As I wrap up this chapter of my academic journey (PhD is now complete! 🥳), I wanted to take a moment to reflect on the research progress I’ve made over the past year, and jot down thoughts for where I might grow next. Writing this out helps me stay grounded, spot new offshoots, and remember how far things have come — especially as I start mentoring students and shaping future projects.
🎶 Dissertation: Action Rules + Music Therapy
This year, I completed and successfully defended my dissertation, Recommendation System for Typhlo Music Therapy in Rehabilitation Work with Visually Impaired Children. I also published two journal papers with my advisor, Dr. Ras, and a third paper—co-authored with Dr. Ras and the collaborating music therapist, Dr. Cylulko. The project focuses on supporting therapists who work with visually impaired children by generating interpretable action rules derived from audio features and session data.
It was my first time working with data provided by a domain expert—a practicing music therapist—and it sparked a genuine interest in interdisciplinary research. I really enjoyed exploring a field so different from my own, and this project deepened my appreciation for music therapy and the potential of AI to support human-centered, therapeutic work. I’d love to continue learning more about this space.
I also spent time extracting music features using tools like librosa, experimenting with different feature types (chroma, tempo, spectral contrast, etc.), and generating action rules from RSES classification rules (specifically using the LEM2 method).
Next Steps & Ideas:
- Expand feature extraction and visualization, especially around emotional or tonal aspects of the music.
- Build interactive dashboards or simple recommendation prototypes to share insights with therapists.
- Experiment with exhaustive search methods for RSES rule generation and compare resulting action rules.
- Develop the Action Schema method for generating more flexible and generalizable action rules.
- Extend action rule generation to optimize for multiple decision attributes at once, rather than treating them independently—better reflecting the multifaceted nature of therapeutic progress.
- Explore graph-based representations of relationships between children, sessions, and music.
- Begin building the actual music recommender system, incorporating insights from action rules to suggest music that aligns with therapeutic goals.
Related papers:
- https://doi.org/10.3390/app14031270
- https://doi.org/10.1007/s10844-025-00961-5
- https://www.mdpi.com/2078-2489/16/8/666
💬 Student Reflection + Support Recommender System
Another major project this year was designing and running an A/B study on our Student Challenge-Solution Recommender System. For this project, I collaborated with Dr. Wiktor and Dr. Dorodchi. The original version (SCS), which surfaces solutions written by past students facing similar challenges, was published in a pilot paper I worked on in 2023. This year, we developed an enhanced version (LLM-SCS) that incorporates GPT-4 to generate more personalized and supportive suggestions. This version was recently submitted to SIGCSE for review.
I enjoyed running a multi-class deployment study across several semesters, and what I found most meaningful was the system’s potential to support students’ sense of belonging. One of the goals behind this project was to show students that they’re not alone in their challenges—and that with the right support, they can get through. The system not only gives students a space to reflect and receive advice, but also offers instructors valuable insights into the types of challenges their students are facing.
I first discovered the value of self-reflection during my undergraduate years, particularly through my experiences in the honors program. It’s something I’ve carried with me ever since—shaping how I approach learning, research, teaching, and even moments like this, as I sit and write this reflection. Taking the time to pause, look back, and look forward has been one of the most grounding and meaningful habits I’ve developed.
Next Steps & Ideas:
- Build a guided reflection system. While current LMS tools like Canvas surveys offer a starting point, I’ve long been interested in developing a more dynamic, agent-based system that can help guide students through deeper, more meaningful self-reflection. Ideally, this system would also summarize and surface key themes to instructors in a digestible format—something especially helpful when working with 100+ student reflections.
- Continue the deployment study. One immediate goal is to scale up the original study with more participants to strengthen the findings and improve generalizability. I’d also like to improve the system’s automation pipeline—right now, spreadsheet generation and batch emailing works, but it could definitely be smoother.
- Collect instructor and external feedback. Adding a layer of instructor insights or feedback from trusted reviewers would give the system more context and further refine the quality of student-facing suggestions.
- Build a centralized database. A longer-term technical goal is to move away from scattered spreadsheets and instead create a structured database for the challenge-solution pairs and user interactions, making it easier to manage, analyze, and expand the system over time.
🎮 Gamified Learning Pathways: A Skill Tree Approach
This idea is still in the very early stages—just a vague concept for now—but I’ve been thinking a lot about how to make course design feel more like a video game skill tree. The goal would be to let students choose their own learning pathways while still meeting key course objectives. I feel like this model could work especially well for an online asynchronous course, but I’d need to brainstorm further to see how it might be adapted to a synchronous setting. I’d love to explore ways to pilot it in my Data Mining course down the line, particularly as a way to increase engagement and give students more ownership over their learning journey.
Next Steps:
- Break course content into modular “skills” or branches.
- Prototype visual or interactive learning maps.
- Collaborate with students on the design process to make it playful and pedagogically sound.
Growing Into Mentorship
One of the most exciting parts of this coming year is that I’ll be mentoring undergraduate students in research. I’m especially interested in inviting students to contribute to:
- Data visualization and feature analysis for the music therapy project, exploring how audio features relate to therapy outcomes.
- Peer feedback categorization, guided reflection design, and email interface improvements as part of the student challenge-solution recommender system.
- Gamified course design experiments, such as prototyping a skill tree model to support flexible and engaging learning pathways.
I’m excited to support students as they bring their own ideas, skills, and creativity into these evolving projects. As I step into this new mentoring role, I’m learning how to balance guidance and autonomy, structure projects thoughtfully, and match tasks to each student’s interests and strengths. I have a feeling these experiences will shape how I approach both research and teaching in meaningful ways moving forward.
Looking Ahead
As I transition into life after the PhD, I’m excited to keep building on this work—with more reflection, more collaboration, and more student-driven design. Whether it’s personalizing therapeutic insights, empowering students with AI, or turning a course into an interactive map, I want to keep making tools and systems that are helpful, thoughtful, and deeply human-centered.
Aileen Benedict 