Home Aims & Objectives NaCTeM Services Text Mining Tools Text Mining Infrastructures Text Mining for Biodiversity Resources Terms & Conditions FAQ News & Events News What others are saying about us Press and Journal Mentions NaCTeM Seminars People Projects Publications Community External Collaboration Vacancies Teaching & Tutorials Feedback How to find us

Talk at Digital Trust and Security Workshop Series

2019-12-03

Dr. Chrysoula Zerva, postdoctoral researcher at NaCTeM, will give a talk entitled "Assessing credibility of health & science news via scientific literature" , as part of the the series of workshops on Digital Trust and Security, organised by the University of Manchester's Digital Futures interdisciplinary network. The workshop will take place from 11:30 to 12:30 on Wednesday 11th December in Room 2.07 of the Humanities Bridgeford Street Building at the University of Manchester.

Abstract:

News on scientific advances, health issues and technological miracles have always been among the most viral and controversial ones, both because of the potential impact on people's lives and because often, the communicated knowledge is complex and hard to verify. Much like political news, health and science news often contain exaggerated statements, misinterpreted statistics and information taken out of context. Thus evaluating the trustworthiness of an article and the credibility of provided information is an involved procedure, requiring both evaluation of the scientific literature and comparison with the information in the original article. Such a process is often too laborious and time consuming to be manually performed by experts, and too demanding for the lay readers. Thus, leveraging information from scientific literature and using it to assess science and health news articles in an automated way could alleviate the issue and help readers and authors alike.

We present some initial efforts on mapping news information to scientific information, using a mixture of deep learning and traditional machine learning approaches. We discuss points of failure for the initial architecture, and propose a new end-to-end framework. The latter uses graph embeddings to achieve an optimal matching between context of sentences across documents and a multi-task architecture, aiming to use heterogeneous information in order to predict the credibility of a news article. We show how we aim to take advantage of manually scored articles for credibility (healthfeedback.org). We also discuss a crowdsourcing plan for further annotations to support deep learning for credibility identification in science news based on BBC articles.

More information...
http://www.digitalfutures.manchester.ac.uk/events/events/dts-workshop-series-chrysoula-zerva-mustafa-a-mustafa/