Using Pre-Trained Language Models for Abstractive DBPEDIA Summarization: A Comparative Study

Purpose: This study addresses the limitations of current short abstracts DBPEDIA entities, which often lack a comprehensive overview due to their creating method (i.e., selecting first two-three sentences from full abstracts). Methodology: We leverage pre-trained language models generate abstractive summaries in six languages (English, French, German, Italian, Spanish, and Dutch). performed several experiments assess quality generated by models. In particular, we evaluated using human judgments automated metrics (Self-ROUGE BERTScore). Additionally, studied correlation between evaluating under different aspects: informativeness, coherence, conciseness, fluency. Findings: Pre-trained more concise informative than existing abstracts. Specifically, BART-based effectively overcome abstracts, especially for longer ones. Moreover, show that BERTScore ROUGE-1 are reliable assessing informativeness coherence with respect also find negative conciseness ratings. Furthermore, fluency evaluation remains challenging without judgment. Value: has significant implications various applications machine learning natural processing rely on resources. By providing succinct summaries, our approach enhances contributes semantic web community.