FactRank: Developing automated claim detection for Dutch-language fact-checkers

Fact-checking has always been a central task of journalism, but given the ever-growing amount and speed news offline online, as well growing amounts misinformation disinformation, it is becoming increasingly important to support human fact-checkers with (semi-)automated methods make their work more efficient. Within fact-checking, detection check-worthy claims crucial initial step, since limits number that require or deserve be checked for truthfulness. In this paper, we present FactRank, novel claim tool journalists specifically created Dutch language. To best our knowledge, first still only such Dutch. FactRank thus complements existing online tools English (a small of) other languages. performs similarly in ClaimBuster, state-of-the-art fact-checking English. Our comparisons baseline also indicate how much even expert disagree, there may natural “upper bound” on accuracy check-worthiness by machine-learning methods. The specific quality derives from interdisciplinary iterative process which was created, includes not high-performance deep-learning neural network architecture, principled approach defining operationalising concept via detailed codebook. This codebook jointly two countries have an official language (Belgium/Flanders Netherlands). We expect very useful exactly because way defined check-worthiness, made explicit traceable.