Deep Learning-Based Abstractive Summarization for Brazilian Portuguese Texts

Abstract

Automatic summarization captures the most relevant information and condenses it into an understandable text in natural language. Such a task can be classified as either extractive or abstractive summarization. Research on Brazilian Portuguese-based abstractive summarization is still scarce. This work explores abstractive summarization in Portuguese-based texts using a deep learning-based approach. The results are relatively satisfactory considering the ROUGE measurements and the quality of the generated summaries. Still, there are some problems regarding coherence, readability, and grammar. We strongly believe they are linked to the inherent complexity of generating an abstract and the degradation of text quality by the translation steps. These results should be seen as preliminary, serving as a basis for future research.