STAMPsy: Towards SpatioTemporal-Aware Mixed-Type Dialogues for Psychological Counseling

TL;DR

STAMPsy introduces the first spatiotemporal-aware mixed-type dialogue dataset for online psychological counseling, collecting 5,006 dialogues across task-oriented diagnosis, knowledge-grounded dialogue, conversational recommendations, empathetic dialogue, and QA, all annotated with personal, spatiotemporal, and psychological knowledge. It then proposes Self-STAMPsy, a spatiotemporal-aware end-to-end framework with Helping Skills Selection, SpatioTemporal Stamp Processing, Adaptive Retrieval Augmented Generation, and Iterative Self-feedback to produce world-aware, skill-guided responses. Experiments show that clarifying client goals and incorporating spatiotemporal state information, along with iterative self-feedback, significantly improve both objective metrics (BLEU, ROUGE, BertSim, STSP accuracy) and human evaluations (GHSC, empathy). The work demonstrates the potential of LLM-based tools to augment accessibility and efficiency in mental health support while emphasizing the necessary role of professionals in safeguarding quality and safety.

Abstract

Online psychological counseling dialogue systems are trending, offering a convenient and accessible alternative to traditional in-person therapy. However, existing psychological counseling dialogue systems mainly focus on basic empathetic dialogue or QA with minimal professional knowledge and without goal guidance. In many real-world counseling scenarios, clients often seek multi-type help, such as diagnosis, consultation, therapy, console, and common questions, but existing dialogue systems struggle to combine different dialogue types naturally. In this paper, we identify this challenge as how to construct mixed-type dialogue systems for psychological counseling that enable clients to clarify their goals before proceeding with counseling. To mitigate the challenge, we collect a mixed-type counseling dialogues corpus termed STAMPsy, covering five dialogue types, task-oriented dialogue for diagnosis, knowledge-grounded dialogue, conversational recommendation, empathetic dialogue, and question answering, over 5,000 conversations. Moreover, spatiotemporal-aware knowledge enables systems to have world awareness and has been proven to affect one's mental health. Therefore, we link dialogues in STAMPsy to spatiotemporal state and propose a spatiotemporal-aware mixed-type psychological counseling dataset. Additionally, we build baselines on STAMPsy and develop an iterative self-feedback psychological dialogue generation framework, named Self-STAMPsy. Results indicate that clarifying dialogue goals in advance and utilizing spatiotemporal states are effective.

Figures (7)

• Figure 1: An example of STAMPsy with spatiotemporal state and reference knowledge.
• Figure 2: The collection workflow of STAMPsy. We extract reference knowledge from context under 9-Box CCM and then gain multi-turn dialogues with a multipart instruction. All the dialogues are annotated and revised by psychological experts.
• Figure 3: Sequence visualization of the common dialogue goal flow patterns within the first 5 counselor-helping skills.
• Figure 4: The framework of the proposed Self-STAMPsy. A detailed prompt template is open-sourced.
• Figure 5: A case of generated answers and the golden answer. It can be observed that without the guidance of counseling techniques, other models tend to explore how to improve children's learning, which appears more like a communication about education rather than addressing the emotional concerns of the visitor, which is different from golden answer in the professional books. Subsequent conversations should revert to focusing on the client's personal issues, so it is better to come into play the "emotional reflection" counseling technique. In our model, the fine-tuned BERT model first correctly categorizes the helping skill. Based on the goal of "Reflection of Feeling", Self-STAMPsy can achieve better response effects, returning to the client's own concerns.