You are here

Commenting Architectures on German News Websites

The next speaker in this session at AoIR 2018 is Christian Strippel, whose focus is on the discourse architecture of German news Websites. The background to this work is a project to develop the tools to automatically detect and mitigate hate speech in comment sections in such sites.

Such hate speech is common in comments sections on such sites, and news publishers have a variety of mechanisms for addressing this, from netiquettes and forum rules through human moderators to automated moderation systems. The technical frameworks and affordances of such commenting spaces play a crucial role here as well, of course, and there are many studies that compare these discourse architectures across sites, but don’t go into much detail as they do so.

Components of such discourse architectures include registration, comment sorting, anonymisation, comment evaluation, reporting of problematic comments, paywalls, topic restrictions, comment closures after some time, character limits, and staff picks of ‘quality’ comments; the presence of these elements has been compared at a whole-of-site level across selected sites.

The present study examined this across some 361 German news sites from a variety of cities, including newspapers, magazines, TV and radio broadcasters, and other sites. The affordances of comment sections were grouped in three dimensions: restrictions to commenting, hurdles before commenting, and moderation after commenting.

Some 48% of all sites had comment sections (95% also had Facebook pages, where presumably commenting might also be possible); 8% offered discussion fora; and 4% offered both; 46% offered neither. 26% of sites restricted comments to topics; 9% closed comment sections after a certain period; and 19% limited the amount of characters per comment. 54% required a separate registration; 9% operated a paywall; and 16% required some form of identification (e.g. email). 5% sorted comments algorithmically; 37% allowed commenters to rate other comments; 40% had a reporting function for inappropriate content; only two sites had an ‘editor’s pick’ comment.

The patterns in these affordances give rise to five broad types of discourse architectures. One type is controlled across all three dimensions (restrictions, obstacles, and regulations); a second is managed but provides no obstacle to access; a third is community-based and has no restrictions, but obstacles and regulations; a fourth is free access, and only offers some regulations; and the fifth only has some obstacles to participation but is not otherwise controlled. Of these, the first three are mainly employed by newspapers; the fourth by local Websites; and the fifth by local radio stations.

Overall, then, half of German news Websites offer comment sections, but almost all have Facebook pages, so Facebook may now be the main discourse architecture for them. Most offer no or few restrictions to commenters, perhaps because heavily discussed articles are good for business; most are also (still) very inclusive, indicating perhaps the high relevance of occasional commenters. There is still a high potential for more (self-)regulation, but this comes at a cost.