Latent Suicide Risk Detection on Microblog via Suicide-Oriented Word Embeddings and Layered Attention

Abstract

Despite detection of suicidal ideation on social media has made great progress in recent years, people's implicitly and anti-real contrarily expressed posts still remain as an obstacle, constraining the detectors to acquire higher satisfactory performance. Enlightened by the hidden "tree holes" phenomenon on microblog, where people at suicide risk tend to disclose their inner real feelings and thoughts to the microblog space whose authors have committed suicide, we explore the use of tree holes to enhance microblog-based suicide risk detection from the following two perspectives. (1) We build suicide-oriented word embeddings based on tree hole contents to strength the sensibility of suicide-related lexicons and context based on tree hole contents. (2) A two-layered attention mechanism is deployed to grasp intermittently changing points from individual's open blog streams, revealing one's inner emotional world more or less. Our experimental results show that with suicide-oriented word embeddings and attention, microblog-based suicide risk detection can achieve over 91% accuracy. A large-scale well-labelled suicide data set is also reported in the paper.

Cross-Lingual Suicidal-Oriented Word Embedding toward Suicide Prevention

“…There has been great interest in developing a model to detect suicide risks based on user behavior such as the number of posts or followers (Kumar et al, 2015;Cao et al, 2019) and linguistic characteristics (e.g., frequently used words like 'family', 'sad', or 'dream') revealed in social media posts (Gaur et al, 2019;Lv et al, 2015). For example, conducted a linguistic analysis on social media data and found a few signals that can be linked to suicide attempts and suicidal ideation.…”

“…We adopt a suicidal-oriented word embedding similar to the prior work (Cao et al, 2019) that refines a word embedding to capture domain knowledge from a pre-built suicide-related dictionary. Figure 2 !…”

“…For training a suicidal-oriented word embedding, we use a pre-built suicide-related dictionary. If such a dictionary contains word-level information that exhibits how much a word is associated with suicidal ideation (like a Chinese dictionary (Lv et al, 2015)), we apply the word-masking classification method similar to the prior work (Cao et al, 2019). To this end, we generate suicidal and non-suicidal expressions for a given input suiciderelated post collected for word embedding, e.g., Weibo Tree Hole data (Cao et al, 2019).…”

“…If a pre-built suicide-related dictionary contains sentence-level information such as Gold Standard Dataset (Gaur et al, 2019), which includes English sentences related to suicidal ideation, directly applying the word-masking method (Cao et al, 2019) is not possible; words for masking cannot be extracted from the given sentence-level dictionary. Hence, we use the sentences belonging to the dictionary as suicidal expressions and non-suiciderelated posts as non-suicidal expressions for developing the sentence-level word embedding.…”

Leverage Social Media for Personalized Stress Detection

“…Stressed dataset. The community where users tell their secrets can be understood as a tree hole [3]. Inspired by this, we searched "stress tree hole" on Weibo to find Stress Tree Holes, where users express their stressful emotions, complain about their difficult challenges and seek for stress relieves.…”