This paper reports on a writing style analysis of hyperpartisan (i.e., extremely onesided) news in connection to fake news. It presents a large corpus of 1,627 articles that were manually fact-checked by professional journalists from BuzzFeed. The articles originated from 9 well-known political publishers, 3 each from the mainstream, the hyperpartisan left-wing, and the hyperpartisan right-wing. In sum, the corpus contains 299 fake news, 97% of which originated from hyperpartisan publishers.We propose and demonstrate a new way of assessing style similarity between text categories via Unmasking-a meta-learning approach originally devised for authorship verification-, revealing that the style of left-wing and right-wing news have a lot more in common than any of the two have with the mainstream. Furthermore, we show that hyperpartisan news can be discriminated well by its style from the mainstream (F 1 = 0.78), as can be satire from both (F 1 = 0.81). Unsurprisingly, stylebased fake news detection does not live up to scratch (F 1 = 0.46). Nevertheless, the former results are important to implement pre-screening for fake news detectors.

Hyperpartisan news is news that takes an extreme left-wing or right-wing standpoint. If one is able to reliably compute this meta information, news articles may be automatically tagged, this way encouraging or discouraging readers to consume the text. It is an open question how successfully hyperpartisan news detection can be automated, and the goal of this SemEval task was to shed light on the state of the art. We developed new resources for this purpose, including a manually labeled dataset with 1,273 articles, and a second dataset with 754,000 articles, labeled via distant supervision. The interest of the research community in our task exceeded all our expectations: The datasets were downloaded about 1,000 times, 322 teams registered, of which 184 configured a virtual machine on our shared task cloud service TIRA, of which in turn 42 teams submitted a valid run. The best team achieved an accuracy of 0.822 on a balanced sample (yes : no hyperpartisan) drawn from the manually tagged corpus; an ensemble of the submitted systems increased the accuracy by 0.048.

No abstract

The segmentation of an argumentative text into argument units and their nonargumentative counterparts is the first step in identifying the argumentative structure of the text. Despite its importance for argument mining, unit segmentation has been approached only sporadically so far. This paper studies the major parameters of unit segmentation systematically. We explore the effectiveness of various features, when capturing words separately, along with their neighbors, or even along with the entire text. Each such context is reflected by one machine learning model that we evaluate within and across three domains of texts. Among the models, our new deep learning approach capturing the entire text turns out best within all domains, with an F-score of up to 88.54. While structural features generalize best across domains, the domain transfer remains hard, which points to major challenges of unit segmentation.

No abstract

No abstract

Web reviews have been intensively studied in argumentation-related tasks such as sentiment analysis. However, due to their focus on content-based features, many sentiment analysis approaches are effective only for reviews from those domains they have been specifically modeled for. This paper puts its focus on domain independence and asks whether a general model can be found for how people argue in web reviews. Our hypothesis is that people express their global sentiment on a topic with similar sequences of local sentiment independent of the domain. We model such sentiment flow robustly under uncertainty through abstraction. To test our hypothesis, we predict global sentiment based on sentiment flow. In systematic experiments, we improve over the domain independence of strong baselines. Our findings suggest that sentiment flow qualifies as a general model of web review argumentation.

This paper proposes a shared task for the identification of the argumentative structure in newspaper editorials. By the term "argumentative structure" we refer to the sequence of argumentative units in the text along with the relations between them. The main contribution is a large-scale dataset with more than 200 annotated editorials, which shall help argumentation mining researchers to evaluate and compare their systems in a standardized manner. The paper details how we model and manually identify argumentative structures in order to build this evaluation resource. Altogether, we consider the proposed task as a constructive step towards improving writing assistance systems and debating technologies.