How can yhumbnail images shape online profanity on web portals

Abstract

This research aims on investigating the cues of comment profanity. The facial salience level in the thumbnail image is especially a key concern of the paper, and hence, CNN based face recognition algorithm was applied to measure the facial salience of the image. Also, manually designed NLP algorithm, robust to noisy online user-generated text was used to determine the swearing level of comments with accuracy of 91.91%. By analyzing the massive online dataset from the web portal of South Korea, we found out that the ratio of faces in thumbnail image is negatively correlated with the average profanity score of comments in general. Depending on whether it is partisan news or not, however, the direction of its impact varied. For partisan news articles, the salient facial images succeeded in regulating harmful comments. On the other hand, it rather increased profanity comments in the case of non-partisan news including national security issues, Thus, this paper argues that facial cue of the media content could take a role as a catalyst for prosocial linguistic behavior in online space.