ihack

Supervised model for tagging twitter users according to their view on XXXX, based on tweets analysis

in short -

24 h project (2 men involved)
Used Twitter API for crawling anti-XXX and pro-XXX users based on publication of hostile or positive hashtags (Python).
resulting 1500 user_data of each containing maximum - last 300 tweets

Engineered features according to -

• lists of expert's adviced word for different conflict position.
• timezone utc offset
• sentiment analysis (using VADER model) of "super tweets"
  resulting in 89 dimension feature vec

Training

A decision tree (100 tree 32 depth) and logistic-regression models for binary classification of pro or anti-XXX opinion using Azure ML

Test precision, recall and accuracy results:

issues -

• non balanced data set (80% - 20%)
• no time for model improvments (CrossVal etc.)
• model was not exported to web app
• low recall
• key autho problems in tweeter API
• learn how to export the model