User Profiling Based Proactive Interaction Manager

This interactive dashboard synthesizes the findings from the research paper by Hoashalarajh Rajendran et al. (University of Moratuwa). The study explores how service robots can transition from reacting to explicit commands to proactively offering services (specifically, reading suggestions) by analyzing user context, posture, and emotion.

The Core Problem

Moving Beyond "Service on Demand"

Traditional robots wait for users to ask for help. This research implements a system where the robot observes the user (e.g., checking if they are relaxing or working), reads their emotional state, considers the environment (time, weather), and proactively initiates interaction to suggest a book. This approach aims to reduce user disturbance and enhance human-robot bonding.

👁

Perceive

Vision & Voice inputs to gather context.

Process

Decision Trees & Online Learning adaptation.

🗣

Proact

Initiate service at the optimal moment.

🎯

Key Achievements

  • High Perception Accuracy

    Emotion and Posture models achieved ~90% accuracy.

  • Proven Adaptability

    System successfully learned and shifted reading preferences over a 4-day trial.

  • Statistical Significance

    User satisfaction rejected the null hypothesis (p < 0.05), proving proactive superiority.

System Architecture

This section illustrates the data pipeline of the robot. The architecture is divided into Perception (gathering data), Cognition (making decisions), and Action. Explore the interactive blocks below to understand how raw sensor data is transformed into a proactive service.

1. Perception Module

📷

Vision Inputs

• MTCNN (Face ID)

• VGG19 EmotionNet

• StateNet (Posture)

🎙

Voice Inputs

• Speech-to-Text

• Sentiment Analyzer

• User Feedback

🌆

Context Data

• Weather API

• Time of Day

• Google Calendar

2. Cognition Module

Proactive Interaction Manager

State Module: Work vs Relax

Fusion Module: Vision + Voice

Decision Tree + Preference Learning

3. Action Module

📖

Service Execution

Robot initiates conversation via text-to-speech, offering a highly tailored book genre recommendation based on the cognitive analysis.

👆 Interactive Diagram

Click on any module above to view detailed information about its role in the proactive interaction pipeline.

The Cognitive Engine

How does the robot actually decide what to suggest? This section unpacks the core intelligence of the system. It utilizes a Decision Tree for baseline suggestions and an Online Learning Algorithm to continuously adapt to the user's changing preferences based on their feedback.

1 Decision Tree Attributes

The initial genre suggestion is derived by traversing a decision tree built on five key contextual variables. Click to highlight attributes.

👥

Demographic

Gender

Male / Female
🙂

Psychological

User Emotion

Happy, Sad, Angry...

Temporal

Time of Day

Morning, Evening...
🌤

Environmental

Weather Condition

Hot, Normal, Cold
📅

Scheduling

Remaining Time

Low, Mod., High

2 Preference Adaptation

The Online Learning Weight Updating Formula ensures the robot learns from its mistakes. If a user rejects a "Fiction" book, the system dynamically lowers the weight of Fiction for that specific user profile in real-time.

// Algorithm 3: Weight Updating
def update_preference(user, genre, feedback):
alpha = 0.5 // Learning rate
current_weight = get_weight(user, genre)

if feedback == 'accept':
new_weight = current_weight + (alpha * 1)
elif feedback == 'reject':
new_weight = current_weight + (alpha * -1)

save_weight(user, genre, new_weight)
Result: Continuous Personalization

Experimental Validation

The system was evaluated in two main phases. First, ensuring the perceptual models could accurately identify human states. Second, measuring how well the proactive manager adapted to user preferences over a 4-day trial period compared to a baseline.

Perception Accuracy

Performance of individual AI models

📈

StateNet (Posture detection) achieved the highest accuracy, crucial for distinguishing between "Working" and "Relaxing" states to prevent unwanted interruptions.

System Adaptability

Mean user satisfaction scores over 4 days

📈

The upward trend demonstrates the effectiveness of the Online Learning Algorithm. By Day 4, the robot was highly accurate in predicting the desired book genre.

🏆

Hypothesis Testing Results

p = 4.28e-11

A one-sample t-test was conducted against a baseline satisfactory score of 3.0. The extremely low p-value (p < 0.05) indicates that user satisfaction with the proactive system was statistically significant and vastly superior to baseline expectations. Users felt the robot understood their needs without being prompted.