Browse

Journals By Subject

Life Sciences, Agriculture & Food
Chemistry
Medicine & Health
Materials Science
Mathematics & Physics
Electrical & Computer Science
Earth, Energy & Environment
Architecture & Civil Engineering
Education
Economics & Management
Humanities & Social Sciences
Multidisciplinary

Books

Publish Conference Abstract Books
Upcoming Conference Abstract Books
Published Conference Abstract Books
Publish Your Books
Upcoming Books
Published Books

Proceedings

Events

Events
Five Types of Conference Publications

Join

Join as Editorial Board Member
Become a Reviewer

Information

For Authors
For Reviewers
For Editors
For Conference Organizers
For Librarians
Article Processing Charges
Special Issue Guidelines
Editorial Process
Manuscript Transfers
Peer Review at SciencePG
Open Access
Copyright and License
Ethical Guidelines
Submit an Article
Research Article | | Peer-Reviewed

Emotion-Aware AI Chatbots for Mental Health Support in Low-Resource Public Health Systems: A Case Study from Ghana

Evans Ankomah,  Regina Esi Turkson*
Published in World Journal of Public Health (Volume 10, Issue 3)
Received: 5 June 2025     Accepted: 8 July 2025     Published: 23 July 2025
Views:       Downloads:
Download PDF
Abstract

Mental health conditions are on the rise globally, yet many low-resource countries face systemic barriers such as stigma, underfunding, and a severe shortage of professionals to provide adequate care. This paper presents the design, implementation, and evaluation of an emotion-aware AI chatbot for mental health support within Ghana's public health context, aiming to bridge the mental health treatment gap. Leveraging deep learning for emotion detection and integrating a generative language model (GPT-3.5), the system delivers culturally relevant responses to users exhibiting symptoms of emotional distress. The study restructured the International Survey on Emotion Antecedents and Reactions (ISEAR) dataset into a 5-emotion model (Joy, Fear, Anger, Sadness, Neutral) to improve classification accuracy. A Convolutional Neural Network (CNN) emerged as the top-performing classifier (76.4% accuracy), outperforming LSTM, BiLSTM, and GRU models. This classifier was integrated with GPT-3.5 to enable context-aware, empathetic interactions. Field testing with 311 participants in Ghana revealed high satisfaction: 89% praised usability, 81% affirmed cultural relevance, and 78% reported emotional support. Notably, 66% felt encouraged to seek professional care, demonstrating the chatbot’s potential as a gateway to formal mental health services. The system’s anonymity and 24/7 accessibility addressed key barriers like stigma and resource limitations. The findings suggest that emotion-aware chatbots can complement mental health services in under-resourced settings and offer an innovative pathway for public health outreach. Future work will expand language options and crisis protocols. This research contributes a scalable, cost-effective model for global public health, emphasizing cultural adaptation and emotion-aware AI as critical tools in mental health innovation.

Published in World Journal of Public Health (Volume 10, Issue 3)
DOI 10.11648/j.wjph.20251003.17
Page(s) 265-272
Creative Commons

This is an Open Access article, distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium or format, provided the original work is properly cited.

Copyright

Copyright © The Author(s), 2025. Published by Science Publishing Group
Previous article
Keywords

Mental Health, Chatbot, Emotion Detection, Deep Learning, AI in Public Health, Ghana, GPT-3.5, Low-Resource Settings

1. Introduction
Mental health has become a critical public health issue worldwide, particularly in low- and middle-income countries (LMICs), where access to professional care is severely limited by stigma, geography, cost, and systemic underfunding. The World Health Organization (WHO) estimates that one in ten individuals globally need mental health care, yet in low-income countries, there are only two mental health professionals for every 100,000 people
[1]
. In Ghana, approximately 2.4 million people suffer from mental illness, but only 2% of them receive psychiatric support
[2, 3]
. Structural challenges—including underfunding (only 1.4% of the national health budget is allocated to mental health), outdated policy frameworks, and a shortage of qualified personnel—compound the problem
[3]
.
In response to these limitations, mental health care in Ghana often relies on informal providers such as traditional and faith-based healers, which vary in quality and effectiveness
[4]
. Meanwhile, community-based surveys show high psychological distress levels, with nearly 30% of Ghanaians reporting mental distress and 12.7% meeting the criteria for depressive disorders
[3]
. These data underscore the need for scalable and accessible solutions to support the nation’s mental health infrastructure.
Chatbot systems have emerged as promising digital tools to address mental health issues, offering anonymity, 24/7 availability, and affordability. Globally, systems like Woebot, Wysa, and Tess have shown effectiveness in reducing symptoms of depression and anxiety
[5, 6]
. In Ghana, research has suggested that both mental health professionals and patients are open to using mobile-based or internet-enabled interventions for mental health management
[7]
. Chatbots are capable of proactive emotional engagement, reminder services, and empathetic responses, and can be trained to understand psychological triggers using deep learning techniques
[8]
.
Despite their potential, existing chatbot systems are often Western-centric and lack cultural sensitivity or language adaptability for African contexts
[9]
. In Ghana, no existing framework robustly integrates localized emotion detection with generative Artificial intelligence (AI) to support mental health.
This research seeks to bridge this gap by developing and implementing an emotion-aware chatbot tailored to Ghana’s public mental health landscape. The goal is to enhance access to care, address anxiety and depression, and complement the limited human resource capacity with AI-based support. In doing so, we contribute to the broader effort to reduce Ghana’s mental health treatment gap and build resilient digital public health infrastructure.
2. Background and Related Work
2.1. Mental Health and the Rise of Digital Interventions
Mental health disorders represent a growing global concern, particularly in low- and middle-income countries (LMICs), where more than 75% of individuals with mental illness receive no treatment
[1, 10]
. The shortfall is largely due to limited human resources, stigma, and weak health infrastructure. In Ghana, only 2% of the estimated 2.4 million people living with mental illness have access to professional care
[3]
. As a result, digital health innovations are increasingly seen as viable tools to extend care in under-resourced settings.
Among these, AI-powered chatbot systems have emerged as promising tools for delivering psychological support
[7]
. Chatbots offer round-the-clock availability, anonymity, and scalability—benefits particularly relevant in environments where human therapists are scarce. During the COVID-19 pandemic, chatbots helped triage mental health concerns and alleviate the burden on traditional care systems
[11]
. Additionally, mental health professionals in both high- and low-resource settings have expressed openness to using digital mental health interventions
[5]
.
2.2. Chatbot Systems in Mental Health
Chatbots such as Woebot, Wysa, and Tess have been used effectively to help individuals manage anxiety, depression, and stress through structured, evidence-based interactions
[6, 12]
. These systems typically follow cognitive behavioral therapy (CBT) frameworks and use scripted or rule-based dialogues. While effective, they lack adaptability, and many fail to resonate with users from diverse cultural backgrounds.
Research by
[9]
that user feedback on mental health chatbots is generally positive, especially in terms of convenience and emotional safety. However, most existing systems are developed with Western populations in mind, raising questions about their cultural sensitivity and efficacy in African contexts
[9]
.
2.3. Evolution of Conversational Agents
Conversational systems have evolved significantly—from early programs like Eliza in the 1960s to modern large language models such as GPT-3.5. These advances have improved chatbots’ ability to handle natural language, maintain context, and demonstrate empathy. Today’s chatbots are broadly categorized as:
1) Retrieval-based: Use pre-scripted responses and rule-based decision trees
[13]
.
2) Generative-based: Use neural networks and deep learning to generate human-like responses dynamically
[14]
.
Generative chatbots are particularly suited to mental health use cases due to their flexibility and capacity for personalization
[15]
.
2.4. Emotion Detection and Cultural Adaptation
Emotion recognition plays a key role in enabling chatbots to deliver empathetic responses. Deep learning models such as Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), Bidirectional LSTM, and Gated Recurrent Units (GRU) have shown high performance in emotion classification tasks using datasets like ISEAR
[16]
. Simplifying the dataset from seven to five emotion classes: Joy, Fear, Sadness, Anger (Anger + Guilt) and Neutral (Shame + Guilt) improves accuracy and interpretability for real-world applications
[17, 18]
.
Despite progress in emotion-aware chatbot development, few systems are designed to accommodate local languages, idioms, or belief systems prevalent in Ghana. The over-reliance on Western datasets and frameworks limits the cultural appropriateness of current AI tools for mental health. This underscores the need for contextualized, culturally sensitive models to improve adoption and therapeutic impact
[19]
.
2.5. Research Gaps and Justification
While chatbot interventions are well-documented in global literature, gaps remain in their deployment within African mental health systems. Particularly underexplored are generative models that adapt to emotional inputs in culturally meaningful ways. Moreover, few studies combine deep learning-based emotion detection with real-time AI-generated counseling dialogue.
This study addresses these gaps by developing an emotion-aware chatbot tailored for Ghana. It integrates CNN-based emotion classification with GPT-3.5-turbo for context-aware, empathetic dialogue generation. The chatbot is designed to support individuals experiencing anxiety and depression, offering scalable, accessible care in line with Ghana’s public health objectives 3.
3. Methodology
This study adopted a structured design methodology to develop and evaluate an AI-powered, emotion-aware chatbot for mental health support in Ghana. The system integrates a deep learning-based emotion classification model with a large language model to generate culturally relevant and emotionally responsive conversations. The methodological framework included four major components: emotion model restructuring, Natural Language Processing (NLP) preprocessing, deep learning training, and system deployment.
3.1. Dataset and Emotion Categorization
The emotion classifier was trained using the International Survey on Emotion Antecedents and Reactions (ISEAR) dataset, which includes over 7,000 manually labeled emotional expressions
[19]
. Originally labeled with seven categories—joy, fear, anger, sadness, disgust, shame, and guilt—the labels were restructured into five core categories: joy, fear, anger, sadness (merging shame), and disgust (merging guilt), following precedent from affective psychology and emotion classification studies
[17]
.
This reclassification was aimed at improving classification accuracy and simplifying deployment in real-time applications. Figure 1 below shows the flow of the Data preparation, training and best model selection.
Figure 1. Best model selection.
3.2. Preprocessing and Feature Representation
To prepare the text data for model training, several standard NLP preprocessing steps were applied:
1) Contraction expansion (e.g., “don’t” → “do not”)
2) Punctuation and stopword removal
3) Text normalization (lowercasing, lemmatization)
4) Tokenization and sequence padding to a fixed length of 300 tokens
Feature representation was achieved using GloVe word embeddings (100-dimensional), which are pretrained on global corpora and proven effective for capturing semantic relationships between words
[20]
.
3.3. Deep Learning Models
Four deep learning models were trained and compared:
1) CNN - Used to extract n-gram-like features from sequences of word embeddings
[21]
.
2) LSTM - Designed to capture long-range dependencies in sequential text
[22]
.
3) BiLSTM - Processes text in both forward and backward directions for richer context
[23]
.
4) GRU - An efficient alternative to LSTM with comparable performance
[24]
.
All models were implemented using Keras and TensorFlow, trained with categorical cross-entropy loss, Adam optimizer, batch size of 500, and early stopping. The CNN model outperformed others, achieving a classification accuracy of 76.4% on the 5-emotion model, with high precision and F1-score.
3.4. Chatbot System Integration
The best-performing model (CNN) was integrated into a conversational pipeline with OpenAI’s GPT-3.5-turbo language model. The chatbot interaction sequence as seen in Figure 2 involved:
1) User input is passed through the CNN model.
2) The predicted emotion is used to construct a prompt.
3) GPT-3.5 generates a response tailored to the emotional context.
4) The message is delivered to the user via a web interface.
This hybrid system leverages the precision of a trained classifier and the flexibility of a generative language model to produce empathetic, context-aware dialogue, enhancing user engagement and realism.
Figure 2. Chatbot Workflow.
3.5. System Deployment and Evaluation
A web-based chatbot interface was developed using Flask (Python backend) and JavaScript frontend, enabling device-agnostic interaction. A pilot study was conducted involving 100 participants from diverse demographics in Ghana. Post-chat evaluations measured:
1) Ease of use
2) Cultural relevance
3) Emotional resonance
4) Likelihood of reuse
5) Willingness to seek professional care
The feedback confirmed the system’s practical utility and public health potential.
4. Results
The performance of the emotion classification models and the overall usability of the chatbot system were evaluated in two phases: (1) quantitative evaluation of deep learning models and (2) user-based evaluation of the deployed chatbot prototype. Figure 4 below shows the interface of the Chatbot built.
Figure 3. Chatbot's interface.
4.1. Model Evaluation
Four deep learning models—CNN, LSTM, BiLSTM, and GRU—were trained and tested using the 5-emotion classification scheme derived from the ISEAR dataset. The CNN model outperformed all others with an accuracy of 76.4%, precision of 78.3%, recall of 77.6%, and an F1-score of 76.9%. BiLSTM and GRU followed closely but were slightly less consistent, particularly in classifying overlapping emotions like sadness and disgust.
The confusion of the CNN model matrix as shown in Figure 4 below showed high classification of confidence for distinct emotions like joy and fear, with most misclassifications occurring between sadness and disgust. This validated the decision to merge related emotion categories and reduce emotional ambiguity during training.
Figure 4. Confusion Metrix.
4.2. Chatbot Usability Evaluation
The integrated chatbot system—comprising the CNN emotion classifier and GPT-3.5 for response generation—was deployed via a web interface and tested by 311 participants in Ghana. Evaluation metrics focused on user experience, cultural appropriateness, and perceived emotional support.
Key user feedback included:
1) Ease of Use: 89% found the system intuitive and accessible. See Figure 6.
2) Cultural Relevance: 81% agreed that the responses reflected Ghanaian social realities and communication styles. Refer to Figure 5.
3) Emotional Support: 78% felt the chatbot provided empathetic and helpful responses. Refer to Figure 8.
4) Help-Seeking Behavior: 66% said the interaction encouraged them to consider seeking professional mental health support. Refer to Figure 9
5) Willingness to Reuse: 84% expressed readiness to engage with the chatbot again in the future. Refer to Figure 7
Users also appreciated the anonymity and 24/7 availability of the system, especially in contrast to in-person therapy sessions, which are difficult to access due to limited personnel and stigma.
Figure 5. Cultural relevance.
Figure 6. Ease of Use.
Figure 7. Willingness to Reuse.
Figure 8. Emotional Support.
Figure 9. Ability to connect to psychiatric experts.
5. Discussion
This study demonstrates the feasibility and impact of deploying an emotion-aware AI chatbot to support mental health in low-resource settings like Ghana. By combining a CNN-based emotion classifier with GPT-3.5, the system delivers context-sensitive and culturally relevant responses to users experiencing emotional distress. The model's strong classification performance and positive user feedback validate the effectiveness of this hybrid approach.
User evaluation revealed high satisfaction with the system’s usability, cultural appropriateness, and emotional responsiveness. These findings are particularly meaningful in a context where mental health stigma and professional shortages hinder access to care. The chatbot offers a private, always-available alternative to traditional support methods.
However, limitations remain. The system currently supports only English, requires continuous internet access, and lacks crisis intervention features. Addressing these challenges—through language expansion, offline capabilities, and escalation protocols—will be critical for wider deployment and long-term sustainability.
6. Conclusions
This study addressed the critical shortage of accessible mental health care in Ghana, where systemic barriers such as stigma and underfunding leave 98% of the affected individuals without treatment. To bridge this gap, we developed an emotion-aware AI chatbot that combines CNN-based emotion classification (76.4% accuracy) with GPT-3.5-generated responses, offering scalable, culturally adapted support. Key findings from field tests with 311 participants demonstrated high user satisfaction, improved perceived access to support, and potential for scaling as a public health intervention. With about 66% reporting increased willingness to seek professional care, emotion-aware AI chatbot proves to be a promising step toward reducing the treatment gap. Despite the chatbot’s 24/7 availability and anonymity in addressing critical barriers in low-resource settings, limitations such as English-only support and internet dependency must be overcome to maximize impact. Future work should explore the integration of local languages (e.g., Twi, Ga) and offline functionality to reach underserved populations. Additionally, more work is needed on developing a strong crisis escalation protocol to connect high-risk users with human providers.
Abbreviations

GPT

Generative Language Model

ISEAR

International Survey on Emotion Antecedents and Reactions

AI

Artificial Intelligence

CNN

Convolutional Neural Network

WHO

World Health Organization

CBT

Cognitive Behavioral Therapy

LSTM

Long Short-Term Memory

GRU

Gated Recurrent Units

NLP

Natural Language Processing
Conflicts of Interest
The authors declare no conflicts of interest.
References
[1] WHO Team, “Mental Health Atlas 2020,” WHO Publication, pp. 1-136, 2021, Accessed: Feb. 13, 2023. [Online]. Available:

https://www.who.int/publications/i/item/9789240036703

[2] M. Roberts, J. Asare, C. Mogan, E. Adjase, and A. Osei, “The mental health system in Ghana-WHO AIMS report,” Accra: Kintampo Project/Ghana Ministry of Health, 2013.
[3] M. Roberts, C. Mogan, and J. B. Asare, “An overview of Ghana’s mental health system: Results from an assessment using the World Health Organization’s Assessment Instrument for Mental Health Systems (WHO-AIMS),” Int J Ment Health Syst, vol. 8, no. 1, pp. 1-13, May 2014.
[4] E. Badu, A. P. O’Brien, and R. Mitchell, “An integrative review of potential enablers and barriers to accessing mental health services in Ghana,” Health Res Policy Syst, vol. 16, no. 1, pp. 1-19, Nov. 2018.
[5] K. K. Fitzpatrick, A. Darcy, and M. Vierhile, “Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): a randomized controlled trial,” JMIR Ment Health, vol. 4, no. 2, p. e7785, 2017.

https://doi.org/10.2196/mental.7785

[6] S. D’Alfonso, “AI in mental health,” Curr Opin Psychol, vol. 36, pp. 112-117, Dec. 2020.

https://doi.org/10.1016/j.copsyc.2020.04.005

[7] N. Berry, S. Bucci, and F. Lobban, “Use of the Internet and Mobile Phones for Self-Management of Severe Mental Health Problems: Qualitative Study of Staff Views,” JMIR Ment Health 2017; 4(4): e52. vol. 4, no. 4, p. e8311, Nov. 2017.

https://doi.org/10.2196/mental.8311

[8] H. Chin et al., “The Potential of Chatbots for Emotional Support and Promoting Mental Well-Being in Different Cultures: Mixed Methods Study,” J Med Internet Res, vol. 25, p. e51712, 2023.

https://doi.org/10.2196/51712

[9] A. A. Abd-Alrazaq, M. Alajlani, N. Ali, K. Denecke, B. M. Bewick, and M. Househ, “Perceptions and Opinions of Patients About Mental Health Chatbots: Scoping Review,” J Med Internet Res 2021; 23(1): e17828 vol. 23, no. 1, p. e17828, Jan. 2021.

https://doi.org/10.2196/17828

[10] K. S. Jacob and V. Patel, “Classification of mental disorders: A global mental health perspective,” The Lancet, vol. 383, pp. 1433-1435, 2014.
[11] Y. Zhu, M. Janssen, R. Wang, and Y. Liu, “It Is Me, Chatbot: Working to Address the COVID-19 Outbreak-Related Mental Health Issues in China. User Experience, Satisfaction, and Influencing Factors,” Int J Hum Comput Interact, vol. 38, no. 12, pp. 1182-1194, Jul. 2022.

https://doi.org/10.1080/10447318.2021.1988236

[12] M. Bjaaland Skjuve and P. Bae Brandtzaeg, “Chatbots as a new user interface for providing health information to young people,” 978-91-88855-02-2, 2018, Accessed: Feb. 13, 2023. [Online]. Available:

https://www.duo.uio.no/handle/10852/71503

[13] T. Sharma, T. Singh Rathore, and N. Singh, “RETRIEVAL BASED CHATBOT SYSTEM,”

www.irjmets.com

[14] S. Hussain, O. Ameri Sianaki, and N. Ababneh, “A survey on conversational agents/chatbots classification and design techniques,” in Web, Artificial Intelligence and Network Applications: Proceedings of the Workshops of the 33rd International Conference on Advanced Information Networking and Applications (WAINA-2019) 33, Springer, 2019, pp. 946-956.

https://doi.org/10.1007/978-3-030-15035-8_93

[15] D. Calvaresi, A. Ibrahim, J.-P. Calbimonte, R. Schegg, E. Fragniere, and M. Schumacher, “The Evolution of Chatbots in Tourism: A Systematic Literature Review,” Information and Communication Technologies in Tourism 2021, pp. 3-16, 2021.

https://doi.org/10.1007/978-3-030-65785-7_1

[16] E. S. Dan-Glauser and K. R. Scherer, “The Geneva affective picture database (GAPED): A new 730-picture database focusing on valence and normative significance,” Behav Res Methods, vol. 43, no. 2, pp. 468-477, Jun. 2011.
[17] J. P. Tangney, J. Stuewig, and D. J. Mashek, “Moral Emotions and Moral Behavior,” 2006.

https://doi.org/10.1146/annurev.psych.56.091103.070145.

[18] H. A. Chapman and A. K. Anderson, “Understanding disgust,” Ann N Y Acad Sci, vol. 1251, no. 1, pp. 62-76, Mar. 2012.

https://doi.org/10.1111/j.1749-6632.2011.06369.x

[19] K. R. Scherer and H. G. Wallbott, “Evidence for Universality and Cultural Variation of Differential Emotion Response Patterning,” J Pers Soc Psychol, vol. 66, no. 2, pp. 310-328, 1994.
[20] J. Pennington, R. Socher, and C. D. Manning, “GloVe: Global Vectors for Word Representation,” EMNLP 2014 - 2014 Conference on Empirical Methods in Natural Language Processing, Proceedings of the Conference, pp. 1532-1543, 2014.
[21] J. Kim, Y. Kim, B. Kim, S. Yun, M. Kim, and J. Lee, “Can a machine tend to teenagers’ emotional needs? A study with conversational agents,” in Extended abstracts of the 2018 CHI conference on human factors in computing systems, 2018, pp. 1-6.

https://doi.org/10.1145/3170427.3188548

[22] H. Sepp and S. Jurgen, “LONG SHORTTERM MEMORY,” Neural Comput, 1997, Accessed: Jun. 04, 2025. [Online]. Available:

https://www.bioinf.jku.at/publications/older/2604.pdf

[23] M. Schuster and K. K. Paliwal, “Bidirectional recurrent neural networks,” IEEE Transactions on Signal Processing, vol. 45, no. 11, pp. 2673-2681, 1997.
[24] K. Cho et al., “Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation,” EMNLP 2014 - 2014 Conference on Empirical Methods in Natural Language Processing, Proceedings of the Conference, pp. 1724-1734, 2014.
Cite This Article
Plain Text BibTeX RIS

  • APA Style

    Ankomah, E., Turkson, R. E. (2025). Emotion-Aware AI Chatbots for Mental Health Support in Low-Resource Public Health Systems: A Case Study from Ghana. World Journal of Public Health, 10(3), 265-272. https://doi.org/10.11648/j.wjph.20251003.17

    Copy | Download

    ACS Style

    Ankomah, E.; Turkson, R. E. Emotion-Aware AI Chatbots for Mental Health Support in Low-Resource Public Health Systems: A Case Study from Ghana. World J. Public Health 2025, 10(3), 265-272. doi: 10.11648/j.wjph.20251003.17

    Copy | Download

    AMA Style

    Ankomah E, Turkson RE. Emotion-Aware AI Chatbots for Mental Health Support in Low-Resource Public Health Systems: A Case Study from Ghana. World J Public Health. 2025;10(3):265-272. doi: 10.11648/j.wjph.20251003.17

    Copy | Download 

  • @article{10.11648/j.wjph.20251003.17,
  author = {Evans Ankomah and Regina Esi Turkson},
  title = {Emotion-Aware AI Chatbots for Mental Health Support in Low-Resource Public Health Systems: A Case Study from Ghana
},
  journal = {World Journal of Public Health},
  volume = {10},
  number = {3},
  pages = {265-272},
  doi = {10.11648/j.wjph.20251003.17},
  url = {https://doi.org/10.11648/j.wjph.20251003.17},
  eprint = {https://article.sciencepublishinggroup.com/pdf/10.11648.j.wjph.20251003.17},
  abstract = {Mental health conditions are on the rise globally, yet many low-resource countries face systemic barriers such as stigma, underfunding, and a severe shortage of professionals to provide adequate care. This paper presents the design, implementation, and evaluation of an emotion-aware AI chatbot for mental health support within Ghana's public health context, aiming to bridge the mental health treatment gap. Leveraging deep learning for emotion detection and integrating a generative language model (GPT-3.5), the system delivers culturally relevant responses to users exhibiting symptoms of emotional distress. The study restructured the International Survey on Emotion Antecedents and Reactions (ISEAR) dataset into a 5-emotion model (Joy, Fear, Anger, Sadness, Neutral) to improve classification accuracy. A Convolutional Neural Network (CNN) emerged as the top-performing classifier (76.4% accuracy), outperforming LSTM, BiLSTM, and GRU models. This classifier was integrated with GPT-3.5 to enable context-aware, empathetic interactions. Field testing with 311 participants in Ghana revealed high satisfaction: 89% praised usability, 81% affirmed cultural relevance, and 78% reported emotional support. Notably, 66% felt encouraged to seek professional care, demonstrating the chatbot’s potential as a gateway to formal mental health services. The system’s anonymity and 24/7 accessibility addressed key barriers like stigma and resource limitations. The findings suggest that emotion-aware chatbots can complement mental health services in under-resourced settings and offer an innovative pathway for public health outreach. Future work will expand language options and crisis protocols. This research contributes a scalable, cost-effective model for global public health, emphasizing cultural adaptation and emotion-aware AI as critical tools in mental health innovation.},
 year = {2025}
}

    Copy | Download 

  • TY  - JOUR
T1  - Emotion-Aware AI Chatbots for Mental Health Support in Low-Resource Public Health Systems: A Case Study from Ghana

AU  - Evans Ankomah
AU  - Regina Esi Turkson
Y1  - 2025/07/23
PY  - 2025
N1  - https://doi.org/10.11648/j.wjph.20251003.17
DO  - 10.11648/j.wjph.20251003.17
T2  - World Journal of Public Health
JF  - World Journal of Public Health
JO  - World Journal of Public Health
SP  - 265
EP  - 272
PB  - Science Publishing Group
SN  - 2637-6059
UR  - https://doi.org/10.11648/j.wjph.20251003.17
AB  - Mental health conditions are on the rise globally, yet many low-resource countries face systemic barriers such as stigma, underfunding, and a severe shortage of professionals to provide adequate care. This paper presents the design, implementation, and evaluation of an emotion-aware AI chatbot for mental health support within Ghana's public health context, aiming to bridge the mental health treatment gap. Leveraging deep learning for emotion detection and integrating a generative language model (GPT-3.5), the system delivers culturally relevant responses to users exhibiting symptoms of emotional distress. The study restructured the International Survey on Emotion Antecedents and Reactions (ISEAR) dataset into a 5-emotion model (Joy, Fear, Anger, Sadness, Neutral) to improve classification accuracy. A Convolutional Neural Network (CNN) emerged as the top-performing classifier (76.4% accuracy), outperforming LSTM, BiLSTM, and GRU models. This classifier was integrated with GPT-3.5 to enable context-aware, empathetic interactions. Field testing with 311 participants in Ghana revealed high satisfaction: 89% praised usability, 81% affirmed cultural relevance, and 78% reported emotional support. Notably, 66% felt encouraged to seek professional care, demonstrating the chatbot’s potential as a gateway to formal mental health services. The system’s anonymity and 24/7 accessibility addressed key barriers like stigma and resource limitations. The findings suggest that emotion-aware chatbots can complement mental health services in under-resourced settings and offer an innovative pathway for public health outreach. Future work will expand language options and crisis protocols. This research contributes a scalable, cost-effective model for global public health, emphasizing cultural adaptation and emotion-aware AI as critical tools in mental health innovation.
VL  - 10
IS  - 3
ER  -

    Copy | Download

Author Information
Download PDF
Submit an Article
  • Abstract
  • Keywords

  • Document Sections

    1. 1. Introduction
    2. 2. Background and Related Work
    3. 3. Methodology
    4. 4. Results
    5. 5. Discussion
    6. 6. Conclusions
    Show Full Outline
  • Abbreviations
  • Conflicts of Interest
  • References
  • Cite This Article
  • Author Information

About Us

Science Publishing Group (SciencePG) is an Open Access publisher, with more than 300 online, peer-reviewed journals covering a wide range of academic disciplines.

Learn More About SciencePG

Products

Information

Important Link

Copyright © 2012 -- 2025 Science Publishing Group – All rights reserved.