Scope

The MuseAI solution will use advanced natural language processing (NLP) techniques to identify factual claims within textual content across multiple languages. This capability will enable fact-checkers to efficiently scan texts of variable lengths to determine and locate the presence of claims that need further verification. The development of this component requires designing, implementing, training, and evaluating the necessary ML and NLP models for claim detection, making use of Natural Language Inference and Stance detection multilingual models, as well as evaluating and benchmarking the developed models. Although we will be focused on text, the goal is to facilitate extension to further languages and other information modalities, such as audio, through audio-to-text transcription. We will depart from multilingual models to ensure reaching the largest possible market share.

Once a claim has been identified, it must be assessed and generate an output that establishes whether it is a true or false fact (or that there is insufficient evidence to make it so). To build this component, we will develop semantic similarity and natural language inference models to compare candidate claims against existing ones in a database of fact-checked facts. In this way, we ultimately rely on the work of the fact-checking entities to produce reliable and up-to-date outputs.

The MuseAI solution will support multiple information modalities, including text, audio, and video, by incorporating automated transcription capabilities using deep learning models. This versatility allows fact-checkers to analyse a wide range of content formats effectively, enhancing the comprehensiveness of the fact-checking process.

An intuitive and user-friendly interface will be developed to provide fact-checkers and journalists with seamless access to MuseAI’s functionalities. The interface will facilitate easy navigation, efficient data input, and insightful visualisation of results, empowering users to interact with the solution effectively and derive actionable insights from the detected claims.

The project scope includes evaluation of the developed solution in terms of accuracy, performance, and usability. Evaluation activities will assess the effectiveness of AI models in claim detection and matching, as well as gather feedback from end-users to refine the solution’s design and functionality iteratively.