Karen Stepanyan, UCL
From Digitised Student Newspapers to Wikidata: Lessons from UCL's Special Collections
Pi is a UCL student periodical that has been in publication since 1946. UCL holds copies of student
periodicals, some of which have been digitised and are accessible as part of UCL Special Collections.
Accessing PDF documents, even when OCR-ed, creates challenges when using them to understand
or search across the corpus rather than within a single issue.
This talk outlines the lessons learnt from a collaborative Knowledge Exchange and Innovation project
between the Department of Information Studies, UCL Special Collections and Wikimedia UK,
exploring solutions for making the special collections more accessible. The project focused on a
subset of digitised student periodicals, namely the issues of Pi published between 1946 and 1954.
Such collections offer a direct record of what students at British universities wrote and read about,
and this subset is particularly interesting as it covers the post-war decade of student life in London
and at University College in particular. The lessons learnt are, however, applicable to other projects
and research strands that use historic periodical archives.
The main aim of the project was to use automation techniques for the semantic annotation of the
digitised artefacts, and to publish some of the extracted information in the form of a graph, more
specifically, on Wikidata. The project explored various routes but settled on a pipeline that carried 88
digitised issues (379 pages, 1946–1954) through LLM-based OCR and article segmentation, named-
entity recognition, geocoding, topic classification, and reconciliation to Wikidata. The project released
3,366 extracted articles openly under a DOI, published catalogue records for every issue used in the
pipeline to Wikidata, and enriched existing Wikidata records with mention statements against 231
person items. The processing pipelines were made openly available and documented on GitHub, on
Wikidata project pages, and on an institutional page. The talk discusses the primary challenges faced
throughout the project, as well as the elements of success in publishing a local collection into a
shared knowledge base. The resulting dataset was structured in the same format as the Newswire
dataset 8 . While the result is a graph of mentions rather than an interconnected knowledge graph, the
exercise mapped the scale and types of challenges faced when working with historic periodicals.
These include layout-aware article segmentation, the disambiguation of people and places, and a
topic scheme that bought interoperability with Newswire at the cost of fit to a British student paper.
The dataset has since been reused in teaching a postgraduate information retrieval module at UCL.
Students work with a repository of writing produced at the institution at which they study, and can
consult the physical or scanned copies of the newspapers from which the dataset was built. The
module introduces Elasticsearch and Kibana, and progresses to the integration of Elasticsearch
indexing with Retrieval-Augmented Generation.
Nicky Nicolson, Kew Jordell Laboratory
Knowledge Structures in Biodiversity - Linking Specimens, Literature and Evolving Interpretations
Our knowledge of biodiversity is grounded in physical, examinable objects: the global corpus of specimen collections. These provide essential "what, where, when" data - what material was collected, from where in space, and when in time. Comprehensive specimen digitisation and data mobilisation have progressed alongside the digitisation of the scholarly record, the scientific literature in which these specimens are described, formally named and contextualised.
We have recently developed the concept of the "digital extended specimen" - a network of links from a specimen to the information and context through which it can be interpreted. The scholarly record includes not only systematic knowledge - our understanding of which species exist and how they interrelate - but the evolving record of evidence, interpretation and changing understandings through which our current state of knowledge has developed. The digital availability of specimen and literature data together with the application of new computational approaches opens up the possibility of investigating our past working practices and developing new ways of working. This talk aims to give an overview of biodiversity informatics as a particularly rich domain for knowledge representation: one in which physical objects, structured data, scholarly literature and evolving scientific interpretations can be connected, explored and transformed through new information architectures.
Martin Poulter, Bristol, Wikidata
Wikidata as a Cultural Aggregator
Many different kinds of sources can help people research or study culture and history: biographies, maps, bibliographies, art catalogs, auction records, and so on. A knowledge graph can make this knowledge findable, useful, and visible in ways that were previously impractical. Wikidata in particular is aggregating a huge amount of data about art, artists, and historic events. This talk reports on some uses of Wikidata as a platform for cultural data at the University of Oxford and a private art collection.
Robert Allen, Organizer
Supporting Exploration of Dynamic, Structured Semantic Models
Hypertexts require structured knowledge and interfaces for browsing that knowledge. We have explored knowledge structures for semantic (qualitative) models, such as those used in scientific research reports. For example, we focus on the Earth system model from the theory of Snowball Earth. Here, we extend the work on visual interfaces to provide interaction with structured system descriptions. We extend a previously developed widget and describe three new ones to support explanations of interaction details among system components. Specifically, we show lower-level mechanisms and the temporal relationships among events. By modeling mechanism details, we can understand possible event interactions. In addition, we describe steps toward state-based semantic programs for those mechanisms.
Aldrich (Zhujun) Pan, UAL
Guiding Generative Storytelling with Knowledge Graphs
Large language models (LLMs) have shown great potential in story generation, but challenges remain in maintaining long-form coherence and effective, user-friendly control. Retrieval-augmented generation (RAG) has proven effective in reducing hallucinations in text generation; while knowledge-graph (KG)-driven storytelling has been explored in prior work, this work focuses on KG-assisted long-form generation and an editable KG coupled with LLM generation in a two-stage user study. This work investigates how KGs can enhance LLM-based storytelling by improving narrative quality and enabling user-driven modifications. We propose a KG-assisted storytelling pipeline and evaluate it in a user study with 15 participants. Participants created prompts, generated stories, and edited KGs to shape their narratives. Quantitative and qualitative analysis finds improvements concentrated in action-oriented, structurally explicit narratives under our settings, but not for introspective stories. Participants reported a strong sense of control when editing the KG, describing the experience as engaging, interactive, and playful.