StageWell: A Process-Aligned Chinese Corpus for Positive-Psychology Support Dialogue

• Positive psychology dialogue targets everyday distress relief and positive resource building through multi-turn support involving acknowledgment, clarification, resource activation, and action advancement.Source: §1 · p. 1
• Existing resources provide limited supervision for multi-turn process progression and rarely localize preference reasons or process-level defects.Source: §1 · p. 2
• StageWell organizes support into a six-stage process and constructs SFT, DPO, and GroundTruth subsets for Chinese positive psychology dialogue.Source: §1 · pp. 2–3
• A Planner, Writer, Patcher, and Refiner workflow rewrites generic mental-health dialogues into coherent positive-psychology dialogues, producing 12,445 SFT training instances.Source: §1 · p. 2
• HQS covers safety, response quality, and stage consistency, supporting both targeted preference repair and structured evaluation.Source: §1 · pp. 2–3
• StageWell contributes 1,849 DPO preference pairs and provides auditable supervision over support progression and local response quality.Source: §1 · p. 3

• Recent datasets and systems support empathetic, topic-aware, intervention-like, report-based, long-term, and Chinese counseling dialogue responses.Source: §2.1 · p. 3
• The lack of process supervision leaves chosen and rejected responses inheriting ambiguity during preference construction.Source: §2.2 · p. 3
• Preference learning resources commonly emphasize instruction following, harmlessness, or broad helpfulness rather than psychological-support process structure.Source: §2.2 · p. 3
• Strategy-aware emotional-support methods improve alignment or interpretability, yet rarely identify violated stage constraints, localized defects, or appropriate repairs.Source: §2.2 · p. 3

Task Definition and Six-Stage Support Process

• The task requires responses to fit both the current concern and the dialogue’s position within the support process.Warmth alone is insufficient when a response advances too early, repeats empathy, or gives ungrounded advice.Source: §3.1 · p. 3
• StageWell defines six ordered support stages spanning rapport and safety, clarification, strengths and resources, future and hope, goals and strategies, and consolidation.The stages operationalize per-turn decisions about process position, support function, and progression.Source: §3.1 · p. 4

Process-Aligned Data Construction

• StageWell rewrites generic mental-health dialogues into process-aligned positive-psychology dialogues through role, thematic, and risk filtering followed by case-card-guided multi-agent rewriting.The construction workflow is used for dataset creation rather than inference.Source: §3.2 · p. 4
• Whole-dialogue rewriting assigns a global stage sequence, turn functions, and factual boundaries before downstream agents realize and refine each turn.This produces coherent ordered support functions rather than isolated turn edits.Source: §3.2 · p. 4
• DPO pairs fix dialogue history and target stage while contrasting an original flawed output with a targeted repair of a localized defect.This tighter comparison primarily isolates one process-relevant correction instead of varying process position, support function, and surface quality simultaneously.Source: §3.2 · p. 4
• HQS separates hard safety and boundary checks, local response-quality assessment, and stage-consistency analysis for both preference construction and evaluation.Its modules cover safety and risk, five response-quality dimensions, and expected stage, progression pace, and transition fit.Source: §3.3 · p. 5, §3.2 · p. 5

Resource Scale and Basic Statistics

• StageWell’s resource statistics are organized by support stages S1–S6 and HQS response-quality dimensions Q1–Q5.Q1–Q5 denote content fit, acknowledgment, feasibility, maturity, and Chinese naturalness.Source: Table 4 · p. 5, B. Stage distribution · p. 5
• HQS links SFT demonstrations, DPO process-localized preference pairs, and GroundTruth expert references to held-out model evaluation.The protocol is used as a shared framework rather than an isolated metric set.Source: B. Stage distribution · p. 5

Experimental setup

• Experiments use a disjoint GroundTruth set of 120 dialogues and 977 expert-revised QA pairs for held-out evaluation.SFT provides staged demonstrations, DPO provides process-localized preference pairs, and GroundTruth provides expert references.Source: §4.1 · p. 6
• DPO pairs keep dialogue history and target stage fixed while repairing localized defects in rejected model outputs.Earlier-stage repairs address acknowledgment or questioning, whereas later-stage repairs address action grounding, premature closure, or unstable boundaries.Source: §4.1 · p. 6

Evaluation

• HQS evaluates safety, response quality, and stage consistency alongside reference-based metrics on the same held-out data.H-critical is dialogue-level, Q scores turn-level quality, and S evaluates predicted support stages.Source: §4.1 · pp. 6–7

Main results

• S-exact rises from 0.424 to 0.646 with SFT, while H-critical falls from 0.202 to 0.040 across backbones.These averaged results indicate the largest process-control gains come from stage-aware supervised training.Source: §4.2 · p. 7
• DPO further raises Q Overall from 3.821 to 3.993 and S-exact from 0.646 to 0.660, while lowering H-critical from 0.040 to 0.035.The paper characterizes DPO as local refinement over an SFT-established support process rather than uniform improvement on every backbone.Source: §4.2 · p. 7
• Reference-based gains are modest, and GLM-4-9B is a non-monotonic case where H-critical increases from 0.033 to 0.042 after DPO.The increase corresponds to roughly 4 versus 5 critical cases among 120 held-out dialogues.Source: §4.2 · p. 7

Exploratory feasibility and usability pilot

• The pilot reports a higher average immediate psychological-state score from 3.24 to 3.48, a relative increase of 7.4%.It includes 28 students and is intended to assess feasibility, usability, and immediate self-reported experience rather than causal effectiveness.Source: §4.3 · p. 8

Post-interaction experience

• The pilot’s small, uncontrolled, short-term self-report design limits interpretation to preliminary feasibility and usability evidence.It is not designed to estimate causal effectiveness.Source: B. Post-interaction Experience · p. 8

Conclusion

• HQS is a structured protocol for preference construction and evaluation in Chinese positive psychology dialogue.It is presented together with StageWell as a reusable process-oriented resource.Source: §5 · p. 8
• StageWell provides staged whole-dialogue SFT instances, process-localized DPO preference pairs, and a held-out GroundTruth evaluation set.The corpus targets non-clinical support for everyday distress relief and positive resource building.Source: §5 · p. 8
• Across four open-source LLMs, StageWell improves process control, response quality, and safety.The conclusion highlights process-aware supervision as practically valuable for supportive dialogue alignment.Source: §5 · p. 8
• Future work will assess the long-term effectiveness of process-aligned dialogue systems in real-world deployments.The current conclusion frames this as a future research direction.Source: §5 · p. 8

• StageWell’s six-stage process and HQS criteria may not transfer directly to other languages, cultures, or specialized support scenarios.Source: §6 · p. 9
• Constrained rewriting and expert revision prioritize clear process supervision and factual control over the conversational diversity of fully naturalistic data.Source: §6 · p. 9
• Targeted DPO repairs under shared context and stage constraints may cover a narrower range of preference variation than open-ended response comparisons.Source: §6 · p. 9
• The main evaluation relies on held-out expert-revised data and structured automatic judging, complemented by a small exploratory pilot.Source: §6 · p. 9
• Larger independent interactive studies are still needed to assess user-facing effects under extended real-world use.Source: §6 · p. 9

• All research procedures were approved by a formal ethics committee.Source: Ethics Statement · p. 9
• StageWell underwent deidentification and manual inspection to protect participant privacy.Source: Ethics Statement · p. 9
• Training models on StageWell introduces inherent risks because of machine-learning black-box behavior.Source: Ethics Statement · p. 9
• Subtle model biases or inaccuracies may inadvertently affect vulnerable individuals.Source: Ethics Statement · p. 9

Dataset scope

• StageWell covers ten common support topics and maps source dialogues to plausible school, family, peer, or growth scenarios.Source: §A.1 · p. 12

Stage-aware rewriting

• Each assistant turn receives a dialogue tool, required information slot, and support stage, allowing the rewriting skeleton to remain auditable.Source: §A.2 · p. 12

Pipeline modules

• The pipeline combines a Case Card, Planner, Writer, JsonFix, Patcher, Refiner, Validator, and Repeat Gate for structured generation, repair, fluency polishing, and acceptance decisions.Source: §A.2.1 · p. 12

Pipeline constraints

• Planner constraints fix topic assignment, translate adult or workplace concerns into plausible student contexts, and enforce monotonic process progression.Source: §A.2.1 · p. 12

Auditability

• The audit trail retains the normalized Case Card, Planner schedule and facts, labeled Writer dialogue, local edits, and final Validator decision.Source: §A.2.1 · p. 12

Expert revision

• GroundTruth experts revise drafts while preserving topic, turn structure, context, and task scope, targeting expression, empathy, appropriateness, and concrete school-scenario suggestions.Source: §A.3 · pp. 12–14

Dataset checks

• The final SFT topic distribution sums to 12,445 instances, while structural diagnostics check that rewritten dialogues remain well-formed.Source: §A.2.1 · p. 12, Table 8 · p. 13

Training configuration

• DPO uses a smaller learning rate of 5 × 10^-7 and larger gradient accumulation than SFT to stabilize preference alignment.Source: §B.1 · p. 14

Checkpoint selection

• SFT selects the checkpoint with lowest validation loss, whereas DPO jointly monitors training loss and evaluation metrics, prioritizing the highest evaluation metric when they diverge.Source: §B.3 · p. 16, §B.4 · p. 16

Training configuration

• The four base models share identical primary SFT and DPO hyperparameters to ensure comparability across models.Source: §B.5 · p. 16

Evaluation protocol

• HQS separates dialogue-level safety screening, stage-conditioned turn-level quality scoring, and turn-level stage recognition; JSON Repair only recovers formatting.Source: §B.6 · pp. 16–17, §B.6.1 · p. 17
• The Quality Judge uses negative reminders against rewarding long, polished generic, over-read, bookish, or checklist-like replies.Source: §B.6.1 · p. 17

Preference analysis

• Chosen DPO responses are often no longer than rejected responses, suggesting the preference objective is not simply rewarding verbosity.Source: §B.6.1 · p. 17

Preference construction

• Targeted rewrites improve only designated Qk dimensions while preserving dialogue context, stage function, and core semantics.Source: §B.6.2 · p. 17

Results visualization

• Figure 9 visually compares reference-based and HQS metrics across Base, SFT, and DPO conditions.Source: §B.7 · p. 17