Αναγνώριση προσβλητικού λόγου σε Ελληνικά tweets, με χρήση τεχνικών Βαθιάς Μάθησης
Date Issued
June 13, 2024
Type
Πτυχιακή Εργασία
Abstract
The frequent appearance of hateful content on social media has created the need for automatic identification of these posts, as moderators do not have the time to process and reject them. The field of computer science that deals with the recognition of meaning in text corpora is artificial intelligence and more specifically the subfield of Natural Language Processing (NLP). Recent developments, based on deep learning architectures such as transformers, have given a huge boost to the understanding of natural language. In this work, I use and adapt a pre-trained transformer model, GreekBERT, which is in turn an adaptation of the original BERT to Greek texts. For the training/adaptation of the pre-trained model, I use a labeled dataset consisting of tweets written in Greek. After preprocessing the dataset, the model is trained with different combinations of hyperparameters in order to achieve the best results. The final model, which is free for everyone to use, has been published on the Hugging Face platform.
Subjects