Using the CRiTERIA tools to carry out better risk assessment in the context of border security

Print
Email
  • The tools and methodologies, which have been developed in the CRiTERIA project, aim to assist analysts in carrying out risk analysis in the field of cross-border crime. It has to be noted here that CRiTERIA is a research project and that the resulting tools and methodologies are still prototypes, which has direct implications for their eventual adoption.
  • The tools have, however, been developed using professional, state of the art development methods. Still, adoption of these tools and methodologies into an operational environment requires some effort because they have to be adapted and further developed from advanced prototypes into products ready for adoption in a concrete context.
  • The main consideration for the adoption of the CRiTERIA tools for border security agencies or NGOs can be structured along the following 3 aspects:
    • Adaptation: CRiTERIA tools will be used in a very specific environment (e.g. border situation). For the best use of the different tools, important adaptations are required e.g. with respect to languages, sources and specific contents considered.
    • Integration: CRiTERIA tools will typically not be the only ones to be used for risk and threat analysis and therefore they would have to be adapted to work in synergy with existing methods and technologies.
    • Organisational embedding: As all new acquisitions, the CRiTERIA tools and methodologies would have to embedded into the organisational structures and processes.
Component Description Benefit Producer
Technical Components
Automatic Speech Recognition (ASR) ASR/Speech-To-Text models for languages relevant to CRiTERIA High quality automatic speech recognition, best for broadcast-news like contents, continuous adaptive vocabulary and language model adaptation (CAVA) HENSOLDT Analytics
Automatic Speech Recognition (ASR) ASR adaptation on selected languages, use of large pre-trained models for its deployment in target domain for target tasks (e.g., multi-task problem), use of novel parameter efficient fine-tuning technologies. Reusable models that can be considered for integration with other existing solutions such as the project platform from ROXANNE allowing to extend set of supported languages. Idiap Research Institute
Event Detection Component Component for detection of events from media content Multilingual event detection in both traditional and social media in the languages relevant to CRiTERIA. HENSOLDT Analytics
Narrative Detection Component Dashboard and data processing pipeline for the identification competing narratives. Automatic multisource narrative detection identifying storyline and framing. Enables the exploration of topic specific discussion threats within the documents, follow threats and identify new patterns. L3S Research Center
Author Verification Component Component for the detection and verification of authors in media content Author verification and detection of potential hijacked accounts in multiple languages. HENSOLDT Analytics
Stance Detection Component Risk Indicators/ Magnitude and likelihood assessment Multilingual social media stance detection. Special emphasis is given to generalizability to less-active users and new events. L3S Research Center
Image/Video Annotation and Explainable AI Component Image/Video Annotation and Explainable AI Component, including video event detection and annotation, logo detection and recognition, text-to-video retrieval, visual indicator annotation, and trend detection. Quality of annotation results. Provision of explanations for the results, allowing the detection of potential biases in the datasets/classifiers, and increasing the trust of the user in the results. CERTH - Centre for Research and Technology Hellas, Information Technologies Institute
Bias Detection Component Political bias detection Systematic bias detection for sources Idiap Research Institute
Risk and Threat Indicator Detection Component Social media post automatic classification into risk indicators. The component is based on a list of keywords identified by experts and augmented using large language models. A machine learning model is trained to generalize the concepts behind the risk indicators. Identification and aggregation of evidence supporting the set of predefined indicators without relying on keywords in the documents. L3S Research Center
News Source Reliability Scoring Reliability score of an individual domain or a list of domains, that reflects their factuality of reporting. Reliability of news media robust to content-related biases since they are both language and content independent (political, religious, racial, etc.). Idiap Research Institute
News Media Political Bias Component that computes political bias score for news media Robust model to calculate political bias score of news media websites. Idiap Research Institute
Explainable Data Validation and Risk Linkage
Image/Video Verification Component Media verification tool for media item validation and gaining additional insight into the sources participating in a specific topic and the relative support for the researched claims. Quality of verification results. Provision of localization information (on the image grid) that accompany the verification outcome, and possibly additional spatial explanations. CERTH - Centre for Research and Technology Hellas, Information Technologies Institute
Fact Checking Component AI/ML module trained on large amount of textual data. As part of the module, the reliability source data analyser is developed. A highly demanded component which requires an iterative re-adaptation and updates. Idiap Research Institute
Semi-Automatic Evidence Validation Tool Component supporting risk analysis, especially validation Risk Analysis Support tool and dashboard for managing and validating threads, vulnerability, and impact. L3S Research Center
Risk Interaction and Propagation Tool Component supporting risk analysis, especially considering risk interaction Identification of risk co-occurrences and interactions based on automatic evidence extraction. L3S Research Center
Platform / Dashboard (accessible/integrated with GUI)
WebLyzard Dashboard / CRiTERIA Visual Analytics The CRiTERIA Dashboard is a visual analytics interface that provides access and insights into the online debate around border intelligence-related topics, ranging from user-generated content to Web content and open-access news coverage. Powerful aggregation, segmentation, search and drill-down features integrated into a responsive and interactive user interface to provide actionable insights from public debate and stakeholder communication. webLyzard technology
CREST CRiTERIA Risk Evaluation Support Tool (CREST) offers several analyses and visualizations aggregated around expert-defined risk indicators. This tool focuses on multimodal evidence analysis and intelligence representation. L3S Research Center
News Search News search with fact-checking, reliability and political bias of source estimation Information verification tool in the context of news articles with fact-checking, reliability and political bias of source estimation Idiap Research Institute
Video Search Free-text video search interface. Exploitation of Image/Video Annotation results and Explainable AI Components for analysing and searching for content. Video searching using cutting-edge tools developed in WP5 for Image/Video Annotation. CERTH - Centre for Research and Technology Hellas, Information Technologies Institute
Filter for target audience

MIRROR has received funding from the European Union’s Horizon 2020 research and innovation action program under grant agreement No 832921.

CRiTERIA has received funding from the European Union’s Horizon 2020 research and innovation action program under grant agreement No 101021866.

© All rights reserved

Imprint | Privacy Policy