EM-PERSONA: EMotion-assisted Deep Neural Framework for PERSONAlity Subtyping from Suicide Notes

Proceedings of the 29th International Conference on Computational Linguistics, pages 1098–1105 October 12–17, 2022. 1098 EM-PERSONA: EMotion-assisted Deep Neural Framework for PERSONAlity Subtyping from Suicide Notes Soumitra Ghosh1, Dhirendra Kumar Maurya1, Asif Ekbal1 and Pushpak Bhattacharyya2 1Department of Computer Science and Engineering, IIT Patna, India 2Department of Computer Science and Engineering, IIT Bombay, India {ghosh.soumitra2,mauryadhirendra563}@gmail.com, asif@iitp.ac.in, pb@cse.iitb.ac.in Abstract The World Health Organization has empha- sised the need of stepping up suicide preven- tion efforts to meet the United Nation’s Sustain- able Development Goal target of 2030 (Goal 3: Good health and well-being). We address the challenging task of personality subtyping from suicide notes. Most research on personality subtyping has relied on statistical analysis and feature engineering. Moreover, state-of-the-art transformer models in the automated personal- ity subtyping problem have received relatively less attention. We develop a novel EMotion- assisted PERSONAlity Detection Framework (EM-PERSONA). We annotate the benchmark CEASE-v2.0 suicide notes dataset with per- sonality traits across four dichotomies: In- troversion (I)-Extraversion (E), Intuition (N)- Sensing (S), Thinking (T)-Feeling (F), Judging (J)–Perceiving (P). Our proposed method out- performs all baselines on comprehensive eval- uation using multiple state-of-the-art systems. Across the four dichotomies, EM-PERSONA improved accuracy by 2.04%, 3.69%, 4.52%, and 3.42%, respectively, over the highest per- forming single-task systems. 1 Introduction Suicide continues to be one of the significant causes of death worldwide (Ghosh et al., 2020). Given the significance of personality as a basis for understand- ing psychopathology (Krueger and Tackett, 2006) and the variability in risk factors associated with suicide, subtyping patients based on their person- alities can provide greater specificity than simple comparisons of suicidal to non-suicidal individuals (Ortigo et al., 2009). Pompili et al. (2008) showed that emotions such as anger, aggressiveness, anxi- ety, and sadness were associated with personality traits of individuals who attempted suicide. We quote a few excerpts from suicide notes (SNs) and online personality posts (PPs) in Table 1 to show how personality discussions on public fo- rums and genuine SNs generally follow similar SN-1 They just do whatever the fuck they want and justify it later. PP-1 They can do anything they like to make any law they like. SN-2 He is an ugly stupid faggot and we should kill him. PP-2 They’re oppressing you, kill them all! Table 1: Sample excerpts from a couple of SNs and PPs. language patterns. We computed cosine similarity (CosSim) scores between SNs (from CEASE-v2.0 corpus (Ghosh et al., 2022)) and PPs (from MBTI dataset1) and observed an alarming number of SNs having a considerable amount of word-based sim- ilarity with generic PPs. The results are shown in Table 2. We observed CosSim scores over 0.6, 0.5, 0.4 for 12, 39 and 113 SNs, respectively with respect to the PPs in MBTI dataset. CosSim >0.30 >0.40 >0.50 >0.60 # Notes 204 113 39 12 Table 2: Cosine Similarity scores between suicide notes and personality posts. Our primary contributions are two-fold: we present a novel corpus of suicide notes annotated with personality traits across four dichotomies and develop …

Content truncated to 500 words