Duluth at SemEval-2019 Task 4: The Pioquinto Manterola Hyperpartisan News Detector

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 949–953 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 949 Duluth at SemEval-2019 Task 4: The Pioquinto Manterola Hyperpartisan News Detector Saptarshi Sengupta and Ted Pedersen Department of Computer Science University of Minnesota Duluth, MN 55812, USA {sengu059,tpederse}@d.umn.edu Abstract This paper describes the Pioquinto Manterola Hyperpartisan News Detector, which partici- pated in SemEval–2019 Task 4. Hyperpartisan news is highly polarized and takes a very bi- ased or one–sided view of a particular story. We developed two variants of our system, the more successful was a Logistic Regression classifier based on unigram features. This was our official entry in the task, and it placed 23rd of 42 participating teams. Our second variant was a Convolutional Neural Network that did not perform as well. 1 Introduction Social media has become a vital source of news for many people. It makes it possible to share use- ful information widely and in a timely fashion, and yet can also be misused to spread biased, mislead- ing, or dangerous content. Hyperpartisan news is a particular worry in that it is premised on absolute allegiance to one par- ticular point of view, and seeks to reinforce po- tentially misinformed opinions held by its read- ers. This has led to very real consequences in this world. A tragic example can be found in Myan- mar, where Buddhist ultranationalists relied on so- cial media to spread hyperpartisan and fake news in order to promote hatred and violence against different Muslim communities (Fink, 2018). While related, Hyperpartisan news is not the same as fake news. The former shows a high de- gree of bias, whereas the latter is more so an out- right fabrication. However, the techniques applied to detecting both are similar. For example, P´erez- Rosas et al. (2018) detected fake news by training Support Vector Machines using ngrams, punctua- tion, and measures of readability. (Tacchini et al., 2017) used likes of articles as features for building a Logistic Regression classifier for fake news de- tection. Potthast et al. (2018) identified hyperpar- tisan news through the use of style and readability features, and also employed a technique known as unmasking (Koppel et al., 2007) to distinguish be- tween hyperpartisan and mainstream news. 2 Task Description SemEval–2019 Task 4 (Kiesel et al., 2019) chal- lenged participants to detect whether an article is hyperpartisan (H) or mainsteam (M). As such it represents a binary classification task. The task or- ganizers provided training data, and so we elected to take a supervised learning approach. There were two datasets provided by the orga- nizers (Kiesel et al., 2018). The by-article data is a smaller corpus of 645 news articles that have been manually assigned to H (238 articles) or M (407 articles). There was also the much larger by-publisher data set with 750,000 articles where classifications were made based on the source of an article. Making classifications in this way is possible since certain publishers are known to …

Content truncated to 500 words