EB-KG: Knowledge Graph of the first 8 eiditions Encyclopaedia Brittanica (1768-1860)

<p>This Knowlege Graph represents the information of the first eight editions of Encyclopaedia Brittanica (years: 1768 to 1860) in RDF (ttl format).</p> <p>The raw dataset is provided by the NLS in this <a href="https://data.nls.uk/data/digitised-collections/encyclopaedia-britannica/">link</a> , and it comprises of eight editions and a total of 195 volumes with a total size of 44GB. It uses two XMLs schemas: METS&nbsp; for descriptive, structural, technical and administrative metadata (Title, Author, Publisher, etc); and ALTO&nbsp; for encoding the OCR text of a page.</p> <p>In this work, we have extracted the information from METS and ALTO XMLS using <a href="https://github.com/francesNLP/defoe">defoe</a> tool and developed <a href="https://github.com/francesNLP/defoe/tree/master/defoe/nlsArticles/queries">novel information extraction heuristics</a>. With the extracted information, we created the EB-KG Knowlege Graph, which&nbsp; uses the <a href="https://francesnlp.github.io/EB-ontology/doc/index-en.html">EB Ontolgy</a>, to represent such information. Furthermore, during the information extraction phase, we have employed several techniques to mitigate two common OCR errors: long-S and the line-break hyphenation.</p> <p>The EB-KG contains 1,638,239 RDF triples. It has information from 8 editions. Each edition can have several Volumes, references to Books, Supplements; it also has an Editor and a Publisher, which can be a Person or an Organization. A Volume has several Pages, which can contain several Terms. And a Term can be either a Topic (a term described across several pages, often combining text, pictures, and tables.) or an Article (a description of the term in one- or two-paragraph long text (similar to an entry in a dictionary)). The data model of the EB-KG can be found <a href="https://francesnlp.github.io/EB-ontology/doc/dataModel.png">here</a>.</p> <p>The original ALTO files do not indicate the start and end of each EB term, the first part of our work involved the<br> automated extraction of all terms (along with their metadata) across editions, so they can be analysed independently without the surrounding text.</p> <p>&nbsp;</p> <p>&nbsp;</p> <p>&nbsp;</p>