AN INTUITIVE AI-DRIVEN SYSTEM TO
EMPOWER EMOTIONAL WELL-BEING
THROUGH SMART JOURNALING AND
PERSONALIZED ACTION PLANS
Courtney Keyi Lee 1, Jonathan Thamrun 2
1 Northwood High School, 4515 Portola Pkwy, Irvine, CA 92620
2 Computer Science Department, California State Polytechnic University,
Pomona, CA 91768
ABSTRACT
Recent years have seen an increase in mental health challenges, particularly depression
and anxiety, exacerbated by barriers such as stigma, cultural differences, and the lack of
readily accessible treatment. This paper proposes an AI-driven journaling app as a
solution to address these issues by offering personalized, culturally sensitive, and engaging
mental health support. The app leverages Natural Language Processing (NLP) and
adaptive learning to generate tailored prompts and advice, fostering emotional awareness
and self-reflection. Key challenges included interpreting complex emotions and addressing
cultural and generational nuances. These were tackled through dynamic AI models, curated
datasets, and iterative refinement. Experiments tested the app’s accuracy in emotion
detection and its ability to adapt advice to diverse contexts, achieving promising results in
relevance and cultural alignment. By integrating inclusivity and real-time feedback, this
app offers a practical, scalable tool for mental well-being, making it an accessible
alternative to traditional mental health support.
KEYWORDS
Natural Language Processing, Flutter, Emotional Analysis, Dart
1. INTRODUCTION
In today’s fast-paced society, the emphasis on mental well-being has become increasingly
prominent as individuals navigate the complexities of balancing personal, academic, and
professional obligations. Depression among U.S. adults rose from 8.7% in 2017 to 14.4% by
April 2020, spurred by the COVID-19 pandemic [1]. This rise in depression has hit certain
groups particularly hard, including individuals with low socioeconomic status, minimal savings,
or those facing unstable employment and related stressors [2][3].
Depression affects the daily lives for nearly two-thirds of those diagnosed [4]. 225 million
workdays and $36.6 billion in lost productivity each year are attributed to major depression, with
major depression resulting in an average loss of 5.6 hours of productive work time per week.
Unfortunately, the treatment of depression in primary care settings is hindered by various factors,
including personal stigma, physician challenges, and limitations within healthcare systems.
People with higher personal stigma preferred to cope with depression alone. High dropout rates
David C. Wyld et al. (Eds): BDAB, SVC, AIBD, ACSTY, NATP – 2025
pp. 163-173, 2025. CS & IT - CSCP 2025 DOI: 10.5121/csit.2024.150114
164 Computer Science & Information Technology (CS & IT)
for both medication and counseling further worsen outcomes, leaving symptoms unmanaged and
often leading to a cycle of worsening mental health [5].
Suicidality and depression severity is often caused by negative problem orientation (pessimistic
view) and an impulsive problem-solving style [6]. Journaling fosters self-reflection, enhancing
awareness and therapeutic skills alongside other treatments [7].
Therapy and medication are essential aspects of mental health treatment, however, they are not
accessible for everyone. In many cases, people are prevented from seeking help due to financial
constraints or stigma. Additionally, not all individuals require intensive treatment; some benefit
greatly from daily mental health practices, like journaling, to manage stress and build resilience.
Methodology A: Emotionally Intelligent Chatbots (e.g., Wysa)
Emotionally intelligent chatbots facilitate the development of emotional rapport and guide users
through evidence-based therapeutic exercises, thereby enhancing engagement with Cognitive
Behavioral Therapy (CBT). However, these chatbots typically rely on decision-tree algorithms,
which can constrain their adaptability and effectiveness, particularly in culturally diverse or
complex scenarios where nuanced understanding is required.
Methodology B:AI-Powered Conversational Agents
Conversational agents simulate human interactions to provide immediate mental health support
and guidance through structured exercises like CBT. They aim to reduce barriers such as stigma
and accessibility. However, they rely on pre-defined scripts and lack nuanced understanding,
cultural sensitivity, and crisis management capabilities.
Methodology C: NLP for Analyzing Mental Health Data
NLP techniques analyze text to detect emotional states and mental health conditions, supporting
early intervention efforts. While effective in large-scale assessments, these systems are limited by
biases in training data, lack of context, and static models.
In an effort to address the rising rates of depression and the barriers to effective treatment, an AI-
driven journaling app is proposed to provide personalized support and practical advice for mental
well-being, designed to complement existing treatment methods. This solution leverages the self-
reflective nature of journaling, which has been shown to foster greater self-awareness and
enhance therapeutic skills among users [7]. By utilizing AI algorithms, the application delivers
personalized prompts and resources, adapting dynamically to the unique needs and emotional
states of each user, facilitating structured self-reflection that mitigates the impersonal nature of
generalized interventions. Moreover, the app's private and discrete format reduces the stigma
often associated with seeking mental health support [8].
Compared to traditional methods such as in-person therapy this solution provides immediate and
on-demand access to mental health resources, enabling users to engage with their emotions
anytime and anywhere. This is crucial for those who may feel hesitant to seek help through
conventional channels.
Plus, compared to mental health apps the app’s AI continuously adjusts based on user
interactions, evolving to meet the user's current emotional state and goals. This adaptability
allows for a more dynamic support system compared to apps that offer a static set of exercises or
Computer Science & Information Technology (CS & IT) 165
one-time assessments. In addition, through this AI component the app allows users to explore at
their own pace and choose what feels relevant to them at any moment.
By facilitating self-reflection and providing practical coping strategies, this innovative approach
has the potential to improve mental health outcomes significantly and create a supportive
community for individuals facing similar challenges.
In Experiment 1, the objective is to assess AI’s ability to detect emotions and generate relevant
advice, rating its performance on a 1 to 5 scale. The results highlight strong performance with
straightforward entries but struggling with complex emotions. Experiment 2 evaluated cultural
and generational sensitivity. The dataset included entries representing collectivist and
individualist cultural orientations, as well as perspectives from both older and younger
generational cohorts. The AI showed high accuracy for general patterns but struggled with
nuanced cultural and generational language styles. These findings emphasize the need for refined
models to improve nuanced understanding and tailored advice.
2. CHALLENGES
In order to build the project, a few challenges have been identified as follows.
2.1. NLP system
A major component of the AI-driven journaling app is the Natural Language Processing (NLP)
system that analyzes user inputs to generate personalized prompts and feedback. This system
must navigate the complex and nuanced landscape of mental health, recognizing that
interventions cannot be uniformly applied across users. An adaptive learning algorithm enables
the AI to refine responses based on user feedback and behaviors, ensuring prompts evolve to stay
relevant and engaging.
Moreover, the NLP system must consider key information about users’ backgrounds, cultural
contexts, and mental health preferences. This is crucial, as individuals from diverse cultural
backgrounds may express themselves differently and have diverse attitudes towards mental
health, as well as different ways to express their thoughts and feelings[9]. To address this
potential bias, a multi-faceted data collection process could be implemented, gathering users’
contextual information through an initial survey at the start of the app and refining this data over
time as users journal more consistently. This approach would enable the AI to better understand
the cultural and personal nuances of each user, allowing for more accurate and culturally
sensitive responses. Additionally, an inclusive language model avoids assumptions about users'
experiences and mental health.
Eliminating biases in training data and updating models ensures AI-generated responses resonate
broadly. Taking a holistic approach to mental health not only enhances user satisfaction but also
creates a more inclusive environment where users feel understood and supported.
2.2. Maintaining User Privacy and Data Security
Another challenge is maintaining user privacy and data security. The user may be reluctant to
share personal information if they are uncertain as to how the data will be used or protected. To
maintain user trust, the app should implement strict encryption, transparent policies, and user-
controlled data-sharing options, with reminders and accessible data management tools
166 Computer Science & Information Technology (CS & IT)
2.3. User Engagement
User engagement is a common issue in mental health apps, as individuals often struggle to
maintain consistent habits, such as journaling, which are essential for long-term mental health
benefits. This challenge stems from factors like motivation fatigue, a lack of immediate results, or
initial discomfort with self-reflection. To boost engagement, the app can include reminders,
achievements, rewards, and community features for shared accountability. By proactively
tackling engagement challenges, the app can more effectively support users on their mental
health journeys.
3. SOLUTION
The program’s main structure is built around three core components: the journaling page, the
calendar page, and the analysis page. Flutter was used to develop the app's User Interface (UI),
while Dart was used to implement the app's logic. A color palette of yellows, pinks, and purples
enhances the user experience by fostering warmth and friendliness, encouraging regular use.
When users first enter into the app they are faced with the journaling page. This page’s primary
use is to record data on a user’s daily reflections, thoughts, and emotions for mental health
analysis, allowing users to freely express their experiences, struggles, and achievements. To
encourage detailed entries, the journaling page includes options for prompts, helping users
document their thoughts in a structured yet flexible way. After journaling, users are redirected to
the calendar page, which serves as a timeline of their mental health journey, allowing quick
reviews of entries and trends.
Clicking a calendar date opens the analysis page, powered by Google’s Generative AI. The
pretrained AI utilizes NLP to analyze journal entries, mood trends, and behavioral patterns to
identify mental well being indicators. This analysis generates insights on the user’s mental state,
offering guidance on potential stress management techniques or highlighting areas for personal
growth through a 5-step actionable plan. By offering personalized reflections, the analysis page
supports users in their self-reflection and personal improvement journey.
The purpose of the journaling page for users is providing a space where they can record daily
reflections and track their mental health over time. But before users can access the journaling
page, the app includes a user authentication page. This ensures that each user’s data is unique and
securely linked to their account, preventing unauthorized access and maintaining the accuracy of
the information.
To store the journal entries, Firebase Firestore is utilized. This cloud-based database service
allows the app to securely store, retrieve, and update journal entries in real time. It ensures that
user data is kept consistent and accessible, even across multiple devices.
The journaling page also includes AI functionality to analyze the user's journal entry and identify
key emotions they may be experiencing. Based on this analysis, the app's UI is modified to
display the detected emotions to the user. The code streamlines the process by sending the journal
entry to the AI, receiving its analysis, and updating the interface accordingly.
Computer Science & Information Technology (CS & IT) 167
Figure 1. Flowchart of code 1
The geminiFunctionCalling is an asynchronous Flutter function (Future<void>) that processes a
user's journal entry, analyzes its content, and returns a list of emotions and actionable advice.
The function begins by defining a systemPrompt, which instructs the AI to act as a therapist,
focusing on identifying five emotions from the journal entry. Simultaneously, the userPrompt
extracts the user's input from _textEditingController.text and directs the AI to respond with a
comma-separated list of emotions.
To have this interaction, the function initializes a conversation with the AI model using
[Link]. This step establishes a chat session that incorporates a history that is both
relevant and aligned with the function's purpose by combining the user's input and the AI's
guiding instructions.
The history consists of two primary elements:
1. [Link](userPrompt): This incorporates the journal entry provided by the user
through _textEditingController.text. By embedding the user's text directly into the
history, the function enables the AI to process and analyze the specific concerns or
emotions described in the journal entry.
2. [Link]([TextPart(systemPrompt)]): This defines the AI's role and boundaries,
effectively instructing it to act as a therapist. The system prompt specifies the AI's focus
on identifying five emotions from the user's journal entry and provides guidance on the
format of its responses.
Once the history is set up, the user prompt is sent to the AI using [Link]. This ensures
the function's execution remains seamless, avoiding potential issues with accessing incomplete or
delayed responses.
168 Computer Science & Information Technology (CS & IT)
The AI processes the userPrompt within the contextual framework established by the
systemPrompt, enabling it to deliver responses tailored to the journal entry. For example, the AI
might return a concise and structured list of emotions, such as: “anxious, overwhelmed, tired,
frustrated, sad.” This structured process enables the function to achieve its core objective of
identifying emotions in the user's journal entry.
The calendar page serves as a central interface for managing and reviewing journal entries. Users
can view a month’s entries, navigate between months, and click on specific days to add or review
entries. At its core, the component uses DatabaseService to get journal entries from a specific
month and year, ensuring the calendar view is updated in real-time. The component is designed
as a StatefulWidget to handle dynamic data and user interactions, ensuring the UI updates
seamlessly as users navigate through months or fetch new data. Functionally, by integrating
backend data services and AI-powered emotion detection, the calendar page acts as a cohesive
and versatile journaling tool that not only organizes user entries but also fosters introspection,
making it an analytical platform for reflecting on both events and emotions.
Figure 2. Flowchart of code 2
The grid is central to the calendar UI, visually representing the month and offering interactivity
for each day. It dynamically adjusts to varying month lengths and weekday alignments without
additional logic, using a [Link] for flexible layouts across screen sizes. This widget is
surrounded by a Padding widget to reduce visual clutter and provide space, particularly above the
grid. The grid is divided into seven columns, each of which represents a day of the week.
Each day is rendered as an interactive cell, where a tap on a day navigates to a detailed view. This
interaction is facilitated through a GestureDetector that makes each day clickable, enhancing the
functionality of the calendar. Additionally, the design uses the Google Fonts Rubik for modern,
clean typography, ensuring readability and a balanced visual hierarchy. The grid automatically
Computer Science & Information Technology (CS & IT) 169
adapts to different month lengths (28, 29, 30, or 31 days) and accounts for varying weekday starts
using the weekdayOffset, ensuring proper alignment.
The analysis page provides an interactive view of a specific journal entry, featuring two
expandable sections: the entry's content and related advice. Users can expand or collapse these
tiles for a focused or detailed experience. Designed with the Rubik font via Google Fonts, the
page offers a modern, readable style that complements the app's clean aesthetic.
Figure 3. Flowchart of code 3
The _buildExpansionTile method uses the ExpansionTile widget to create collapsible content
sections. Each tile includes parameters like title (displayed text), content (expanded content),
isExpanded (boolean tracking the tile's state), and onExpansionChanged (a callback that updates
the state using setState). The booleans isAdviceExpanded and isJournalEntryExpanded track
whether the "Advice" and "Journal Entry" sections are expanded. In the event that either section
is expanded or collapsed, the state is updated, and the widget is rebuilt based on the new
expansion state. A SizedBox with a height of 20.0 adds spacing between the tiles, while an
Expanded widget ensures that content exceeding the screen size is scrollable via a
SingleChildScrollView. The Column arranges the two expandable tiles vertically, allowing the
user to interact with them to reveal or hide content.
4. EXPERIMENT
4.1. Experiment 1
Journaling is an extremely personal activity, which makes accurate AI analysis the greatest blind
spot in a journaling application. Misinterpreting these sensitive entries could result in
inappropriate or dismissive advice, potentially harming the user. This sensitivity ties into the
app's ethical responsibility, as journaling apps handle private content, and inaccurate advice
170 Computer Science & Information Technology (CS & IT)
based on flawed prompts could raise serious ethical concerns, especially if it influences important
decisions.
To test AI’s emotion detection and advice relevance, the study will compare AI-generated results
with pre-labeled journal entries. The independent variable is the journal entry text, and the
dependent variable is the accuracy of emotion detection and advice relevance. A control group of
pre-labeled journal entries will be used to ensure consistency and provide a benchmark for
comparison.
The AI system will process the same dataset, categorizing emotions and generating advice for
each entry. The performance of the AI will be evaluated afterwards based on advice relevance,
scored on a 1 to 5 scale.
Scores range from 1, indicating irrelevant advice, to 5, indicating highly relevant advice. There
will be a checklist to determine whether the advice directly addresses the emotion, reflects the
entry's details, and is actionable. Each “Yes” contributes to the final relevance score.
The following journal entry is an example of what was created to assess the AI's ability to
accurately detect emotions and generate relevant advice.
1. Boredom
i feel like I’m stuck in a loop. every day is just copy-paste of the last one, and i feel like
an npc. i want something new, something exciting but i don’t even know where to start.
ugh.
Human Advice: Start small by introducing changes to your daily activities, like trying a
new recipe, walking a different path, or picking up a hobby you’ve been curious about. If
you include others in these activities, you might find joy not just in the change but in the
shared experiences.
For the complete list of journal entries used in this study and their scores, please visit the full
dataset at:
[Link]
nXxeyC8bu3gbVGPwY6VuCo/edit?tab=t.0
The AI's average score of 3.975 and median score of 4 suggest its advice is generally relevant,
empathetic, and actionable. However, occasional lower scores highlight areas where the AI
struggles to fully align with the highest quality criteria, pointing to specific opportunities for
improvement. These results underscore the importance of refining emotion recognition models
and tailoring advice.
4.2. Experiment 2
Different cultures have different perspectives on mental health and emotional well-being, and
what is considered helpful or appropriate in one may not be considered the same in another,
therefore the AI must be able to address them all. For instance, East Asian cultures emphasize
collectivism, while the United States values individualism. Additionally, generational differences
influence mental health views, with younger generations more open to emotional expression and
older generations holding more traditional views. The app's advice may be ineffective or
alienating if it fails to analyze these nuances.
Computer Science & Information Technology (CS & IT) 171
To evaluate the app’s handling of cultural and generational differences in mental health, this will
be done through a series of tests focused on cultural sensitivity and generational awareness. The
first test will assess the AI’s ability to provide culturally sensitive advice, using 20 entries from
collectivist and 20 from individualist cultures. The AI’s response will be assessed based on its
ability to recognize cultural context and offer appropriate advice. For example, someone from a
collectivist culture should receive advice focused on family-oriented support. Success will be
measured by calculating the percentage of journal entries correctly attributed to the right culture,
with a target of 90% cultural alignment and 85% satisfaction with tone and relevance.
Next, journal entries will reflect the language styles of different generations, from comparing
older users (50s-60s) with younger users (20s-30s). These entries will express emotional
struggles differently, with older individuals showing reluctance to discuss emotions openly, and
younger individuals being more open to therapy and modern self-care practices. The AI will then
be evaluated for its ability to match advice to each generation, with a target of 80% accuracy in
identifying generational context and 85% satisfaction with the relevance and tone of the response.
In order to determine the success of the project, a percentage of journal entries correctly
attributed to a particular age group will be calculated.
Note that the entries represent generalizations, and may not fully capture individual variations in
language and expression.
Figure 4. Screenshot of experiment 2
The results will reveal how well the AI recognizes cultural and generational nuances.
Misalignment indicates areas requiring further contextual adaptation. This process ensures the
app remains inclusive and effective for a diverse user base.
5. RELATED WORK
The study found that users of Wysa, a chatbot designed for mental health support, develop an
emotional bond with the AI that is comparable to those formed with human therapists[10]. This
bond is key to promoting positive mental health outcomes, including better engagement with
CBT [11].
The AI uses emotionally intelligent conversations to guide users through exercises for managing
symptoms like anxiety and depression, providing emotional support and encouraging continued
interaction. However, one limitation of this solution is its reliance on pre-programmed content
and decision-trees, which can be limiting in more complex or urgent mental health scenarios [11].
172 Computer Science & Information Technology (CS & IT)
Additionally, while Wysa shows promise for addressing the global mental health crisis and the
shortage of therapists, there is still a need for more clinical validation through robust
experimental design. Additionally, the chatbot may not be fully aware of the cultural and
linguistic differences between different populations, which may limit its effectiveness.
By incorporating more personalized AI responses, integrating real-time data from validated
mental health assessments, this study’s AI can improve. Moreover, tackling sociocultural factors
such as linguistic nuances could further enhance the AI’s ability to provide effective support
across diverse user groups.
AI-powered conversational agents utilize NLP to provide mental health support through human-
like interactions. These systems interpret user inputs, analyze sentiment, and offer personalized
coping strategies, including mindfulness and cognitive behavioral therapy exercises.
Conversational agents are effective in addressing mild to moderate mental health issues by
providing immediate, stigma-free, and accessible support. However, limitations include biases in
training data, lack of cultural sensitivity, and an inability to manage crisis effectively. Despite
these challenges, they represent a valuable tool for expanding access to mental health resources.
Future advancements must prioritize personalization and inclusivity to enhance their impact [12].
By identifying patterns in language, sentiment, and tone from textual data, NLP techniques assess
mental health. Detecting early signs of mental health conditions, providing insight into emotional
states, and supporting large-scale mental health studies are all possible with these systems. While
NLP demonstrates moderate to high accuracy, it is limited by data biases, lack of contextual
understanding, and challenges in interpreting complex emotions. Additionally, static models may
not adapt to evolving user language. Addressing these limitations requires dynamic, user-focused
approaches that integrate cultural and personal contexts for more nuanced and actionable mental
health support [13].
6. CONCLUSIONS
The AI-driven journaling app has limitations in understanding complex emotions, adapting to
diverse cultural and generational contexts, and sustaining user engagement. Addressing these
requires refining NLP models with diverse datasets, integrating region-specific linguistic models,
and adding gamified features. With more time, efforts would focus on creating a dynamic
framework incorporating user feedback, expanding datasets, and iterative testing to improve
accuracy, inclusivity, and engagement. Despite these challenges, the app holds potential to
revolutionize mental health support by enhancing personalization, cultural sensitivity, and
accessibility, empowering users to effectively manage their mental well-being.
REFERENCES
[1] Hidaka, Brandon H. "Depression as a disease of modernity: explanations for increasing prevalence."
Journal of affective disorders 140.3 (2012): 205-214.
[2] Robinson, Robert G., and Gianfranco Spalletta. "Poststroke depression: a review." The Canadian
Journal of Psychiatry 55.6 (2010): 341-349.
[3] Ettman, Catherine K., et al. "Prevalence of depression symptoms in US adults before and during the
COVID- 19 pandemic." JAMA network open 3.9 (2020): e2019686-e2019686.
[4] McLaughlin, Katie A. "The public health impact of major depression: a call for interdisciplinary
prevention efforts." Prevention Science 12 (2011): 361-371.
[5] Boerema, A. M., Kleiboer, A., Beekman, A. T., van Zoonen, K., Dijkshoorn, H., & Cuijpers, P.
(2016). Determinants of help-seeking behavior in depression: a cross-sectional study. BMC
psychiatry, 16, 78. [Link]
Computer Science & Information Technology (CS & IT) 173
[6] Becker-Weidman, E. G., Jacobs, R. H., Reinecke, M. A., Silva, S. G., & March, J. S. (2010). Social
problem-solving among adolescents treated for depression. Behaviour research and therapy, 48(1),
11–18. [Link]
[7] Hwang, B., Choi, H., Kim, S., Kim, S., Ko, H., & Kim, J. (2018). Facilitating student learning with
critical reflective journaling in psychiatric mental health nursing clinical education: A qualitative
study. Nurse education today, 69, 159–164. [Link]
[8] Horsfall, J., Cleary, M., & Hunt, G. E. (2010). Stigma in mental health: clients and professionals.
Issues in mental health nursing, 31(7), 450–455. [Link]
[9] Srivastava, R., & Srivastava, R. (2019). Impact of cultural identity on mental health in post-
secondary students. International Journal of Mental Health and Addiction, 17(2), 520–530.
[Link]
[10] Beatty, C., Malik, T., Meheli, S., & Sinha, C. (2022). Evaluating the Therapeutic Alliance With a
Free-Text CBT Conversational Agent (Wysa): A Mixed-Methods Study. Frontiers in digital health,
4, 847991. [Link]
[11] Chew, A. M. K., Ong, R., Lei, H. H., Rajendram, M., K V, G., Verma, S. K., Fung, D. S. S., Leong,
J. J., & Gunasekeran, D. V. (2020). Digital Health Solutions for Mental Health Disorders During
COVID-19. Frontiers in psychiatry, 11, 582007. [Link]
[12] Guntuku, S. C., Ramsay, J. R., Merchant, R. M., & Ungar, L. H. (2019). Language of ADHD in
Adults on Social Media. Journal of attention disorders, 23(12), 1475–1485.
[Link]
[13] DeYoung, J., Jain, S., Rajani, N. F., Lehman, E., Xiong, C., Socher, R., & Wallace, B. C. (2020,
April 24). ERASER: A Benchmark to Evaluate Rationalized NLP Models. [Link].
[Link]
[14] Nepal, S., Pillai, A., Campbell, W., Massachi, T., Heinz, M. V., Kunwar, A., Choi, E. S., Xu, O.,
Kuc, J., Huckins, J., Holden, J., Preum, S. M., Depp, C., Jacobson, N., Czerwinski, M., Granholm,
E., & Campbell, A. T. (2024). MindScape Study: Integrating LLM and Behavioral Sensing for
Personalized AI-Driven Journaling Experiences. [Link]. [Link]
[15] Kesavan, R., Gay, D., Thevessen, D., Shah, J., & Mohan, C. (2023). Firestore: The NoSQL
Serverless Database for the Application Developer. 2023 IEEE 39th International Conference on
Data Engineering (ICDE). [Link]
© 2024 By AIRCC Publishing Corporation. This article is published under the Creative Commons
Attribution (CC BY) license.
