Pro Yearly is on sale from $80 to $50! »

Inferring Social Media Users’ Mental Health Status from Multimodal Information

Inferring Social Media Users’ Mental Health Status from Multimodal Information

Worldwide, an increasing number of people is suffering from mental disorders such as depression and anxiety. In the United States only, one in every four adults suffers from a mental health condition, which makes mental health a pressing concern. In this paper, we explore the use of multimodal cues present in social media posts to predict a user's mental health status. Specifically, we focus on identifying social media posts that indicate either a mental health condition or its onset. We collect posts from Flickr and apply a multimodal approach that consists of jointly analyzing image, language, and post cues and their relation to mental health. We conduct several classification experiments aiming to discriminate between (1) healthy users and users affected by a mental health illness; and (2) healthy users and users prone to mental illness. Our experimental results indicate that using multiple modalities can improve the performance of this classification task as compared to the use of one modality at a time, and can provide important cues into a user's mental status.

8be6e1c8c9c6a183d36aad17cb4f310c?s=128

Zhentao Xu

May 18, 2018
Tweet

Transcript

  1. Inferring Social Media Users’ Mental Health Status from Multimodal Information

    Zhentao Xu, Veronica Perez-Rosas, Rada Mihalcea University of Michigan {frankxu, vrncapr, mihalcea}@umich.edu The Problem Background: Growing number of people suffer from mental disorders. Current diagnosis is limited. Assumption: People who often use words/tags related to mental disorders are more likely to suffer from these disorders. Research Questions: Can we identify a person’s mental health status from his/her social media posts? Can we predict the potential onset of disorder in future from current social media posts? Data Collection Modalities Top Features in Characterizing Mental Disorder & Research Conclusions mentalillness, bipolardisorder, schizophrenia, selfharm, depression, bipolar, psychosis, insanity, anxiety, depressed, insane, disorder, suicide, despair Visual modalities from photos - low-level: color, brightness, texture, lines, - high-level: faces, objects, scenes Language modalities from captions - stylistic features, (part-of-speech) n-grams, LIWC Meta modalities from posts - views, activity time, EXIF Classification Experiments and Performances Experiments: Identify social media users with mental disorder (under our assumption) Predict users prone to mental disorders using both machine learning classifiers and a two-layer neural network. Performance: Top visual features Top language features Top meta features Significant Visual Features ( Healthy users prefer features w/ positive effect size) Significant Language Features Significant Meta Features Temporal Pattern between healthy, mental disorder, and potential mental disorder posts Sample Data mentalillness 12056 tags Extend the seed tag with co-occurance Filter tags using (1) popularity, (2) PMI, (3) manually checking The Final Mental-Illness Tag Set Tag Level Post Level Final Dataset: 15,000 healthy posts; 12,056 mental illness posts; 11,828 pre-mental illness posts Mental Disorder Identification Performance Mental Disorder Prediction Performance Conclusions: Individuals suffering from or prone to mental disorders prefer darker images with high contrasts showing indoors scenes and fewer faces The derived features are useful for the prediction of user’s mental health status The combination of visual, language, and meta information lead to better performance as compared to the use of individual modalities This material is based in part upon work supported by the Michigan Institute ​ for Data Science (MIDAS)​