关键词: 自杀意念检测, 社交媒体, 机器学习, 深度学习, 特征提取

Abstract: Suicide constitutes a grave societal quandary, an eminent worldwide public health predicament, and a pivotal determinant of global mortality. In recent years, with the rapid development of the Internet, social media have emerged as a novel domain for individuals to express their emotions, including suicidal ideation, attempts, and behaviors. Consequently, these platforms have evolved into crucial data repositories and essential assessment criteria for detecting suicidal ideation. With the advent of artificial intelligence technology, the utilization of machine learning in detecting suicidal ideation among social media users has emerged as a scintillating subject. The field in China, however, lacks sufficient research and has yet to establish a comprehensive system. To systematically review the research status and development context of suicide ideation detection, this paper presents a systematically summary of machine learning technology's application in empowering suicide ideation detection, marking the first review in this field conducted in China over the past three years. Firstly, this paper provides an overview of the definition, process, commonly employed methods, and evaluation indicators for detecting suicidal ideation. Secondly, this paper provides a comprehensive overview of suicide ideation detection techniques, encompassing both traditional machine learning and deep learning approaches. The key methodologies, fundamental concepts, merits, and limitations of each method are thoroughly compared and analyzed. Furthermore, the urgent issues and innovative solutions in this field are summarized, with a particular focus on the application of large language models such as ChatGPT and multi-modal models. Finally, the limitations of machine learning in the application research of suicide ideation detection on social media are comprehensively discussed, and future research directions are proposed, in order to further promote the formation of a new paradigm of data-driven, human-computer collaboration, interdisciplinary integration, and cross-cultural domain of suicide ideation detection, so as to provide reference and reference for researchers in related fields.

Key words: Suicidal ideation detection, Social media, Machine learning, Deep learning, Feature extraction