Abstract
Mark Michael Hall, Oier Lopez de Lacalle, Aitor Soroa Etxabe, Paul Clough, Eneko Agirre. Proceedings of the 6th Workshop on Language Technology for Cultural Heritage, Social Sciences, and Humanities.
2012.
Introduction
Digital Cultural Heritage (CH) collections, such as Europeana, represent vast repositories of historical knowledge. However, these archives often face a significant "discovery" challenge. While traditional search interfaces serve experts well, they can be intimidating for novice users who may not know which keywords to use. This lack of a clear starting point often leads to high "bounce rates," where casual browsers leave a digital library almost immediately.
To bridge this gap, researchers developed Wikiana, a system that transitions from a traditional search paradigm to an exploratory one. By injecting relevant cultural heritage objects directly into Wikipedia articles, Wikiana leverages a familiar interface to guide users toward deeper archival exploration.
The Wikiana System
Wikiana serves as a middleware layer between users and data providers, specifically Wikipedia and Europeana. The system architecture is designed to facilitate seamless data flow and augmentation.
The system performs augmentation at two distinct levels:
Article-Level Augmentation
At the top of a Wikipedia page, Wikiana displays a gallery of up to 20 relevant items. This is achieved by mapping Wikipedia anchor text to articles and matching Europeana metadata (primarily the dc:title field) against a custom "Wikipedia dictionary."
To ensure the gallery remains engaging, the authors implemented an "Interestingness" metric ($\alpha_i$) based on Inverse Document Frequency (IDF). This formula identifies "unusual" items by analyzing the frequency of metadata values across the entire collection, effectively surfacing unique "hidden gems" that might otherwise be buried by common search results.
Paragraph-Level Augmentation
As users read, individual paragraphs are augmented with a single relevant thumbnail. The system extracts keywords from hyperlinks within the text to generate boolean queries for the Europeana OpenSearch API. This contextual augmentation provides immediate visual hooks related to the specific topics being read.
Evaluation and Results
The effectiveness of Wikiana was evaluated through a user study and a three-month analysis of web logs comparing Wikiana to the main Europeana portal.
- Engagement: Users originating from Wikiana viewed a median of 11 items per session, notably higher than the 8 items viewed by general Europeana users.
- Retention: Only 19% of Wikiana users left after viewing three or fewer items, compared to a 30% bounce rate for general users.
- Relevance: In the user study, 25% of automatically injected paragraph images were rated as "appropriate," and 36% were deemed "interesting." While these figures highlight room for improvement, they establish a strong baseline for fully automated discovery.
Limitations and Future Work
The current system faces challenges regarding the "appropriateness" of paragraph-level images, often due to noisy keyword extraction or sparse metadata. Future work will explore applying the "interestingness" ranking to paragraph-level results and utilizing advanced Natural Language Processing (NLP) to enrich archive metadata, further refining the mapping between Wikipedia content and cultural heritage objects.
Summary updated