doing to put social media to use in identifying public safety incidents, including traffic collisions, crime incident patterns, emergencies, terrorism incidents and pandemics, and to develop tools that can use these insights to improve public safety.
At a media briefing on Monday, acting CSIR Research Group Leader for Data Science Dr Vukosi Marivate highlighted that the work focused on identifying patterns in crime and public safety incident posts on social media and using that information for more effective public safety management.
The initiative, ‘Data Science for Public Safety’, led by the CSIR’s data science research group, uses machine learning, artificial intelligence systems, natural language and behaviour modelling to mine social media posts to build intelligent and responsive systems to support public and private organisations to more effectively allocate resources and to respond more effectively to public safety incidents.
“The aim is to lead to real-time, on-the-ground awareness for when incidents occur. A challenge South Africa faces is that we typically only get quarterly public safety data from the South African Police Service and other organisations.
“Researchers need more granular data than this. Social media presents an opportunity to identify when someone posts something about public safety and to use these available data points to build analytics tools to extract insight from this.”
Marivate showed a Twitter post that described a smash-and-grab incident in Bree street (Lilian Ngoyi street), in Johannesburg, which includes an indicator of the time it occurred – half an hour before the tweet – and the location, as well as the incident.
While social media is not effective in combating individual crimes, patterns can be identified that can lead to more effective policing.
Additionally, because the public safety analytics systems run in real-time, it can help users, such as the police, to identify short-term trends and potentially catch a criminal or a gang conducting a crime spree.
The analytics tools require natural language processing to identify information from the posts.
The team also taps into information often posted on social media by public safety organisations, such as Arrive Alive and emergency services. The public safety tools can also process short-hand notes posted on social media, which is often used by these organisations.
“The Data Science for Public Safety initiative looked at the use case of social media data. Important for any public safety tool is the context of the incident, which can be filled in with social media information, where available.”
Marivate noted that these tools would only be effective in areas with high social media use, typically urban areas, and are not as effective in rural areas.
The work over the past few years has aimed to identify whether the post is a description of a crime or not and then use machine learning and artificial intelligence to disambiguate the information available to provide useful information.
“[It] is interesting to determine how we can use data science for science and government applications. At the start, we labelled posts ourselves and then used publicly available data to train the predictive models to improve performance.”
Further, the work is compelling because it not only aims to classify when a post describes an incident and identify normal topics related to crime or public safety, such as traffic incidents, but to identify unique information and build a system that can automatically identify unique information, explained Marivate.
“The work we have been doing over the past year has focused on differentiating between crime and public safety in social media posts, which can look very similar. Determining the context is difficult; however, the models can effectively differentiate between the different categories. Mostly the posts are about a hijacking, and/or recovery of a hijacked vehicle, traffic lights that are out (possibly posing a public safety risk) or a crash on the highway, for example.
“What this has allowed us to do is to build the capability. We are working with law enforcement and we have access to data from the City of Tshwane. We contributed to the Tshwane Safety App, which has been out for two years, and helped build the analytics for them to better understand reports over the app and allow the reports to be used.”
The CSIR is also working with the Tshwane Metro Police to do data resource planning to determine what has entered their call centre, read and analyse text messages and make the systems responsive to data that comes in and is fed back in from the analytics engines.