Feature Attention Network: Interpretable Depression Detection from Social Media

Feature Attention Network: Interpretable Depression Detection from Social Media Hoyun Song∗ Jinseon You∗ Jin-Woo Chung Jong C. Park† School of Computing Korea Advanced Institute of Science and Technology {hysong, jsyou, jwchung, park}@nlp.kaist.ac.kr Abstract Although depression is one of the most com- mon mental disorders, the depressed individ- uals may not be aware of their symptoms at all so that they sometimes miss the appropri- ate time for treatment. In order to prevent this problem, many researchers looked into so- cial media to figure out depressed individu- als by analyzing the differences in language use. While they have recently achieved rea- sonable performance in detecting depression, especially using deep learning methods, such methods still do not provide a clear way to explain why certain individuals have been de- tected as depressed. To address this issue, we propose Feature Attention Network (FAN), in- spired by the process of diagnosing depres- sion by an expert who has background knowl- edge about depression. We evaluate the per- formance of our model on a large scale general forum (Reddit Self-reported Depression Diag- nosis) dataset. Experimental results demon- strate that FAN shows good performance with high interpretability despite a smaller num- ber of posts in training data. We investigate different aspects of posts by depressed users through four feature networks built upon psy- chological studies, which will help researchers to investigate social media posts to find useful evidence for depressive symptoms. 1 Introduction Depression is one of the most common mental dis- orders which not only hampers an individual’s daily ∗ Equal contribution (alphabetically ordered) † Corresponding author function, but also causes a variety of serious social problems such as suicide. The depressed individuals may not be aware of their depressed symptoms at all so that they sometimes miss the appropriate time for taking care of depression. To prevent this prob- lem, many researchers used social media to identify depressed users by analyzing the differences in lan- guage use. The researchers reported good performance in de- tecting depression by analyzing a variety of features such as lexical features including symptom lexicons (De Choudhury et al., 2013a; De Choudhury et al., 2013b; Coppersmith et al., 2014), syntactic features (Nambisan et al., 2015; Gkotsis et al., 2016), sen- timent analysis (Wang et al., 2013; Preot¸iuc-Pietro et al., 2015), or topic modeling (Resnik et al., 2015; Preot¸iuc-Pietro et al., 2015). However, these ap- proaches are mainly based on labor-intensive engi- neering with handcrafted features, which show lim- ited performance for classification. To further en- hance the performance of depression detection, re- searches have recently proposed deep learning meth- ods (Kshirsagar et al., 2017; Yates et al., 2017; Orabi et al., 2018). However, despite the improved perfor- mance, they still do not present a clear way to ex- plain why certain individuals have been detected as depressed, which is critical for a follow-up study on specific diagnosis and prevention. We propose a model that can interpret the process of classifying social media users when it derives the result of depression …

Content truncated to 500 words