School of
Information Technology and Electrical Engineering

Speaker: Mr Shaoxiong Ji
Seminar Date: Thu, 17/10/2019 - 11:45
Venue: 78-631
Host: Prof Xue Li; Assoc Prof Helen Huang

Seminar Type: Thesis Review


Mental health issues, such as anxiety, depression and suicidal ideation, are becoming increasingly concerning in modern society. Without effective treatment, severe mental disorders but without effective treatment are very likely to turn to suicide. The reasons people commit suicide are complicated, including social factors like social isolation; personal issues, such as alcoholism or career failure; or the influence of negative life events. Thus, developing effective suicide prevention techniques is urgently needed. Detecting suicidal ideation early is one of the most effective methods to prevent suicide. Research on suicidal intention understanding and suicide prevention mainly concentrates on its psychological and clinical aspects and classifying questionnaire results via supervised learning. However, collecting data and/or patients is typically expensive, from both a psychological and a clinical perspective.

With the advances in social media, more and more individuals are expressing their feelings and suffering on the internet. Anonymous online websites provide a comfortable place for people to interact with others using asynchronous communication. The social content in online communities for depression provides topic features and psycholinguistic clues for the automatic detection and prediction of suicidal ideation. Using data mining techniques on social networks and applying neural networks provide an avenue to understand the intention within online posts and even relieve a person's suicidal intentions. In this thesis, we firstly have a comprehensive content analysis to discover knowledge from suicide-related text and preforms a benchmarking on binary classification of suicidal ideation including using feature extraction based classifiers and deep neural networks. The reasons of committing suicide are complicated, and suicidal factors vary from individuals. To incorporate suicidal factors for suicidal intention understanding, we consider sentimental clues and topics in people's posts and propose to reason the relations between those factors and posts with attention relation networks for fine-grained suicidal ideation detection. Lastly, we study suicidal ideation detection in another scenario of private chatting. To tackle the challenge of isolated data in private chat rooms, we develop a knowledge transferring framework to train a global model for knowledge sharing with distributed agents. Overall, this thesis develops methods with content analysis, feature engineering, and deep learning techniques in the hope of using effective suicidal ideation detection to prevent suicide and save people's life.


Shaoxiong is an MPhil candidate at the School of Information Technology & Electrical Engineering, UQ.