Time and place:
24.11.2020 г. (Thursday), 18:15

Meeting ID meet.google.com/fia-zdtx-kiu; Phone Numbers(‪BG‬) ‪+359 2 907 4849‬ PIN: ‪452 034 234#‬

Zlatin Stanimirov

Is it possible to automate the tracking of public opinion on current events ?

Modern natural language processing solutions can give high accuracy, but require significant resources for training and implementing inside software. Linear models give smaller accuracy than models with deep learning, but they need significantly less resources to train and use.

The seminar will showcase a web application which uses both linear and deep models using real data from Twitter.

There will be a comparison between the resources, performance and accuracy of different models.

Lecture topics:

  • Data preprocessing
  • Linear models and representations – Logistic Regression, Support Vector Machine, Bag of Words, TfIdf, Doc2Vec
  • Deep learning – BERT via ktrain and HuggingFace
  • Using the official Twitter API to fetch real data.
  • Using the trained models inside web servers
  • Visualising the results of the classification

This lecture takes place within the Machine Learning seminar of IAPS.

Share This