The increasing prevalence of anxiety and depression, exacerbated by the COVID-19 pandemic and long waiting lists for mental healthcare, highlights the urgent need for automated mental health detection. Social media platforms offer a rich source of self-reported data, making them valuable resources for Mental Health Analysis (MHA). Existing Computational Intelligence Techniques (CIT) show promise, but lack explainability. This paper advocates for a shift towards Level 1 studies in MHA, focusing on causal analysis (cause detection, causal inference, cause categorization) and perception mining to understand the psychological perspectives shaping user intentions. The authors emphasize the importance of discourse analysis and pragmatics within NLP to bridge the gap between CIT and clinical psychology practice. Collaboration between NLP researchers and clinical psychologists is crucial to maintain the integrity of this approach, aiming to develop explainable AI models for real-time applications in personalized mental healthcare, such as conversational AI agents.

The authors review existing literature on machine learning and deep learning techniques for mental health identification and prediction from digital data, medical records, and social media text. They highlight the limitations of current approaches, including demographic bias in data collection, user consent issues, and a lack of theoretical underpinnings about human behavior in user-generated content. Existing studies are categorized into Level 0 (algorithm-focused) and Level 1 (in-depth analysis of cause-and-effect relationships and perceptions). While Level 0 studies have made progress in mental health classification, they often lack explainability. The paper positions itself as a Level 1 study, aiming to provide a deeper understanding of human behavior through a comprehensive analysis of social media language.

The paper proposes a methodology built upon two core approaches: Causal Analysis and Perception Mining. **Causal Analysis** is broken down into three sub-tasks: 1. **Cause Detection:** Identifying whether a text contains reasons for a user's mental state. This involves identifying indicators of causes like job loss, family issues, or financial problems. 2. **Causal Inference:** Extracting explanations from the text to understand the reasons behind mental illness. This involves identifying text segments that explain the causes. 3. **Cause Categorization:** Classifying the identified causes into pre-defined categories (e.g., job-related, relationship-related, etc.). The authors suggest using discourse parsing, knowledge graphs (KGs), and techniques like semantic dependency information-guided sentence simplification (SISS) to analyze complex text and extract causal relationships. Knowledge graphs represent relationships between events, objects (aspects of mental well-being), and situations to reveal cause-and-effect relationships. Discourse analysis helps determine connections between different text segments. **Perception Mining** focuses on understanding the psychological perceptions of users reflected in their social media posts. This approach involves analyzing a user's historical timeline of posts to identify evolving attitudes and beliefs. Psychological theories like self-perception theory and structural balance theory are used as a framework for understanding how users interpret sensory information and form their perspectives. The authors advocate for utilizing pragmatics and discourse analysis to uncover deeper nuances in users' perceptions and beliefs, including moral foundations and the influence of interpersonal relationships. Existing datasets and models in moral sentiment classification and personality analysis are mentioned as relevant resources. The methodology emphasizes integrating these two core approaches, highlighting the importance of explainability in AI models for mental health analysis. The authors propose an output representation for explainable AI that includes causal relationships and perception mining insights.

The paper's main contribution is its proposition of a two-pronged approach using causal analysis and perception mining to improve the interpretation and explainability of AI models for mental health analysis on social media. Existing research largely focuses on predicting mental health states without delving into the underlying reasons or psychological perspectives. This paper addresses this gap by advocating for a deeper, more nuanced analysis. Key elements are: * **The importance of discourse analysis and pragmatics:** The paper emphasizes moving beyond simple linguistic features and semantic analysis to incorporate the structure and relationships within text segments (discourse) and the contextual implications of language use (pragmatics). * **Integration of causal analysis and perception mining:** The authors argue that these two approaches are complementary, with perception mining providing context and deeper understanding for causal analysis. * **Explainable AI:** The paper strongly advocates for developing AI models that not only predict mental health states but also provide clear and understandable explanations for their predictions. * **Use of Knowledge Graphs (KGs):** KGs are proposed to represent relationships between events and aspects of mental well-being, allowing for better understanding of causal connections. * **Longitudinal Analysis:** The importance of analyzing a user's historical timeline (longitudinal studies) is highlighted for perception mining to track changes in attitudes and beliefs over time. * **Ethical considerations:** The authors acknowledge the importance of addressing ethical concerns related to privacy, responsibility, transparency, and fairness in the development and application of AI models for mental health analysis. The paper provides a framework and suggests several potential avenues for future research, including the development of new datasets with annotations specifically designed for causal analysis and perception mining. The authors also mention several existing datasets that could be expanded or repurposed for this purpose.

The proposed approach directly addresses the limitations of existing research in mental health analysis on social media by integrating causal analysis and perception mining to enhance explainability and accuracy. The integration of discourse analysis and pragmatics offers a more comprehensive understanding of the nuances of human language and its relationship to mental health. The framework’s emphasis on explainability is critical for building trust and facilitating the adoption of AI-based tools by mental health professionals. Future research is crucial in further developing and validating this framework, which can have significant impacts on personalized mental healthcare, early detection of mental health issues, and the development of more effective interventions.

This position paper advocates for a paradigm shift in mental health analysis using social media data, proposing a framework that integrates causal analysis and perception mining within an NLP context. The authors emphasize the crucial role of explainability in developing trustworthy and impactful AI models. Future research should focus on developing robust datasets, refining the proposed methodology, and exploring the ethical implications of using AI in mental healthcare. This approach promises to advance our understanding of mental health conditions and contribute to the development of effective and ethical AI-driven interventions.

The authors acknowledge that this paper presents a theoretical perspective and lacks empirical studies. While it provides a valuable framework, future work needs to include implementation and evaluation of the proposed methodology. The study is limited to analyzing language in social media and does not directly address clinical diagnoses or symptoms. The ethical considerations discussed are general guidelines, and more specific ethical protocols are needed for particular applications.