Pervasive Chatbots: Investigating
Chatbot Interventions for Multi-
Device Applications
The 32nd ACM Conference on User Modeling, Adaptation and Personalization
Cagliari, Sardinia, Italy
Mayowa Olapade, PhD Candidate University of Tartu
4TH JULY 2024
Tarlan Hasanli, Abdul-Rasheed Ottun, Adeyinka Akintola
Mohan Liyanage and Huber Flores
Background
• Increased use of IoT and smart devices within digital landscape
➢ Device(s) in-built infrastructures making connectivity easier to use
2
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
‘In the digital
world of today,
one can say
almost nothing is
local’ – John W.
Henry
[Source] https://blockchain.infovys.com/blockchain-consensus-algorithm-focus-fundamentals/ [Source] https://en.wikipedia.org/wiki/BitTorrent [Source] https://blog.google/products/android/io22-multideviceworld/
Bitcoin
Cross-device ads
BitTorrent
3
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
Importance
• Enhanced functionality in opportunistic environments
➢ More efficient devices (QoS)
➢ Improved user experience (QoE)
[Source] https://www.freepik.com/premium-photo/young-business-people-
using-their-smartphones-workshop_16751373.htm
[Source] https://www.westend61.de/en/photo/MEUF00791/group-of-
friends-using-smartphones-at-bus-stop-in-the-city
Meetings Bus stations
Farms
[Source] https://maker.pro/blog/various-sensors-and-
their-applications-for-smart-farming-and-robotics
Limitation
• Heavily relies on human expertise or know-how for example battery saving task in mobile phones
Do people know about these methods?
Micro-data centers
Sensing scope
File storage
Gauging perception before intervention
4
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots“32nd ACM UMAP 2024,
Cagliari, Italy
• Distributed and collaborative environments
o Two-week online survey
o 56 participants
o 4 different applications
o Likert-scale style
o 29 questions
- High knowledge gap
Top suggestions
1. Close all background apps
2. Activate battery saving mode
3. Computational offloading
The Idea: Pervasive Chatbots
• The rise of AI chatbot for everyday interventions. E.g. LLMs (text based)
➢ Context-aware responses based on suggestions/recommendations.
➢ Chat-induced guidance between multi-devices with the option of
automating the suggestion provided.
5
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
User input Analyze requests and identify suggestions
from recommendations
Reply user with
suggestions and
option to trigger
automation within
chat
Regular
chatbot
Pervasive
chatbot
Reply user with
suggestions
Chatbot flow
6
Contributions
• Chatbot solution: We designed and developed a proof-of-
concept pervasive chatbot solution that fosters the easy
interconnection of multiple devices for end users.
• Insights: By utilizing the pervasive chatbot, real time technical
assistance can be triggered for time constraint collaborative
task like battery saving in mobile devices.
• Evaluation: Shows pervasive chatbots can aid user
performance when compared to reference chatbots when
performing collaborative tasks.
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
[Source] https://tenor.com/en-GB/view/sleep-kaboo-gif-
8665269083093919363
7
Chatbot design and development - Implementation
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots“32nd ACM UMAP 2024,
Cagliari, Italy
• Java 17
• Android SDK 33
• Gradle 8.1.2
• Apache OpenNLP 2.3.0
Custom-built to save battery power for energy-constraint
device.
Welcome to PervasiveBot. I’m here
to help you learn how to save
energy and preserve battery life.
Just ask me questions like “How to
save energy?” and I’ll provide you
with useful steps and advice. Feel
free to get started
Top recommendations/suggestions
1. Close all background apps
2. Activate battery saving mode
3. Computational offloading
8
Pervasive vs. Reference mode interface
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
The experiment: Qualitative evaluation of pervasive chatbot
9
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
o Participants: 24 between
subject design
o Apparatus: Huawei android
device
o Design concept: Priming
battery life with a
background service running
continuously
o Metrics: Comparison
performance, user
awareness level and user
satisfaction score(SUS and
TLX)
o Question: Likert scale (1-5)
• User study Energy saving task
3 mins
10
Results – Comparison performance
• Pervasive chatbot vs Reference chatbot
• Significant 57% improvement over reference chatbot.
• 3 participants failed to finish using reference chatbot.
In terms of efficiency, pervasive chatbots performs better
with time constraint tasks.
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
Results – User awareness level
11
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
• Pervasive chatbot vs Reference chatbot
• Participants were aware of a new method to save battery life.
• Accuracy of new method learned improved with reference chatbot.
In terms of user awareness level, reference chatbots raises awareness
level better.
Results - User satisfaction scores
12
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
• Pervasive chatbot vs Reference chatbot
• Usability SUS score 92% vs 83%, both showing high satisfaction score.
• Perceived workload (mental demand, effort and frustration
level) PC - 45.2% RC - 79.8%.
In terms of user satisfaction scores, pervasive chatbots is more satisfying
to use.
13
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
Lesson learned
• Multi-device applications are easy to construct with the aid of pervasive chatbot
• Privacy, ethical and security risks related concerns when establishing multi-device connections
• The need for lightweight chatbots in constrained devices
• Chatbot level of automation can be vital for edge intelligence
14
Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024,
Cagliari, Italy
• Pervasive chatbot is an innovative method that fosters easy collaborative tasks by interconnecting
multiple devices.
• It significantly performs better than regular chatbots when interconnecting devices by reducing time
effort needed.
• It shows there is a significant trade-off between automated function and knowledge gained during use.
• Lastly, we contributed to a prototype as open source available in GitHub (https://github.com/mobile-
cloud-computing/Pervasive-Chatbot)
Summary & conclusion
Questions?
Thank you! (Do not hesitate to contact us)
Grazie!
14
Adeyinka Akintola
(akintola.adeyinka@ut.ee)
Mayowa Olapade(olapade@ut.ee)
Tarlan Hasanli
(tarlan.hasanli@ut.ee)
Huber Flores
(huber.flores@ut.ee)
Abdul-Rasheed Ottun
(rasheed.ottun@ut.ee)
Mohan Liyanage
(mohan.liyanage@ut.ee)

Pervasive AI-chatbots for multi-device and decentralized applications

  • 1.
    Pervasive Chatbots: Investigating ChatbotInterventions for Multi- Device Applications The 32nd ACM Conference on User Modeling, Adaptation and Personalization Cagliari, Sardinia, Italy Mayowa Olapade, PhD Candidate University of Tartu 4TH JULY 2024 Tarlan Hasanli, Abdul-Rasheed Ottun, Adeyinka Akintola Mohan Liyanage and Huber Flores
  • 2.
    Background • Increased useof IoT and smart devices within digital landscape ➢ Device(s) in-built infrastructures making connectivity easier to use 2 Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy ‘In the digital world of today, one can say almost nothing is local’ – John W. Henry [Source] https://blockchain.infovys.com/blockchain-consensus-algorithm-focus-fundamentals/ [Source] https://en.wikipedia.org/wiki/BitTorrent [Source] https://blog.google/products/android/io22-multideviceworld/ Bitcoin Cross-device ads BitTorrent
  • 3.
    3 Mayowa Olapade (Olapade@ut.ee),“Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy Importance • Enhanced functionality in opportunistic environments ➢ More efficient devices (QoS) ➢ Improved user experience (QoE) [Source] https://www.freepik.com/premium-photo/young-business-people- using-their-smartphones-workshop_16751373.htm [Source] https://www.westend61.de/en/photo/MEUF00791/group-of- friends-using-smartphones-at-bus-stop-in-the-city Meetings Bus stations Farms [Source] https://maker.pro/blog/various-sensors-and- their-applications-for-smart-farming-and-robotics Limitation • Heavily relies on human expertise or know-how for example battery saving task in mobile phones Do people know about these methods? Micro-data centers Sensing scope File storage
  • 4.
    Gauging perception beforeintervention 4 Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots“32nd ACM UMAP 2024, Cagliari, Italy • Distributed and collaborative environments o Two-week online survey o 56 participants o 4 different applications o Likert-scale style o 29 questions - High knowledge gap Top suggestions 1. Close all background apps 2. Activate battery saving mode 3. Computational offloading
  • 5.
    The Idea: PervasiveChatbots • The rise of AI chatbot for everyday interventions. E.g. LLMs (text based) ➢ Context-aware responses based on suggestions/recommendations. ➢ Chat-induced guidance between multi-devices with the option of automating the suggestion provided. 5 Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy User input Analyze requests and identify suggestions from recommendations Reply user with suggestions and option to trigger automation within chat Regular chatbot Pervasive chatbot Reply user with suggestions Chatbot flow
  • 6.
    6 Contributions • Chatbot solution:We designed and developed a proof-of- concept pervasive chatbot solution that fosters the easy interconnection of multiple devices for end users. • Insights: By utilizing the pervasive chatbot, real time technical assistance can be triggered for time constraint collaborative task like battery saving in mobile devices. • Evaluation: Shows pervasive chatbots can aid user performance when compared to reference chatbots when performing collaborative tasks. Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy [Source] https://tenor.com/en-GB/view/sleep-kaboo-gif- 8665269083093919363
  • 7.
    7 Chatbot design anddevelopment - Implementation Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots“32nd ACM UMAP 2024, Cagliari, Italy • Java 17 • Android SDK 33 • Gradle 8.1.2 • Apache OpenNLP 2.3.0 Custom-built to save battery power for energy-constraint device. Welcome to PervasiveBot. I’m here to help you learn how to save energy and preserve battery life. Just ask me questions like “How to save energy?” and I’ll provide you with useful steps and advice. Feel free to get started Top recommendations/suggestions 1. Close all background apps 2. Activate battery saving mode 3. Computational offloading
  • 8.
    8 Pervasive vs. Referencemode interface Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy
  • 9.
    The experiment: Qualitativeevaluation of pervasive chatbot 9 Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy o Participants: 24 between subject design o Apparatus: Huawei android device o Design concept: Priming battery life with a background service running continuously o Metrics: Comparison performance, user awareness level and user satisfaction score(SUS and TLX) o Question: Likert scale (1-5) • User study Energy saving task 3 mins
  • 10.
    10 Results – Comparisonperformance • Pervasive chatbot vs Reference chatbot • Significant 57% improvement over reference chatbot. • 3 participants failed to finish using reference chatbot. In terms of efficiency, pervasive chatbots performs better with time constraint tasks. Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy
  • 11.
    Results – Userawareness level 11 Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy • Pervasive chatbot vs Reference chatbot • Participants were aware of a new method to save battery life. • Accuracy of new method learned improved with reference chatbot. In terms of user awareness level, reference chatbots raises awareness level better.
  • 12.
    Results - Usersatisfaction scores 12 Mayowa Olapade (Olapade@ut.ee), “Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy • Pervasive chatbot vs Reference chatbot • Usability SUS score 92% vs 83%, both showing high satisfaction score. • Perceived workload (mental demand, effort and frustration level) PC - 45.2% RC - 79.8%. In terms of user satisfaction scores, pervasive chatbots is more satisfying to use.
  • 13.
    13 Mayowa Olapade (Olapade@ut.ee),“Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy Lesson learned • Multi-device applications are easy to construct with the aid of pervasive chatbot • Privacy, ethical and security risks related concerns when establishing multi-device connections • The need for lightweight chatbots in constrained devices • Chatbot level of automation can be vital for edge intelligence
  • 14.
    14 Mayowa Olapade (Olapade@ut.ee),“Pervasive Chatbots" 32nd ACM UMAP 2024, Cagliari, Italy • Pervasive chatbot is an innovative method that fosters easy collaborative tasks by interconnecting multiple devices. • It significantly performs better than regular chatbots when interconnecting devices by reducing time effort needed. • It shows there is a significant trade-off between automated function and knowledge gained during use. • Lastly, we contributed to a prototype as open source available in GitHub (https://github.com/mobile- cloud-computing/Pervasive-Chatbot) Summary & conclusion
  • 15.
    Questions? Thank you! (Donot hesitate to contact us) Grazie! 14 Adeyinka Akintola (akintola.adeyinka@ut.ee) Mayowa Olapade(olapade@ut.ee) Tarlan Hasanli (tarlan.hasanli@ut.ee) Huber Flores (huber.flores@ut.ee) Abdul-Rasheed Ottun (rasheed.ottun@ut.ee) Mohan Liyanage (mohan.liyanage@ut.ee)