Crowdsourcing, as one of the most promising techniques for distributed problem-solving, requires sustained human involvement. Therefore, it also brings new challenges to data management, fundamentally data input and its quality. In this paper, we looked at various forms of user motivations and quality control of crowdsourcing when building accessibility maps mobile applications. We discuss how motivations could be used to contribute to our accessibility maps scenarios, and how data can be improved for two types of participants: individual participants and organization participants. We identified three useful techniques for improving data quality: qualification-based, reputation-based, and aggregation-based. In addition, based on our own mobile application (named WEMAP), we evaluated our approaches through focus group discussions and in-depth interviews.
Information spreads at a pace never seen before on online platforms, even when this information is fake. Fake news can have substantial impact, for instance when it concern politics and influences the results of legislations or elections. Finding a methodology to verify if some piece of news is true or false is hence essential. In this work, we propose a methodology to create task-generic features that are paired with textual features in order to detect fake news. Task-generic features are created by elaborating on metadata attached to answers from Google's search engine, and by using crowdsourcing for missing values. We experimentally validate our method on a dataset for fake news detection based on the PolitiFact website. Our results show an improvement in F1-Score of 3% over the state of the art, which is significant for a 6-class task.
The Internet is nowadays a fantastic source of information thanks to the quantity of the information it provides and its dynamicity. However, these features also represent challenges when we want to consider trustworthy information only. On the Internet, the process of verifying information, known as factchecking, cannot be performed by human experts given the scale of the information that should be manually checked, and the speed to which it changes. In this paper, we propose an approach to evaluate the trustworthiness of online information modeled as RDF Triples. Given a use case, we select a specific ontology (in the following we use movie reviews as a use case) and match its object properties with WordNet. This allows us to understand, for each input triple, which class the subject and the object belong to. We associate SPARQL queries to each class, which are then used by our approach to search for additional evidences in Wikidata. By doing so, our approach generates feature vectors that are used by machine learning classification models to predict the trustworthiness of new input triples. Experiments on real movie data show that our approach provides results that are on par or better than the state of the art in fact checking.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.