Cognitive computing approaches
for human activity recognition
from tweets
A case study of Twitter marketing campaign
Dr. Jari Jussila & Prashanth Madhala
Applied Data Analytics Workflow
#moodmetricstressinhallintakeino
https://azure.microsoft.com/en-us/services/cognitive-services/computer-vision/#analyze
Caption: A woman sitting in a room
Translation of tweet
https://translate.google.com/
Sentiment Analysis
See e.g. Thelwall (2017)
Activities and Emotions in 2D Valence
& Arousal Space
Example of False Positive Anger
”Play Paranoid”
Discussion and conclusions
• No libraries or API’s were found that support activity recognition
adequately
• Combining activity recognition and sentiment analysis provides
deeper understanding of customer behavior – which is helpful for
marketing
• Algorithms provide means for strong (small data) and weak
personalization (big data) of stress management service
• Current cognitive computing API’s support a limited number of
languages - and more importantly are only commodity to users that
have programming skills

Cognitive computing approaches for human activity recognition