Actionable companion · Learning design & outcomes

Learning with agency: Graph learning based suicidal ideation detection via tree-drawing test

AI-assisted editorial implementation guide, separate from the original research. Proposed actions are not claims or recommendations attributed to the source authors. No independent peer review of this guide is asserted.

Read original paper Search paper agent CC BY 4.0

Evidence and context

Title and abstract screening; stage labels describe reported settings, not verified numerical ages. Design labels require full-text appraisal. Reported stage(s): Secondary school. Verify the participant population and task before transferring the findings.

Original authors’ abstract, reproduced under the linked source license:

Introduction Adolescent suicide is a critical public health concern worldwide, necessitating effective methods for early detection of high suicidal ideation. Traditional detection methods, such as self-report scales, suffer from limited accuracy and are susceptible to personal concealment. Automatic methods based on artificial intelligence techniques are more accurate, while they often lack scalability due to strict data requirements. In order to achieve a balance between accuracy and scalability, this paper introduces the Tree-Drawing Test (TDT) as an effective tool for suicidal ideation detection, and proposes a novel graph learning approach to enable its automatic application. Methods The proposed method first constructs a semantic graph based on psychological features annotated automatically from tree-drawing images, and leverages a Graph Convolutional Network (GCN) model to realize individual suicidal ideation detection. To evaluate this method, a real dataset of 806 students from primary and secondary school in Shaanxi Province, China, is collected, and some metrics including macro-F1, G-mean, and false positive rate are used. Results The results demonstrate that the proposed method significantly outperforms traditional machine learning and convolution neural network approaches. The ablation study demonstrates the effectiveness of feature “leaves and fruits” in detecting suicidal ideation. Further experiments demonstrate that the proposed method remains stable even when the model is disturbed, such as when a tree-drawing image cannot be fully represented. Discussion The proposed method highlights its effectiveness in large-scale suicidal ideation screening, as it not only achieves high detection performance but also maintains model stability while remaining flexible and adaptable.

What kind of evidence is this?

Supporting AI learning science · Design requires full-text appraisal

Stage: Secondary school. Screening labels describe reported context and study-design language. They are not a risk-of-bias rating or proof of effectiveness for an age band.

Study methods · source passage

To conduct this research, we established a dataset involving lots of students from primary and secondary schools in Shaanxi Province, China. There are totally 806 participants (392 males and 414 females). Each participant completed the Beck Scale for Suicide Ideation and drew a tree on A4-sized white paper as part of the Tree-Drawing Test. To ensure the effectiveness of scale, we evaluated itemtotal correlation and refined it to nine items, achieving a Cronbach’s α of 0.89, indicating high internal consistency and reliability (21, 22). The scoring ranges from 0 to 10 points for each item, with a total score ranging from 0 to 90 for the nineitem scale. The score distribution is displayed in Figure 2 . The average score for suicidal ideation across all subjects was 17.19 (SD = 7.01). To categorize the subjects into groups based on their level of suicidal ideation, we applied a threshold of mean plus standard deviation, a common practice in psychological research (23, 24). Consequently, 94 subjects were classified into the high suicidal ideation group, while the remaining 712 were categorized as low suicidal ideation.

Limitations and uncertainty · source passage

As for the comparison with other graph learning models, GCN outperforms in terms of all metrics. We hypothesize that, due to the limited number of node types and edge types in the image-feature semantic graph, heterogeneous graph learning models are unable to fully demonstrate their advantages. Compared to GAT, which assigns adaptive edge weights through attention mechanisms, GCN still demonstrates superior performance. This discrepancy may suggest that maintaining equal edge weights is more suitable for the current task. Additionally, this could indicate that the current dataset size might not be sufficient for graph learning to learn the feature importance. Further exploration is warranted with a larger dataset in the future research.

Design the system before the pilot

Use this study to inform a learning mechanism or implementation question. It is not direct evidence that autonomous agents improve learning. Keep the proposed agent behavior separate from what the source actually tested.

Specify a measurable learning objective, a comparator and a task without tool assistance before adopting a system.

A 90-day district pilot

Days 1–15 · Appraise. Have the curriculum lead and a teacher read the methods and limitations. Record the study design, sample, comparator, outcomes and evidence gaps. Define a local question that the study can help inform.
Days 16–30 · Design. Ask learners to attempt a problem unaided, compare an agent hint, explain a revision and verify the final reasoning against a trusted source. Make the permitted use explicit for each assignment. Name an accountable owner and set the baseline, success threshold and budget before implementation.
Days 31–60 · Test. Start with a small voluntary cohort, preserve an accessible non-AI route and require teacher review. Use synthetic or de-identified examples; approve data handling and tool terms through district processes.
Days 61–90 · Decide. Measure unaided transfer and delayed retention, calibration of confidence and the ability to identify an incorrect AI explanation. Record whether help-seeking becomes answer copying. Compare with baseline, document uncertainty and stop or revise if learning worsens, access is unequal or critical errors persist.

The board decision brief

Report the learning problem, evidence quality, participating population, total cost including review time, measured benefit, errors, access differences and remaining uncertainties. Make a dated recommendation to stop, revise or expand. A positive perception survey alone does not establish learning improvement.

Attribution and reuse

Ye Liu; Jiashuo Zheng; Yang Zeng; Fang Luo; Xuetao Tian. Graph learning based suicidal ideation detection via tree-drawing test. Frontiers in Psychiatry (2025-07-18). https://doi.org/10.3389/fpsyt.2025.1617650

Companion prepared 2026-10-30. Original authorship and DOI are preserved. The abstract is an attributed source extract; the implementation plan is new AI-assisted editorial material, offered under the same CC BY 4.0 license. This guide has not undergone independent peer review.