Machine Learning Approach for Depression Detection in Japanese

Machine Learning Approach for Depression Detection in Japanese Yuka Niimi Graduate School of Social Informatics Aoyama Gakuin University c8121005@aoyama.jp Yutaka Miyaji Graduate School of Social Informatics Aoyama Gakuin University uta@si.aoyama.ac.jp Abstract In this paper, we describe the results of our research aimed at detecting depression in Japanese sentences. In the United States and elsewhere, machine learning approaches to detect depression from language have been demonstrated. However, in Japanese text, there are only two studies that have addressed the detection of depression. In this study, to detect depression based on linguistic features, even in documents that do not explicitly men- tion the topic of depression, we build a ma- chine learning model that detects depression in Japanese by eliminating topics that suggest depression or depression. We also examine the accuracy when the parts of speech are lim- ited and when the number of labels is aligned. In the performances of our models with 5-fold cross-validation, we were able to obtain a high evaluation. 1 Introduction Depression is a common illness that affects over 264 million people worldwide. Depression and other mental health conditions are becoming more com- mon around the world. At its worst, depression can lead to suicide. A resolution passed by the World Health Assembly in 2013 advocated for a compre- hensive, coordinated approach to mental illness at the national level1. According to a patient survey conducted by the Ministry of Health, Labor and Welfare (MHLW), 1https://www.who.int/news-room/fact- sheets/detail/depression an official organization in Japan, the total number of patients with mood disorders such as depression in Japan has increased 2.9 times in 21 years, from 433,000 in 1996 to 1,276,000 in 2009. Depression also affects the economy, and the annual social loss due to depression in Japan is estimated to be 2 tril- lion yen (Sato, 2014). Therefore, ”stress check sys- tem”2 has been introduced into the public system as a mandatory requirement for business establish- ments with 50 or more employees since 2015 to pre- vent mental health problems such as depression. As can be seen from these examples, depression is an issue of high importance in Japan. In psycholinguistics, one of the behavioral indica- tors of depression is linguistic features. (Fine, 2008) reported that the idea that language use reflects the speaker’s mind. Psychiatrists also have used it to assess mental health conditions. Many studies fo- cus on the linguistic features of depression and ana- lyze natural language. (Yip, 2018) manifested the communicative patterns of anxiety and depression communication using the framework of discourse analysis. Unfortunately, lack of resources, a short- age of trained healthcare providers, and the social stigma associated with mental illnesses are all obsta- cles to the effective treatment of depression. There- fore, They sometimes miss the appropriate time for taking care of depression. Detecting depression with machine learning will become increasingly impor- tant in the future. In this study, we use blog data to limit topics with LDA, vectorize documents with TF-IDF, reduce di- mensions with SVD, and build classifiers with SVM. 2https://www.mhlw.go.jp/stf/houdou/0000082587.html Our …

Content truncated to 500 words