WikiDes: A Wikipedia-based dataset for generating short descriptions from paragraphs

As free online encyclopedias with massive volumes of content, Wikipedia and Wikidata are key to many Natural Language Processing (NLP) tasks, such as information retrieval, knowledge base building, machine translation, text classification, summarization. In this paper, we introduce WikiDes, a novel dataset generate short descriptions articles for the problem The consists over 80k English samples on 6987 topics. We set up two-phase summarization method — description generation (Phase I) candidate ranking II) strong approach that relies transfer contrastive learning. For generation, T5 BART show their superiority compared other small-scale pre-trained models. By applying learning diverse input from beam search, metric fusion-based models outperform direct significantly ≈ 22 ROUGE in topic-exclusive split topic-independent split. Furthermore, outcome Phase II supported by human evaluation 45.33% chosen 23.66% I against gold descriptions. aspect sentiment analysis, generated cannot effectively capture all polarities paragraphs while doing task better automatic new reduces efforts creating them enriches Wikidata-based graphs. Our paper shows practical impact since there thousands missing Finally, expect WikiDes be useful related works capturing salient paragraphs. curated is publicly available at: https://github.com/declare-lab/WikiDes . • Wikipedia- Two-phase – ranking. Sentiment consistency analysis between paragraph description.