Cross-Lingual Suicidal-Oriented Word Embedding toward Suicide Prevention

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 2208–2217 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 2208 Cross-Lingual Suicidal-Oriented Word Embedding toward Suicide Prevention Daeun Lee1, Soyoung Park2, Jiwon Kang1, Daejin Choi3, Jinyoung Han1∗ 1Department of Applied Artificial Intelligence, Sungkyunkwan University 2National Assembly Research Service 3Department of Computer Science & Engineering, Incheon National University {delee12, jiwonkang, jinyounghan}@skku.edu sypark@assembly.go.kr, djchoi@inu.ac.kr Abstract Early intervention for suicide risks with so- cial media data has increasingly received great attention. Using a suicide dictionary created by mental health experts is one of the effec- tive ways to detect suicidal ideation. How- ever, little attention has been paid to validate whether and how the existing dictionaries for other languages (i.e., English and Chinese) can be used for predicting suicidal ideation for a low-resource language (i.e., Korean) where a knowledge-based suicide dictionary has not yet been developed. To this end, we pro- pose a cross-lingual suicidal ideation detection model that can identify whether a given so- cial media post includes suicidal ideation or not. To utilize the existing suicide dictionar- ies developed for other languages (i.e., English and Chinese) in word embedding, our model translates a post written in the target language (i.e., Korean) into English and Chinese, and then uses the separate suicidal-oriented word embeddings developed for English and Chi- nese, respectively. By applying an ensemble approach for different languages, the model achieves high accuracy, over 87%. We be- lieve our model is useful in accessing suicidal ideation using social media data for preventing potential suicide risk in an early stage. 1 Introduction As online social media has become the norm to share our daily lives, people often share their emo- tions, feelings, and mental state. This has spurred scholars to identify diverse mental health problems such as depression, anxiety, bipolar disorder, or sui- cidal thoughts using plenty of user behavior data on online social media (Ji et al., 2019; Pavalanathan and De Choudhury, 2015; Kim et al., 2020). Such user behavior data can provide a cue for identifying individual mental state or even suicide risk (O’dea ∗Corresponding author. et al., 2015; Ren et al., 2015; Coppersmith et al., 2018), which can be used to support mental health care (Shen and Rudzicz, 2017; Suhara et al., 2017). Among the diverse mental health problems, sui- cide has become one of the big and emerging con- cerns worldwide. The OECD (Organization for Economic Cooperation and Development) reported 11.2 deaths per 100,000 population in OECD coun- tries in 2017 (OECD, 2020). In particular, the suicide rate of Korea and the USA was 24.6 and 13.9 deaths per 100,000 population in 2016, which ranked 1st and 8th, respectively. The awareness of the severity of suicide has led researchers to assess mental health using so- cial media data for recognizing potential warning signs of suicide in an early stage (Pavalanathan and De Choudhury, 2015; O’dea et al., 2015). In partic- ular, linguistic characteristics (e.g., frequently used words like ‘family’, ‘sad’, or ‘dream’) of social me- dia …

Content truncated to 500 words