PUBLISHED RESEARCH · PRODUCT DESIGN · HUMAN-CENTERED AIMusic Recommendation using Emotion Recognition
I designed and evaluated an emotion-aware music experience that used facial emotion recognition to adapt recommendations to the user’s current state.
PUBLICATION
IEEE MysuruCon
YEAR
2022
CONTRIBUTION
Research → UX → Prototype → Publish
LINK
AT A GLANCEPROBLEM
During COVID-19, people were emotionally drained while digital devices remained constant companions.
DESIGN QUESTION
How can an interface respond to emotional context in a way that feels supportive rather than intrusive?
MY CONTRIBUTION
Research, playlist curation, interaction design, implementation, documentation, evaluation, and publication.
MY PROCESSAcross six stages, I moved from research and curation into interaction design, implementation, evaluation, and publication.
01
RESEARCH
Understand the problem, symptoms, and research fundamentals.
02
CURATE
Translate music-therapy research into emotionally aligned playlists.
03
DESIGN
Create a simple, accessible interaction from recognition to recommendation.
04
IMPLEMENT
Connect the recognition model to the playlist experience.
05
DOCUMENT
Capture method, decisions, model comparison, and limitations.
06
PUBLISH
Prepare the work for peer review and publication with IEEE.
HOW THE SYSTEM WORKED01
CAPTURE
INPUT DESIGN · USER AGENCY
02
RECOGNIZE
Translate feedback into a usable stateThe CNN classifies the facial expression; the interface turns that output into a clear interaction signal.
DATA ANALYSIS · USER FEEDBACK
03
CURATE
Turn research into iterative prototypes
Music-therapy research informed how each emotional state mapped to playlists designed to support, not simply mirror, the user’s mood
RESEARCH SYNTHESIS · PERSONALIZATION
04
RECOMMEND
Make adaptation understandable and actionablePresent the designs with minimal friction while preserving emotional context and user control.
ADAPTIVE INTERACTION · USER CONTROL
SYSTEM VALIDATIONModel performance mattered because every design recommendation depended on the quality of the emotional signal.
89%
Accuracy
SVM 86% · +3 percentage points
79%
Specificity
SVM 73% · +6 percentage points
79%
Sensitivity
SVM 92% · +4 percentage points
OUTCOME + DESIGN VALUETurning emotion recognition into a clear, usable interaction.
I structured the experience around three UX priorities: clear system feedback, low-friction task flow, and an accessible transition from recognition to recommendation.
system feedback → interaction flow → accessibility → information clarity → design judgment
UX VALUE
The design challenge wasn’t detecting an emotion.
It was deciding how that prediction should appear in the experience:
what the user needs to understand, what happens next, and how to keep the interaction simple.
SCOPEThis was a first step toward more emotionally aware experiences. The study was limited to a core set of emotions and basic playlist delivery, but the interaction model could extend to smart devices, richer emotional models, or real-time adaptive experiences.
PEER-REVIEWED PUBLICATION
Published through IEEE MysuruCon
Conference: 16–17 October 2022 · IEEE Xplore: 13 December 2022