Crisis Detector is a tool which allows an early detection of an image crisis. It is being developed by scientists from our Faculty, and more specifically, from the Department of Artificial Intelligence.

Prof. Maciej Piasecki and Dr. Jan Kocoń from the Department of Artificial Intelligence are working on a tool that will help detect first signs of a situation potentially harmful to the image of a company, brand or a specific person presented in the media. As part of their scientific project, they are developing the "Crisis Detector" – a system for scanning the media in search for the early signals of an emerging image crisis.

– The aim of the project is to build a solution that would be based on new research in the field of machine learning and would automatically recognize such signals in media streams that indicate a potential image crisis threatening the security of the company. The system would also note which threads of the information feed are critical and whether the information is potentially fake news (FN ) – explains Dr. Jan Kocoń from the Department of Artificial Intelligence.

Tracking, detecting, analysing

The system will track news streams appearing in various media channels of the cyberspace. It will indicate texts related to the analyzed entities and recognize various linguistic anomalies which could indicate that their linguistic image is changing. The research goal, i.e. a system based on the developed AI model, will be achieved by completing individual research and implementation tasks. They will include, among others:

• preparing databases for annotation,
• designing an AI-based anomaly detection mechanism,
• developing machine learning methods for automatically describing text in terms of overtones, emotions, situation markers, temporal expressions, occurrences of proper names and text similarity to verified FNs,
• ensuring an appropriate level of model effectiveness for operating on real data.

The research result will be a validated prototype ready for implementation.

Support for PR specialists, politicians and companies

– Crisis Detector will be available both to customers from the areas of politics and public affairs (political parties, foundations, ministries, governmental and non-governmental organizations) and to commercial entities (production companies, service companies, consulting companies, PR and marketing agencies) – says Dr. Jan Kocoń. – The activity of such entities is related to and dependent on information present in cyberspace. If a potential crisis can be detected early and if it can be decided to be fake news (FN) or not, image security and image creation will be positively influenced.

The expected research results include the development of a tool which would support digital risk management, increase the level of cybersecurity and improve the ability to react quickly to the first signs of a crisis.

The result of the project will be an innovative service allowing crises detection in Polish-language media. It will be a unique solution that will automatically scan the media (i.e. press, radio, television, the Internet and social networks) and recognize signals suggesting that an entity is in a situation of a potential image crisis.

Multi-tasking, multi-functional…

Crisis Detector is a complex system that consists of several elements:

1. A system for detecting image crises in text streams of media messages. It will make it possible to track streams of texts appearing in various channels of cyberspace media in given periods and identify texts related to the investigated aspects (including entities, products, brands, situations, etc.). It will help to recognize anomalies appearing in the dynamically changing linguistic image of a person, institution or organization.
2. A module for extracting a stream of messages focused around a selected topic along with its vector-symbolic representation of the content. It will allow analyses of media life and media relations triggered by potential image crises.
3. A module for recognizing targeted overtones and emotions in the text in relation to the indicated semantic features of a text fragment, which will allow an emotive relationship between the sender and the indicated features to be determined and extracted from the text.
4. A system for the continuous monitoring of selected social channels in order to ensure the protection of social media accounts and to alert customers when syndromes of a crisis situation occur.

Watch out... fake news!

The system will not track fake news in order to expose them, rather it will treat them as a signal of an upcoming problem. The module responsible for checking whether a given piece of information is a verified fake news will provide a potential additional feature for predicting whether the entity monitored by Crisis Detector (e.g. a politician, company or organization) is facing an image crisis in the media.

– This crisis detection is based on the one hand on detecting anomalies (suddenly a lot more than usually is written on a given topic and negative messages start to grow in the comments), and on the other hand on the annotations of media monitoring experts – explains Dr. Jan Kocoń.

When ready, the service will be provided by the PSMM company, which deals with media monitoring. As an extension of the services, it will have a module for alerting about potential crises for a tracked, predefined topic (a person, brand, product, company, etc.).