Extracted and NER-ed Pi Newspaper Articles

JSONL records for each issue of digitised Pi (student periodical from UCL Special Collections) at UCL*. The issues are grouped into folders by publication date.<i>*Disclaimer: The dataset was compiled by extracting articles from scanned images using an LLM model. Permission to do so was requested from UCL Library Services. However, the dataset cannot be used as an accurate representation/copy of the information published in Pi. The dataset should be used for educational purposes and for benchmarking future work on OCR-ing, semantic processing and information retrieval. </i>Each record is formatted using JSON resembling the Newswire dataset of historical newspapers for interoperability.Each JSONL file contains:a list of articles OCR-ed from high scanned images using OpenAI's LLM model (i.e. o4-mini-2025-04-16). Due to the limitation of the LLMs the dataset cannot be treated as complete or accurate.results of the Named Entity Recognition processing that used spaCy library with labels that are not limited to people and locations.list of Wikidata Q-IDs and labels for individuals in each article as linked by combining NER processing and Wikidata search query (with Fuzzy matching).list of geocoded locations mentioned in each article.list of the 12 topics an article is associated with using Newswire pre-trained topic classification models.Additional information is on:GitHub: Newspaper-OCR-LLM - https://github.com/kstepanyan/Newspaper-OCR-LLM/tree/mainGitHub: Newspaper-Semantic-Enrichment - https://github.com/kstepanyan/Newspaper-Semantic-Enrichment/tree/main/2-semantic-processingWikidata Project Page: https://www.wikidata.org/wiki/Wikidata:UCL_periodicals<br>Dataset: v. 0.1 (pilot)