A large number of individuals are suffering from suicidal ideation in the world. There are a number of causes behind why an individual might suffer from suicidal ideation. As the most popular platform for self-expression, emotion release, and personal interaction, individuals may exhibit a number of symptoms of suicidal ideation on social media. Nevertheless, challenges from both data and knowledge aspects remain as obstacles, constraining the social media based detection performance. Data implicitness and sparsity make it difficult to discover the inner true intentions of individuals based on their posts. Inspired by psychological studies, we build and unify a high-level suicide-oriented knowledge graph with deep neural networks for suicidal ideation detection on social media. We further design a two-layered attention mechanism to explicitly reason and establish key risk factors to individual's suicidal ideation. The performance study shows that: 1) with the constructed personal knowledge graph, the social media based suicidal ideation detection can achieve over 93% accuracy; and 2) among the six categories of personal factors, post, personality, and experience are the top-3 key indicators. Under these categories, posted text, stress level, stress duration, posted image, and $ruminant thinking$ contribute the most to one's suicidal ideation detection.
Stress has become an increasingly serious problem in the current society, threatening mankind’s well-beings. With the ubiquitous deployment of video cameras in surroundings, detecting stress based on the contact-free camera sensors becomes a cost-effective and mass-reaching way without interference of artificial traits and factors. In this study, we leverage users’ facial expressions and action motions in the video and present a two-leveled stress detection network (TSDNet). TSDNet firstly learns face- and action-level representations separately, and then fuses the results through a stream weighted integrator with local and global attention for stress identification. To evaluate the performance of TSDNet, we constructed a video dataset containing 2092 labeled video clips, and the experimental results on the built dataset show that: (1) TSDNet outperformed the hand-crafted feature engineering approaches with detection accuracy 85.42% and F1-Score 85.28%, demonstrating the feasibility and effectiveness of using deep learning to analyze one’s face and action motions; and (2) considering both facial expressions and action motions could improve detection accuracy and F1-Score of that considering only face or action method by over 7%.
Objective: Coronavirus disease 2019 (COVID-19) has caused substantial panic worldwide since its outbreak in December 2019. This study uses social networks to track the evolution of public emotion during COVID-19 in China and analyzes the root causes of these public emotions from an event-driven perspective. Methods: A dataset was constructed using microblogs (n = 125,672) labeled with COVID-19-related super topics (n = 680) from 40,891 users from 1 December 2019 to 17 February 2020. Based on the skeleton and key change points of COVID-19 extracted from microblogging contents, we tracked the public’s emotional evolution modes (accumulated emotions, emotion covariances, and emotion transitions) by time phase and further extracted the details of dominant social events. Results: Public emotions showed different evolution modes during different phases of COVID-19. Events about the development of COVID-19 remained hot, but generally declined, and public attention shifted to other aspects of the epidemic (e.g., encouragement, support, and treatment). Conclusions: These findings suggest that the public’s feedback on COVID-19 predated official accounts on the microblog platform. There were clear differences in the trending events that large users (users with many fans and readings) and common users paid attention to during each phase of COVID-19.
27 June 2022 in IEEE Transactions on Information Forensics and Security
Automated deception detection has been found as a vital and concerned task, capable of assisting human users to assess truthfulness and detect deceptive behaviors in several situations (e.g., medical, legal, as well as occupational domains). As contact-free video cameras and data analysis techniques are leaping forward, leveraging one’s visual, acoustic and textual information captured in a video can be cost-effective for deception detection. In this study, we aim at a fine-grained question-level deception detection task, focusing on detecting whether a subject lies or not in answering each question rather than providing an overall “lie or not” judgement. A Graph-based Cross-modal Fusion Model (GCFM) is presented to learn the inherent associations among the subject’s reactions to different questions, plus a novel cross-modal attention mechanism to enhance the model’s learning capability. As revealed by the experimental results on the two datasets, the proposed graph-based GCFM outperformed eight other methods, and its two alternatives (clustering K-means based and sequential learning LSTM based methods), achieving accuracy 88.14% and 86.91% on the two datasets, respectively. Besides, through association learning, GCFM could increase the accuracy by 1.87% and 4.33% on the two datasets, respectively. Furthermore, its cross-modal attention mechanism led to the improvement of accuracy by 2.44% and 2.95% on the two datasets, respectively.
28 September 2021 in IEEE Transactions on Affective Computing
Suicide has become a serious problem, hurting the well-being of human society. Thanks to social media, from peopleslinguistic posts, suicide risk detection has achieved good performance. The aim of this study is to investigate whether more significantaccuracy could be achieved. Motivated by the observation that the prior solutions strived to detect suicide risk based on users explicitouter post expressions on social media, and no attempt was made to infer users inner true thoughts and emotion changes from theirnormal open posts for suicide risk detection, we propose to firstly learn the correlations between users normal open posts and hiddencomments, trying to understand users inner true thoughts and emotion changes from his open posts, and then detect users suiciderisk upon the generated intermediate results. The performance study on the microblog dataset (3,652 at-risk microblog users and 3,652ordinary microblog users) and forum dataset (392 at-risk forum users and 108 ordinary forum users) shows that, based on the inferredhidden thoughts and inner emotion changes, the proposed approach can achieve 95% and 90% detection accuracy, respectively.