Skip to content

Continuous Conditional text generation of clickbait newspaper articles. Using fine-tuned Pegasus text summarizer, with Fudge and PPLM conditional text generation models.

License

Notifications You must be signed in to change notification settings

dsvilarkovic/clickbaitonator

Repository files navigation

clickbaitonator

NLP & LSS 2022 project aimed to create conditional clickbait titles based on the article content.
Check the Huggingface Spaces app at : https://huggingface.co/spaces/Dusan/clickbaitonator

Files used for running models

ClickbaitClassifier.ipynb - notebook for training clickbait classifier.
Fudge_running.ipynb - Fudge based implementation for generating controllable clickbait titles. It contains setup for generating titles, rating them according to our clickbait classifier and saving in respective dataframes. Our Huggingface Spaces implementation uses this model due to realtime response. At the end of this notebook there is also a Gradio section where you can launch in-notebook Gradio application.
PPLM_running.ipynb - PPLM based implementation for generating controllable clickbait titles. It contains setup for generating titles, rating them according to our clickbait classifier and saving in respective dataframes.
DataSet_sentence_collector.ipynb - sentence collector on Fudge and PPLM dataframes generated by Fudge_running.ipynb and PPLM_running.ipynb. It was used for sampling best titles that were generated.
DatasetAnalysis.ipynb - notebook for running wordcloud and LDA topic modelling analysis on ground truth titles.

Folders used for running analysis and models

DataProcessing - used for generating all relevant tables in a report and presentation. Contains fudge_csv, PPLM_csv and pegasus_clickbait_summarizer_csv folders for .csv files that were used for the report and visualizations of respective models.
fudge - Fudge adapted to our clickbait summarizer.
PPLM - PPLM adapted to our clickbait summarizer.

About

Continuous Conditional text generation of clickbait newspaper articles. Using fine-tuned Pegasus text summarizer, with Fudge and PPLM conditional text generation models.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published