Chatbots for Mental Health Support: Exploring the Impact of Emohaa on Reducing Mental Distress in China

1. Introduction

• The introduction situates the study amid rising mental-health concerns and service challenges, including limited professional training and stigma that can contribute to low diagnostic accuracy and treatment delays.It cites pandemic-era increases in depression and anxiety and reports that about 17.4% of 22,330 adults in an international study met criteria for probable insomnia disorder.Source: §1 · p. 2
• In China, 16.57% of adults had lifetime prevalence of mental disorders and 4.98% had 12-month anxiety prevalence; the paper also cites limited appropriate treatment and inadequate service quality and quantity.The passage attributes limited treatment to low mental-health awareness and poor service accessibility.Source: §1 · pp. 2–3
• Conversational agents could increase support availability and reduce the relative costs of seeking it, and prior work found that people were willing to disclose emotional problems to machines.The paper presents self-disclosure as important for providing support and as evidence of user rapport with agents.Source: §1 · p. 3
• The paper distinguishes cognitive support, which helps people reassess situations, from emotional support, which offers validation and understanding to relieve distress.The passage describes human supporters as limited in number and professional help as unaffordable for many.Source: §1 · p. 3
• Prior conversational-agent research mainly delivered CBT-based cognitive support, while machine-based emotional support research was limited and much of the field’s agent research was conducted in English.Prior emotional-support prototypes either lacked empirical effectiveness testing or used predefined statements rather than responses based on users’ messages.Source: §1 · p. 3
• The study investigates a Chinese agent providing both CBT-based cognitive support and open-ended emotional support, assessing effectiveness, acceptability, and practicality among mainland Chinese participants.The authors hypothesized that frequent use involving daily exercises and emotional venting would improve distress symptoms.Source: §1 · p. 3

2.1. Emohaa

• Emohaa has two platforms: a template-based CBT system with predefined options and exercises, and a generative dialogue platform for open-ended emotional support.Source: §2.1 · p. 3

2.1.1. CBT-based

• The CBT platform offers automatic-thought training to identify and replace harmful thoughts with rational perspectives, alongside more than 20 guided expressive-writing exercises.The exercises cover varied topics and guide users through each step via interactive conversations.Source: §2.1.1 · p. 4

2.1.2. ES-based

• For the emotional-support platform, the authors translated the English ESConv dataset into Chinese and used its Helping Skills Theory-based strategies, including self-disclosure, affirmation, and suggestions.The design aimed to let users freely discuss emotional problems and receive responses based on their situation.Source: §2.1.2 · p. 4
• The emotional-support platform uses a Chinese dialogue model to select support strategies from conversation history and generate strategy-conforming responses relevant to users’ messages.The platform also uses a classifier to detect signs of suicidal thoughts and recommends relevant authorities’ contact information when such signs are detected.Source: §2.1.2 · p. 4

2.2.1. PHQ-9

• Depression was measured with the 9-item PHQ-9, which rates symptom frequency and severity over the previous two weeks on a 0–3 scale.The study reported pre-test and post-test Cronbach’s alpha values of .78 and .85.Source: §2.2.1 · p. 4

2.2.2. GAD-7

• Anxiety was measured with the 7-item GAD-7, assessing symptom frequency and severity over the previous two weeks on a 0–3 scale.The scale’s pre-test and post-test Cronbach’s alpha values were both .84.Source: §2.2.2 · p. 4

2.2.3. PANAS

• Affect was measured with the 20-item PANAS, with separate positive- and negative-affect dimensions scored on a 5-point Likert scale.Higher scores indicate higher affect; the passage reports pre-test and post-test Cronbach’s alpha values for both dimensions.Source: §2.2.3 · p. 5

2.2.4. ISI

• Insomnia was measured with the 7-item ISI, which assesses sleep difficulties, interference with daily functioning, and distress from sleep problems.Items were scored from 1 to 4; summed scores classify insomnia from none (0–7) to severe (22–28).Source: §2.2.4 · p. 5

2.3. Participants

• The study randomly assigned eligible participants to Emohaa (CBT-based), Emohaa (Full), or control after recruiting adults able to use smartphones who were not in therapy or experiencing physical issues that might affect psychological state.Of 412 registrants, 301 met the criteria; 30 extra participants were allocated to control because of expected waiting time and potential control-group attrition.Source: §2.3 · p. 5
• Sixteen participants at suicide risk and 38 with more than half their data missing were excluded; 72 CBT-based, 70 full-Emohaa, and 105 control participants completed baseline questionnaires.Participants were asked to complete the same measures one day after the intervention; one-month follow-up data were unavailable for control because participants could receive other interventions.Source: §2.3 · p. 5
• Of 247 randomized participants, 134 (54.2%) provided partial or complete T2 data, and dropout analyses found no significant baseline differences on the reported demographic or symptom measures.The tested measures included age, gender, working tenure, PHQ-9, GAD-7, positive and negative affect, and insomnia.Source: §2.3 · p. 5

2.4.1. Guidelines

• All participants completed mental-distress questionnaires at T1 and T2.Source: §2.4.1 · p. 5
• Intervention participants used the CBT platform daily for an automatic-thinking exercise and guided diary; the Full group also conversed daily for 5–10 turns.Source: §2.4.1 · pp. 5–6

2.4.2. Quality Control

• The researchers checked platform use every three days and monitored conversations; 3 of 34 participants were asked to repeat duplicated messages.Source: §2.4.2 · p. 6
• Table 1 presents user demographics for the studied sample.Source: Table 1 · p. 7
• Table 2 reports analyses of variance in mental-health outcomes; the F-test assesses Condition × Time differences in change rates.Source: Table 2 · p. 7, §2.4.2 · p. 7

• Participants were instructed not to share personal information that could identify them in conversations with Emohaa.Source: §2.5 · p. 7
• Experiment data were anonymized and stored securely, with research access available by request to the corresponding author.Source: §2.5 · p. 7
• The study received Institutional Review Board approval, and participants were informed of its purpose and consented to take part.Source: §2.5 · p. 7

3.1. User Demographics

• Among 134 participants, 107 were female; mean age was 31.81 years, and baseline scores indicated mild depression and anxiety and sub-threshold insomnia.All participants were from mainland China.Source: §3.1 · p. 7

3.2. Effects of Emohaa Intervention

• Both Emohaa groups improved depression, anxiety, and negative affect, while positive affect showed no group-by-time interaction; insomnia declined significantly only for CBT-based Emohaa.The control group’s depression increased, while its anxiety and insomnia did not significantly change; full Emohaa’s insomnia decrease was marginal (p = .053).Source: Figure 2 · p. 8, §3.2 · pp. 8–9

3.2.1. Supplemental Analyses

• At three-week follow-up, insomnia returned to baseline in the CBT-based group but continued improving in the full Emohaa group; anxiety rose again in both groups, with a longer-lasting effect for full Emohaa.Only 16 CBT-based and 27 full Emohaa participants returned follow-up questionnaires; the control group was not followed because participants received other interventions.Source: §3.2.1 · p. 9

3.3. Conversation Analysis

• Participants discussed feelings, work, mood, and study pressure most often, and commonly sought mindfulness, rest, health, and emotional support.They averaged seven conversation sessions, with 17 turns per session on average.Source: §3.3 · p. 9

3.4. Acceptability and Feasibility

• Most users reported satisfaction with CBT-platform functionality (56/69, 81.16%) and rated the emotional-support platform an appropriate chatting partner (24/31, 77.42%); feedback cited technical and content problems.Participants also reported rigid or incoherent generated responses and difficulty maintaining conversational context.Source: §3.4 · pp. 9–10

4.1. Main Findings

• Compared with controls, Emohaa users had significant reductions in depression and anxiety and considerable reductions in negative affect and insomnia.Source: §4.1 · p. 10
• No significant short-term outcome differences emerged between Emohaa groups, but the emotional-support group showed fewer insomnia indicators and a smaller T3 anxiety increase.Source: §4.1 · p. 11
• Most participants were satisfied and would recommend Emohaa; many described the emotional-support platform as understanding, attentive, and a place to vent.Source: §4.1 · p. 11
• Open conversation about users’ chosen topics may complement CBT-based templates and exercises in improving mental distress.Source: §4.1 · p. 11

4.2. Limitations and Future Work

• The short study and limited sample cannot support generalizing the findings to the public; control-group follow-up data were also unavailable.Source: §4.2 · p. 11
• Participants reported technical problems, unclear instructions, and limited exercise content and choices, prompting plans to improve the platform.Source: §4.2 · p. 11
• The emotional-support platform sometimes forgot prior turns or gave contextually irrelevant replies, potentially impairing engagement and rapport.Source: §4.2 · p. 11

• Emohaa combines template-based CBT exercises and guided conversations with a platform for discussing emotional problems.Source: §5 · p. 11
• Participants experienced fewer mental-distress symptoms after using Emohaa during the study.Source: §5 · p. 11
• Adding generative emotional-support dialogue had a complementary effect on improvements, mainly for depression and insomnia.Source: §5 · p. 12

• The CBT interface includes introductory choices and staged diary writing for guided expressive writing.Source: A. User Interface · p. 12, Figure 3 · p. 12
• Automatic-thinking exercises use hypothetical scenarios for perspective-taking, confidence ratings, and learning a different approach.Source: A. User Interface · p. 12, Figure 4 · p. 13