Romanian text summarization using BERT and clustering algorithms

The Workflow is divided in 2 parts, training (1 to 2’) and evaluation (3 to 7):

1. From the wiki corpus a dictionary is generated containing all the words and their count
2. After tokenization and preparation for learning the model loads the corpus 2.1) Model loads the dictionary with words that appear more than a set parameter
3. GET request from browser extension sent to API
4. The API processes the request and evaluates the sentences from the article
5. Clustering algorithm evaluates the word embeddings generated by the model
6. Generated summary is sent to API
7. Browser extension receives the response and displays the summary to the use