«««< HEAD Textile research helps us understand how people live, adapt, and express themselves. By studying what people wear and how fabrics are made and traded, we uncover stories of culture, identity, innovation, and global connection. In 2025, as environmental and social sustainability become ever more urgent, learning from the past and present of textiles can inform better decisions in fashion, labour, and the environment. It’s not just about clothing—it’s about how threads connect us across time and place.

The development of TextileBase compels us to address complex challenges in data curation, validation, and interoperability—issues that are highly relevant across both commercial and academic domains. Because most textiles have a relatively short lifespan, their study often relies on secondary sources such as technical and artistic drawings, colour plates, and photographs that document how textiles were made, used, and worn. These sources introduce multimodality: our system must integrate and relate knowledge drawn from physical artefacts, digitised images, and descriptive texts. Moreover, the data originates from a wide range of institutions—research centres, museums, archives, and libraries—each employing distinct metadata standards. TextileBase is therefore not only a tool for textile historians, but also a model for handling complex, multimodal, and decentralised data in broader sectors.

The first published dataset, Linked Open Datasets on Garments from the Latgale Region contains data on Latvian traditional shirts and skirts from the Latgale region in Eastern Latvia. The female and male shirts, and the skirts in the dataset are handmade and were worn in the 19th century. They represent both festive and daily wear of the local female and male peasants. The shirts are stored at the National History Museum of Latvia and the Ethnographic Open-Air Museum of Latvia. The data contain information on the locality of their origin, their approximate date of creation with various precisions, the materials they are made of, and the way of their fabrication, as well as their purpose of wearing (festive or daily wear) and wearer’s ethnicity and gender. They also include the name of the museum each shirt is stored at, supplemented with its unique inventory number. Data on some sample shirts also include a photo of the shirt.

A Semantic Knowledge Graph for Textile and Dress History

2e3d90dfbfa8fae4b2d532c2d8067737a2653b7c

TextileBase is a knowledge graph that interconnects databases and secondary sources—such as artefact photographs, contemporary images, drawings, and written descriptions—related to textile and dress history. It supports diverse areas of textile research, from historical dress studies to sustainable fashion.

«««< HEAD

• See every entry in the database: All items

How TextileBase Works

TextileBase is an intelligent, modular system designed to support collaborative textile research and data sharing. Its main interface is a user-friendly Wikibase instance—built on the same software that powers Wikimedia projects—which allows researchers and textile enthusiasts to record, link, and validate knowledge.

Behind the scenes, TextileBase integrates advanced methods from digital humanities. It uses an ontological data model based on CIDOC-CRM, RiC, and DCTERMS, ensuring semantic interoperability with museum, archive, and library systems.

TextileBase s supported by a graph database and an experimental SPARQL endpoint. This enables translation of its structured data into linked data graphs, which can be enriched and exchanged with heritage institutions.

The system is multilingual by design. By leveraging international thesauri and computational linguistics, TextileBase supports cross-language discovery—essential in regions like Latvia, where historical textile records exist in multiple languages and scripts, and artefacts have travelled widely across Europe.

Data import and export are powered by Reprex’s data curation and cleaning libraries. This modular design supports workflow automation and the integration of diverse, hard-to-find, multimodal sources. TextileBase is built to scale: it is cost-efficient, open-source, and adaptable for academic, commercial, or industry research teams of any size.

Technically speaking, building an interactive website for textile history is like setting up a webshop—except all the product information is either missing or fragmented. There are no barcodes, producers, or standard descriptions. Instead, the relevant information is buried in academic knowledge, secondary sources, or scattered fragments of primary data.

This is where AI can significantly assist: by gathering, linking, and inferring data at a speed and scale beyond human capacity.

Connecting Surviving Artefacts

Unlike metal objects or buildings, textiles are fragile and rarely survive intact. They are worn, used, and discarded. Even from the 19th century, very few textiles remain.

Our goal is to create a tool that connects fragmented museum inventories and catalogues across institutions—linking rare textile artefacts in a coherent, searchable network.

The Challenges of Linked Open Data in Textile History

While digital humanities frequently refer to “linked open data” or the “Internet of Things,” textile history is still at the early stages of adopting these frameworks. A key goal of TextileBase is to give each artefact—like the Seto shirt—a unique URI. But generating URIs for garments in small rural museums or private collections is often time-consuming and expensive.

TextileBase seeks to streamline this process—making it faster, cheaper, and more reliable.

The Importance of Secondary Sources

Historical collecting practices were often biased. Textiles worn by everyday people—peasants, laborers, minority groups—were rarely preserved. Instead, most surviving collections reflect the lives of elites or were gathered during ethnographic expeditions with colonial or exoticizing perspectives.

That’s why secondary sources—such as period photographs, illustrations, and books—are so valuable. They often unintentionally document everyday garments, offering insights missed by early collectors.

Case Study: Finno-Ugric Traditional Clothing

We focus on the clothing traditions of small Finno-Ugric communities—such as the Setos and Livonians—due to the technical and linguistic challenges involved.

For example, Livonian communities in present-day Latvia were largely assimilated by the late 19th century. Setos are now found in southeastern Estonia and Russia’s Pskov oblast. Their artefacts are scattered across Finnish, Estonian, and Hungarian archives, usually documented in those languages—rarely in Livonian, Seto, or even English.

We must build a historical namespace of place names, languages, even spelling history, and garment classifications to search and understand this data. This includes terms like “shirt” in Finnish (paita), Hungarian (ing), or Estonian, and accounts for changing styles, materials, and local terminologies. This is one of the uses of TextileBase: it contains knowledge about words, placenames, their morphology, so that researchers not familiar with the collection’s language (and its historical changes, abbreviations) can work globally.

Building a Wikimuseum for Dispersed Collections

Inspired by Wikimedia Estonia’s multi-language, open-access model, we propose a virtual museum—a Wikimuseum—that brings together:

• Artefacts from rural museums (e.g., Mõniste, Saatse, Värska)

• Items in national museums (Estonia, Finland, Hungary)

• Private collections that would never be physically exhibited together

This approach connects dispersed knowledge—across time, languages, and borders—into a shared digital space.

How AI Supports Textile Research

TextileBase incorporates AI-driven tools—language models, translation engines, inference systems—that can:

• Search for historical placenames across multilingual archives

• Identify garments in photos or scanned books

• Detect and match synonyms or spelling variants

• Flag potentially relevant images and documents for expert review

By operating within the TextileBase framework, our AI tools remain human-controlled and explainable, reducing the risk of hallucinated or misleading results.

2e3d90dfbfa8fae4b2d532c2d8067737a2653b7c