top of page
Machine Learning for Adaptive System Design - AI-nonymous
This case study details the design and implementation of 'AI-nonymous,' a mobile anonymous chat application leveraging Machine Learning to match users based on personality traits, aiming to explore if algorithm-supported matches enhance user experience compared to random pairing.
Background
The dynamics of social interaction in online environments are driven by the innate human tendency to seek connection and understanding. This project addresses the fundamental need for meaningful social interaction by exploring the use of anonymous chat applications. By leveraging Machine Learning to match users based on personality traits in 'AI-nonymous', the aim is to enhance user experience and facilitate quicker, more comfortable connections with similar individuals.
Project Details
Individual Contribution
-
Timeline: November 2022 - February 2023
-
Team size: 4
-
Location: Eindhoven, The Netherlands
-
Data Preparation
-
Machine Learning Modeling
-
Information Architecture
Design Methodologies
Technologies Used
-
Online Survey
-
Unsupervised Learning (Clustering)
-
ML-enabled Recommendations for Similar Personality
-
Miro
-
Python & JupyterLab
-
Figma
-
MS Office
Design Concept
The design concept involves utilizing Machine Learning to enhance user experience in the 'AI-nonymous' application. Users are recommended others to chat with based on their profile, employing AI clustering to group similar individuals according to personality traits and views. Conversation starters are provided based on selected topics, with further customization through questions tailored to the user's perspectives. Feedback from user experiences is then integrated back into the Machine Learning algorithm for continuous improvement of recommendations.
Machine Learning Modeling
The project utilizes the Young People Survey dataset, redesigned using factor analysis to identify the Big Five personality traits. Additional data for these variables is collected through an online survey. The machine learning model, implemented using Scikit-learn in Python, employs unsupervised K-means clustering to group similar profiles based on personality traits. Recommendations are made based on interests and preferences from the survey data, with the clustering model periodically retrained to accommodate new inputs.
Information Architecture & System Design
Conceptualizing the Adaptive System
The user flow begins with initial Likert-scale questions upon first login, followed by suggested chat topics tailored to the user's personality profile, thus minimizing setup time. Chat partner selection is based on an adaptive system, prioritizing users estimated to be better conversation partners, with statements derived from their app usage to initiate conversation. Once a chat partner is selected, users rate their agreement with the partner's statement to further build their personality profile. The chat interface allows conversations to go off-topic, with options for automatic chat destruction and user feedback to enhance the chatting experience indirectly through data analysis.
Limitations & Future Work
Prototype designed by Mats Erdkamp
The study's reliance on unsupervised learning necessitates further data gathering to assess the accuracy of the proposed algorithm in real conversational contexts. Additionally, the use of the 'Big Five' personality traits for matching may not fully capture user preferences, warranting exploration of alternative options and continual algorithm adaptation over time. Considering potential user preferences for conversing with dissimilar individuals and accommodating for changing personality traits are important aspects for future iterations. Further enhancements may include implementing feedback mechanisms to optimize the recommendation system and allowing users to introduce new interests and hobbies beyond recommended ones.
More Projects
bottom of page