Detecting controversy in general web pages is a daunting task, but increasingly essential to efficiently moderate discussions and effectively filter problematic content. Unfortunately, controversies occur across many topics and domains, with great changes over time. This paper investigates neural classifiers as a more robust methodology for controversy detection in general web pages. Current models have often cast controversy detection on general web pages as Wikipedia linking, or exact lexical matching tasks. The diverse and changing nature of controversies suggest that semantic approaches are better able to detect controversy. We train neural networks that can capture semantic information from texts using weak signal data. By leveraging the semantic properties of word embeddings we robustly improve on existing controversy detection methods. To evaluate model stability over time and to unseen topics, we asses model performance under varying training conditions to test cross-temporal, cross-topic, cross-domain performance and annotator congruence. In doing so, we demonstrate that weak-signal based neural approaches are closer to human estimates of controversy and are more robust to the inherent variability of controversies.

The increasing volume of images published digitally requires social science and communication researchers to employ methods able to perform visual content analysis at a large scale. Ongoing advances in machine vision and the ability to automatically detect objects, concepts and features in images provide a promising opportunity to address this challenge, yet it is often not feasible for social science researchers to develop their own custom classifier given the volume of images, resources and technical expertise needed. We therefore propose a research protocol with which existing pre-trained (commercial) models can be used for theory-building purposes despite their black box approach.

We present INCA (short for INfrastructure for Content Analysis), a Python module for collecting, storing, processing, and analyzing a wide variety of media content, including but not limited to news, political debates, social media, forums, and customer reviews. Using Elasticsearch as a database backend and Celery for task management, it makes automated content analysis scalable. INCA's main objective is to enable and promote an integrated workflow. INCA focuses on re-usability of data, processors, and analyses; making all steps of automated content analysis (ACA) accessible to social scientists, without requiring advanced programming skills. Here, we present the aim, implementation and recommended workflow for INCA.

This paper sets out to detect controversial news reports using online discussions as a source of information. We define controversy as a public discussion that divides society and demonstrate that a content and stylometric analysis of these debates yields useful signals for extracting disputed news items. Moreover, we argue that a debate-based approach could produce more generic models, since the discussion architectures we exploit to measure controversy occur on many different platforms.

Social media are becoming increasingly important for communication between government organisations and citizens. Although research on this issue is expanding, the structure of these new communication patterns is still poorly understood. This study contributes to our understanding of these new communication patterns by developing an explanatory model of message diffusion on social media. Messages from 964 Dutch police force Twitter accounts are analysed using trace data drawn from the Twitter™ API to explain why certain police tweets are forwarded and others are not. Based on an iterative human calibration procedure, message topics were automatically coded based on customised lexicons. A principal component analysis of message characteristics generated four distinct patterns of use in (in)personal communication and new/versus reproduced content. Message characteristics were combined with user characteristics in a multilevel logistic general linear model. Our main results show that URLs or use of informal communication increases chances of message forwarding. In addition, contextual factors such as user characteristics impact diffusion probability. Recommendations are discussed for further research into authorship styles and their implications for social media message diffusion. For the police and other government practitioners, a list of recommendation about how to reach a larger number of citizens through social media communications is presented.