taln-ls2n/pubmed
收藏资源简介:
--- annotations_creators: - unknown language_creators: - unknown language: - en license: - unknown multilinguality: - monolingual size_categories: - 1k<n<10k task_categories: - text-generation task_ids: [] pretty_name: PubMed tags: - keyphrase-generation - keyphrase-extraction - text-mining --- # Schutz 2008 PubMed dataset for keyphrase extraction ## About This dataset is made of 1320 articles with full text and author assigned keyphrases. Details about the dataset can be found in the original paper: Keyphrase extraction from single documents in the open domain exploiting linguistic and statistical methods. Alexander Thorsten Schutz. Master's thesis, National University of Ireland (2008). Reference (indexer-assigned) keyphrases are also categorized under the PRMU (<u>P</u>resent-<u>R</u>eordered-<u>M</u>ixed-<u>U</u>nseen) scheme as proposed in the following paper: - Florian Boudin and Ygor Gallina. 2021. [Redefining Absent Keyphrases and their Effect on Retrieval Effectiveness](https://aclanthology.org/2021.naacl-main.330/). In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4185–4193, Online. Association for Computational Linguistics. Text pre-processing (tokenization) is carried out using spacy (en_core_web_sm model) with a special rule to avoid splitting words with hyphens (e.g. graph-based is kept as one token). Stemming (Porter's stemmer implementation provided in nltk) is applied before reference keyphrases are matched against the source text. ## Content The details of the dataset are in the table below: | Split | # documents | # keyphrases by document (average) | % Present | % Reordered | % Mixed | % Unseen | | :--------- | ----------: | -----------: | --------: | ----------: | ------: | -------: | | Test | 1320 | 5.40 | 84.54 | 9.14 | 3.84 | 2.47 | The following data fields are available: - **id**: unique identifier of the document. - **title**: title of the document. - **text**: full article minus the title. - **keyphrases**: list of reference keyphrases. - **prmu**: list of <u>P</u>resent-<u>R</u>eordered-<u>M</u>ixed-<u>U</u>nseen categories for reference keyphrases. **NB**: The present keyphrases (represented by the "P" label in the PRMU column) are sorted by their apparition order in the text (title + text).
The Schutz 2008 PubMed dataset contains 1,320 articles with full texts and author-assigned keywords. The keywords of this dataset are categorized under the PRMU (Present-Reordered-Mixed-Unseen) scheme. Text preprocessing was performed using spaCy for tokenization and NLTK's Porter Stemmer for stemming. The dataset includes unique document identifiers, titles, full text bodies, keyword lists, and the PRMU classification of the keywords.
Schutz 2008 PubMed dataset for keyphrase extraction
数据集概述
- 语言: 英语 (en)
- 许可证: 未知
- 多语言性: 单语种
- 大小: 1k<n<10k 文档
- 任务类别: 文本生成
- 标签: 关键词生成, 关键词提取, 文本挖掘
数据集内容
- 文档数量: 1320篇
- 平均每文档关键词数量: 5.40
- 关键词分类:
- Present (P): 84.54%
- Reordered (R): 9.14%
- Mixed (M): 3.84%
- Unseen (U): 2.47%
数据字段
- id: 文档唯一标识符
- title: 文档标题
- text: 全文(不含标题)
- keyphrases: 参考关键词列表
- prmu: 参考关键词的<u>P</u>resent-<u>R</u>eordered-<u>M</u>ixed-<u>U</u>nseen分类列表
数据预处理
- 分词: 使用spacy (en_core_web_sm模型),特殊规则避免分割带有连字符的单词
- 词干提取: 使用nltk中的Porters stemmer实现



