Towards a Comprehensive Assessment of the 
Quality and Richness of the 

Europeana Metadata of food-related Images

Yalemisew Abgaz1, Amelie Dorn2, José Luis Preza Díaz2 & Gerda Koch3 
1Adapt Centre DCU, 2ACDH-CH OeAW, 3Europeana Local - AT

AI4HI-2020 Virtual Workshop @ LREC2020
 26.05.2020

https://drive.google.com/open?id=1NenlAcA4q298-lzb4cgIZe3mjpTnxlaT
https://drive.google.com/open?id=1NenlAcA4q298-lzb4cgIZe3mjpTnxlaT
https://drive.google.com/open?id=1NenlAcA4q298-lzb4cgIZe3mjpTnxlaT
https://chia.acdh.oeaw.ac.at/ai4hi-2020-workshop/
https://lrec2020.lrec-conf.org/en/


Background

ChIA

• Interdisciplinary Digital Humanities project (2019-2021)
• Involved expertise: 

Digital Humanities, AI & NLP (ACDH-CH OeAW, AT)
Semantic technologies (Adapt Centre, IE)
Cultural Image aggregation (Europeana Local - Österreich, AT)

• Projet aim & results: the ChIA system - enabled increased 
access and analysis possibilities of cultural (food) images 
for content providers and educational purposes



Background

Europeana data set
Total: 58.6 Mio digital objects
Includes: 34.2 Mio digital images
from: 3.500 institutions in 42 countries 



The Problem
Dataset:
Selection based on food context of images
42.969 images (available with Free Access licenses) were selected 
in form of various sets (baskets) for later download & analysis of 
metadata and images



The Problem
Metadata
Although descriptions are available they seldom tell all details on 
what is depicted in the images. In most cases content descriptions 
use iconographic phrases like “fruits”, “flowers”, “still life”.



The Problem

Vocabularies
- Some institutions deliver metadata to Europeana that 

already includes vocabulary URIs. 
- Europeana enrich semantic connections with vocabularies 

such as AAT, ULAN, IconClass, VIAF, LCSH
http://data.europeana.eu/concept/base/222
http://iconclass.org/41A671 

- No specific food or drink related vocabularies are used.
- Irregularity in the use of semantics across the dataset.

http://data.europeana.eu/concept/base/222
http://iconclass.org/41A671


The problem

The current image collection
- mostly have metadata that is focused on bibliographic and 

format related but lacks domain-specific metadata

- most records are not interlinked on the basis of content 
(only when joint vocabularies are used)

- Thus,the current metadata needs analysis on the basis of
- The quality of the current metadata
- The use of multiple domain-specific vocabularies
- The gap between what the image depicts and the metadata 

express



Proposed Solution

Analyse the richness of the metadata using
- Quantitative approach

- using objective quality assessment metrics

- Qualitative approach
- using expert judgement on the expressiveness of the 

metadata
- Semantic enrichment to fill the gap

- Computer vision
- Semantic annotation



Quality Analysis Metrics

Metrics in four categories are selected 
- Contextual

- indication of used vocabularies
- Intrinsic

- extensional conciseness 
- Accessibility

- links to external LD providers
- Representational



Semantic Richness Analysis 

Semantic Richness: The availability of multiple descriptors of a 
resource particularly representing the main concepts 
represented by the target. 

 
More semantics for this image

- Fruits
- apple
- grapes
- Rose flower? etc

- Objects
- vase
- bowl?

- Culture
- what culture does it 

represent?
- rich/poor?



Results so far

Our initial analysis shows that the metadata
- is rich in bibliographic information
- provides labels with multiple languages
- however, lacks semantic richness 

Our Current work
- use of selected vocabularies to quantify semantic richness
- Analysis of the images with computer vision has a potential 

to address the richness problem
- Preparation of a training set is underway



Conclusion

- It is important to provide quality metadata to improve 
search and retrieval of historical images

- Semantic richness is a key for the search and 
exploration of historical images

- Understanding the gap is crucial to do the semantic 
annotation

- The use of computer vision combined with expert 
annotation and evaluation has a potential to improve 
both semantic richness and quality.



Thank you for listening

Any Questions?

Or ideas for collaboration?

Yalemisew.Abgaz@adaptcentre.ie
Amelie.Dorn@oeaw.ac.at

kochg@europeana-local.at

#chia4dh

@yalemisew
@adooorn

@Europeanaeu
@jlprezadiaz

Check out our website →  https://chia.acdh.oeaw.ac.at 
13

https://chia.acdh.oeaw.ac.at