A Deep Learning Approach for Multimodal Deception Detection

Cyberbullying describes bullying using mobile phones and the internet. Most previous studies have focused on the prevalence of text message and email bullying. Two surveys with pupils aged 11-16 years: (1) 92 pupils from 14 schools, supplemented by focus groups; (2) 533 pupils from 5 schools, to assess the generalisability of findings from the first study, and investigate relationships of cyberbullying to general internet use. Both studies differentiated cyberbullying inside and outside of school, and 7 media of cyberbullying. Both studies found cyberbullying less frequent than traditional bullying, but appreciable, and reported more outside of school than inside. Phone call and text message bullying were most prevalent, with instant messaging bullying in the second study; their impact was perceived as comparable to traditional bullying. Mobile phone/video clip bullying, while rarer, was perceived to have more negative impact. Age and gender differences varied between the two studies. Study 1 found that most cyberbullying was done by one or a few students, usually from the same year group. It often just lasted about a week, but sometimes much longer. The second study found that being a cybervictim, but not a cyberbully, correlated with internet use; many cybervictims were traditional 'bully-victims'. Pupils recommended blocking/avoiding messages, and telling someone, as the best coping strategies; but many cybervictims had told nobody about it. Cyberbullying is an important new kind of bullying, with some different characteristics from traditional bullying. Much happens outside school. Implications for research and practical action are discussed.

We consider the automated recognition of human actions in surveillance videos. Most current methods build classifiers based on complex handcrafted features computed from the raw inputs. Convolutional neural networks (CNNs) are a type of deep model that can act directly on the raw inputs. However, such models are currently limited to handling 2D inputs. In this paper, we develop a novel 3D CNN model for action recognition. This model extracts features from both the spatial and the temporal dimensions by performing 3D convolutions, thereby capturing the motion information encoded in multiple adjacent frames. The developed model generates multiple channels of information from the input frames, and the final feature representation combines information from all channels. To further boost the performance, we propose regularizing the outputs with high-level features and combining the predictions of a variety of different models. We apply the developed models to recognize human actions in the real-world environment of airport surveillance videos, and they achieve superior performance in comparison to baseline methods.