Cross-Lingual Suicidal-Oriented Word Embedding toward Suicide Prevention

Abstract

Early intervention for suicide risks with social media data has increasingly received great attention. Using a suicide dictionary created by mental health experts is one of the effective ways to detect suicidal ideation. However, 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 propose a cross-lingual suicidal ideation detection model that can identify whether a given social media post includes suicidal ideation or not. To utilize the existing suicide dictionaries 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 Chinese, respectively. By applying an ensemble approach for different languages, the model achieves high accuracy, over 87%. We believe our model is useful in accessing suicidal ideation using social media data for preventing potential suicide risk in an early stage.

Detecting Suicidality with a Contextual Graph Neural Network

“…Specifically, prior research showed that linguistic characteristics revealed in social media posts (Sawhney et al, 2020(Sawhney et al, , 2021a) could be linked to suicidal ideation. In particular, utilizing suicide dictionaries made by domain experts has been demonstrated as effective (Lv et al, 2015;Cao et al, 2019;Gaur et al, 2019;Lee et al, 2020), and such lexicon-based methods are known to be fast, explainable, and easy to implement (Kotelnikova et al, 2021;Razova et al, 2021). For example, Lv et al (2015) developed and validated that a Chinese suicide dictionary made by domain experts helps predict suicidality.…”

“…With the recent advancement of deep learning technologies, high-performing deep learning models have been proposed for accurately assessing suicidality (Sawhney et al, 2021a,b;Cao et al, 2020). In this way, incorporating a suicide dictionary into a deep learning model has received great attention (Cao et al, 2019;Lee et al, 2020). For example, Cao et al (2019) built suicide-oriented word embeddings to intensify the sensibility of suiciderelated lexicons and employed a two-layered attention mechanism.…”

“…For example, Cao et al (2019) built suicide-oriented word embeddings to intensify the sensibility of suiciderelated lexicons and employed a two-layered attention mechanism. Lee et al (2020) proposed a deep learning method to utilize existing suicide dictionaries for the low-resource language where a knowledge-based suicide dictionary has not yet been developed. However, the prior work focused on how each word in a post is associated with the words/phrases in a suicide dictionary, e.g., via lexical matching (Lv et al, 2015;Gaur et al, 2019) or fixed word embeddings (Cao et al, 2019;Lee et al, 2020), which may fail to capture the semantic information of suicide lexicons in the suicide-related context.…”

“…The awareness of the severity of suicide has led researchers to develop suicidality detection models using a deluge of user activity data on social media, which can help capture latent warning signs of suicide in an early stage (Sawhney et al, 2020;Lee et al, 2020;Shing et al, 2020). For example, the prior work showed that linguistic characteristics revealed in social media posts could be linked to suicide risks (De Choudhury et al, 2016; Shing et al, 2018).…”

Towards Suicide Prevention from Bipolar Disorder with Temporal Symptom-Aware Multitask Learning

“…However, while the prior work has revealed valuable insights into understanding the behavior of BD patients revealed on social media, little attention had been paid to developing a model that can predict the future suicidality of a BD patient. Although a few studies have proposed methods to identify the current risk of suicide in a given social media post [30,43,44], suicidal ideation can often quickly lead to an actual attempt, thereby making them ineffective in preventing suicide [9,10,41,52]; hence, exploring the BD's risk factors that can lead to suicide ideation for predicting future suicidality is crucial. Therefore, this paper aims to predict the future suicidality of BD patients based on their mood I've wanted to die since I was in 4th grade.…”