This paper presents my thesis proposal on automatic metadata curation of cultural heritage resources. Metadata curation represents one of the aspects of the "metadata management life-cycle" together with ingestion, maintenance and dissemination. The objective of metadata curatorship in general is to ensure the user can access objects of interest to him/her from a repository, digital library, catalogue, etc. using well-assigned metadata values aligned with an appropriately chosen schema. Ideally, in a repository or digital archive, all the objects should be described using the same accuracy and data structure. In this way the user can retrive all the informations, objects of interest and the related items as result of a single search. However, this is very rare. Objects are described using di erent levels of details and di erent data model design might be used by the cataloguers. In such cases, after the metadata quality control process, the curator must correct or normalize all the objects with errors or incosistencies in the metadada schema. New approaches based on the use of the semantic enrichment of the digital resources can help the resources aggregator during the metadata curation process having an impact also in the improvement of resources discovery.
The traditional concept of library has undergone a profound change: from a collection of physical information resources (mostly books) to a collection of digital resources. In addition, the notion of digital resource includes not only texts in digital form, but also, in general, any kind of multimedia resources3. Hence, making accessible cultural heritage resources requires metadata schemas rich in semantics and a structure able to cover the material heterogeneity and the variety of memory institutions (libraries, archives, museums). However, often 3 Such collections may be composed of text, written on di erent materials, paintings, photographs, 3D objects, sound recordings, maps or even digital object. we are facing with problems related to the low quality of the metadata used for the description of the digital resources: wrong de nition, inconsistence of the resources or resources described according only to the minimal mandatory metadata entities4. There may be many reasons for that, all completely valid, e.g.; in many cases these institutions have few human resources to work on improved metadata, they are often not themselves the sources of the metadata.
The goal of the present work is to improve metadata quality integrating semantic web principles into the metadata curation process. 2
Digital curation, broadly interpreted, is about maintaining and adding value to a trusted body of digital information for both current and future use: in other words, it is the active management and appraisal of digital information over its entire lifecycle.
The necessity of a lifecycle approach, to ensure the continuity of digital
material, is discussed by Pennock [
The curation framework developed by Bruce and Hillmann [
In the context of linguistic resources, the CLARIN7 consortium supports
metadata curation developing a metadata curation module to facilitate the
metadata ingestion and curation process of the Virtual Language Observatory (VLO).
In the context of Semantic Web the metrics were often not explicity de ned and
did not consist of precise statistical metrics. Moreover, only few approaches were
actually accompained by an implemented tool and none of them covered all the
data quality dimensions [
From the literature analysis it can be inferred that the existing approaches
are either too abstract or extremely focused on one dimension e.g.; completeness
[
Wrong and incomplete mappings a ect the discoverability and acessibility of the resources for the users: metadata curation plays an essential role to the improvement of the metadata quality. In a standard cycle, the resurces are checked by the curators which, applying the metrics described in the previous paragraph, analyze the metadata quality. Depending on the types of issues curators intervene as follows : { Send back to the content provider the datasets or records in order to x the incosistence/lacks/errors. { Fix by hand the incosistence/lacks/errors.
{ Normalize the resources using a controlled vocabulary.
This approach, even if it can be considered as a standard procedure, leads to problems related to e ective impact of the curation process: In most of the cases curation is done by hand by the curators, quality analysis involves just the mandatories metadata elements8 and metadata quality analysis does not consider all the parameters suggested by Bruce and Hillmann.
For a better contextualization of the problems we can brie y refers to Cultura Italia9, the italian digital library. Resources are integrated in Cultura Italia in the form of metadata using PICO10 pro le. It is based on the international standard language Dublin Core11 that can describe, in a single scheme, every type of cultural resource, both physical and digital. Dublin Core consist of 15 main elements. PICO pro le adds to these others 37 elements conceived for the application of the Dublin Core in Cultura Italia. We can identify two types of errors: { Low metadata completeness: Objects are described using the 6 mandatory metadata elements from PICO for metadata harvesting. Elements like dc:description or dc:author which are considered as optional are under represented E.g in the dataset from "Regione Piemonte" which consists of 71.710 records the element dc:description is never used. { Low accuracy: Metadata are lled using non exaustive informations like dc:title "photo".
In the light of the problems underlined above our research question will be: "Can we improve metadata quality with automatic curation techniques and semantic web technologies?". This topic will be treated in the domain of the cultural heritage. 8 Crucial information like "description" lacks quality evaluation 9 http://www.culturaitalia.it 10 http://www.culturaitalia.it/opencms/documentazione_tecnica_it.jsp? language=it&tematica=static 11 http://dublincore.org
The solution we are going to propose is a framework which, using the
metrics from Bruce and Hillmann, aims to check automatically12 the issues derived
from wrong cataloging processes in order to optimize the metadata curation
work ow. Moreover it aims to improve the metadata quality by suggesting the
missing metadata elements or errors to the curators. Considering the scenario
from Cultura Italia described in the previous paragraph the curation process
will be characterized by the following main tasks:
{ Automatization of the quality metrics: This PhD project will be mainly
focused on the quantitative de nition of the Completeness, Accuracy and
Logical consistency and coherence metrics. For each of these metrics we will
be de ne a customized algorithm[
Then, the metadata creator can decide whether to accept the suggestions. { Evaluation methodology: validation will concern two aspects: Metadata schema: compliance of the new elements with the standards structure of the metadata pro le.
Consistence of the new elements with respect to the context of the digital object: evaluation made by the logical consistency and coherence parameter. 12 Manual curation processes are of course limited in their coverage: the amount of objects that can be curated in datasets like the ones of Cultura Italia or other aggregators is too high for the human resources available In order to achieve the best results two di erent methods will be tested: De nition of a "gold standard" dataset: will be used in order to train the framework about how the correct metadata schema should be in terms of structure and content.
Insert (intentionally) some errors in the well de ned resources in order to check if the proposed system can recognize the errors. 5
Considering the amount of digital archieves, problems related to metadata curation becomes evident. Reasons may be di erent: There is no curation task force, the metadata curation activity is delegated to the content providers or the metadata curation activity is made by hand. The development of an automatic process will enable the curators to not only obtain snapshots of the quality of a repository, but also to constantly monitor its evolution and how di erent events a ect it without the need to run costly human e ort. This could lead to the creation of innovative applications based on metadata quality that would improve the nal user experience.