Irina Bejan


2023

pdf bib
Make Every Example Count: On the Stability and Utility of Self-Influence for Learning from Noisy NLP Datasets
Irina Bejan | Artem Sokolov | Katja Filippova
Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing

Increasingly larger datasets have become a standard ingredient to advancing the state-of-the-art in NLP. However, data quality might have already become the bottleneck to unlock further gains. Given the diversity and the sizes of modern datasets, standard data filtering is not straight-forward to apply, because of the multifacetedness of the harmful data and elusiveness of filtering rules that would generalize across multiple tasks. We study the fitness of task-agnostic self-influence scores of training examples for data cleaning, analyze their efficacy in capturing naturally occurring outliers, and investigate to what extent self-influence based data cleaning can improve downstream performance in machine translation, question answering and text classification, building up on recent approaches to self-influence calculation and automated curriculum learning.

2020

pdf bib
MemoSYS at SemEval-2020 Task 8: Multimodal Emotion Analysis in Memes
Irina Bejan
Proceedings of the Fourteenth Workshop on Semantic Evaluation

Internet memes are one of the most viral types of content in social media and are equally used in promoting hate speech. Towards a more broad understanding of memes, this paper describes the MemoSys system submitted in Task 8 of SemEval 2020, which aims to classify the sentiment of Internet memes and provide a minimum description of the type of humor it depicts (sarcastic, humorous, offensive, motivational) and its semantic scale. The solution presented covers four deep model architectures which are based on a joint fusion between the VGG16 pre-trained model for extracting visual information and the canonical BERT model or TF-IDF for text understanding. The system placed 5th of 36 participating systems in the task A, offering promising prospects to the use of transfer learning to approach Internet memes understanding.