by Shiyang Jiang

The Media Ecology Project and Metadata
Schema
Crosswalk and Cleanup
Digital Assets
Database
Conclusion

The Media Ecology Project and Metadata

In its effort to provide access to valuable moving image resources, the Media Ecology Project (MEP) has worked closely with cultural heritage organizations to assemble a remarkable collection for scholarly research. Some highlights in the collection include extremely high-resolution digital scans of the paper prints at the Library of Congress and all volumes of American Mutoscope and Biograph Picture Catalogue. As the MEP collection grows impressively, more effort is required to improve the accessibility and usability of these digital assets. This article summarizes the progress made in MEP’s metadata organization strategies from 2020 to 2022. While the work is still ongoing, this article summarizes what MEP has achieved thus far and reflects on some remaining limitations.

Metadata—the information we create, store, and share to describe things—allows researchers to interact with these collections to obtain the knowledge they seek. Institutions such as libraries, archives, and museums have long been creating, using, and sharing structured metadata. Our archive partners, including many such institutions, have been generous in their partnership with MEP by supplying the digital files and the metadata that describe them. For this purpose, MEP designed a spreadsheet template based on Dublin Core. It includes mandatory fields such as title, identifier, creator, and date, as well as optional fields, such as location, source, country of origin, languages, and more, that archivists could add as they see fit. MEP has been able to collect these spreadsheets and use them for further research.

However, as the collections at MEP grew more and more diverse, some issues with this workflow arose. The original spreadsheet template was set up only for newsfilm collections, so it features a small set of controlled vocabulary like raw footage and the completed program that may not apply to other collections (see Table 1). Furthermore, each metadata spreadsheet delivered by MEP archive partners was in a separate file instead of consolidating information in a central resource. As a result, team members spent excessive time looking for the correct file. Finally, different collections have different cataloging rules, which results in varying degrees of detail. The lack of consistency hindered MEP’s goal of using machine-learning analysis to unlock the collection’s potential.

By improving metadata organization, researchers can better access and connect MEP’s collections to digital humanities platforms like the Minimum Viable Annotator and Scalar. With consistent metadata, users can easily reference a film while creating annotations in the Minimum Viable Annotator or articles on Scalar. To address these concerns and facilitate better collection management, MEP needs new strategies for organizing, cleaning, and consolidating the metadata.

Schema

Schema is an essential metadata component, describing the overall structure used to capture information about a resource. According to ISO 23081’s definition, “A schema is a logical plan showing the relationships between metadata elements, normally through establishing rules for the use and management of metadata specifically as regards the semantics, the syntax and the optionality (obligation level) of values.”¹

We looked closely at the metadata MEP received to lay the groundwork for the new schema. Some items are described in standard archival language, while others have more unique descriptions like shot lists, probably from the hand of local television crews or stock footage archivists. We recognize that preserving the original records is the best way to capture the stories of how these materials were created, distributed, and archived. As such, we chose to retain the original descriptions as faithfully as possible. 

Following this approach, we modified the existing schema and created a new application profile (see Table 2). The new schema contains most elements of the old one, as well as additional fields that reflect the research needs of film historians. The new schema has multiple “title” and “date” elements to capture the nuances better. Allowing multiple titles is especially important for early cinema. Due to the low survival rate of early cinema, titles are often the only thing that has withstood the ravages of time. In addition, many early films were referred to by different titles in different publications, making every title valuable to historical research. Furthermore, each item can also have metadata about its production date, release date, and copyright date, which are vital for understanding film production and distribution history. The new schema also defines more comprehensive sets of controlled vocabularies, such as the “color” fields. It covers more than “B&W” and “color” but also includes “tinted” and “toned”—a detail especially relevant for early cinema collections.
An important decision is whether to have one set of metadata elements (single entity) or establish groups of metadata elements (multiple entities). Multiple-entity models have the advantage of grouping elements around what we need to describe.² We chose the multiple-entity model for MEP as we can cover more details about collections, assets (films), files (local video files), streaming (online URLs), and annotation records. Additionally, we established rules about the relationships between entities, which would provide linkages between records and agents. For example, we define the relationship between assets and files as “one to many.” That way, even when a video has multiple parts, there will still be only one comprehensive record that describes it. These changes make digital assets much more manageable.

Crosswalk and Cleanup

The metadata crosswalk is an important procedure for maintaining good metadata quality. Crosswalk is the process of mapping the elements and values from one schema to those of another, but mistakes can occur regardless of how careful you are. We keep all original spreadsheets as a last resort for this issue. For example, identifiers for the United States Information Service films have leading zeros, which get automatically removed when the spreadsheets are in comma-delimited formats (CSV). It took us some time to discover the root of the problem. Thankfully, we kept all the original files and developed a workaround to preserve these zeros. 

Consistency is vital to high-quality metadata, so we ensured the new application profile was applied consistently. We used Open Refine for many procedures, including formatting all date fields to YYYY-MM-DD, implementing controlled vocabularies, and consolidating duplicated records. 

So far, the work has primarily focused on structured metadata. However, certain collections present unique challenges involving valuable data buried in plain text. The issue results from using an obsolete system that hosts the metadata or relying on optical character recognition (OCR) texts from books. To transform the plain text into a spreadsheet, we employed Python Regular Expressions to identify recurring headings in the texts and retrieve the necessary data. Additionally, Tesseract, an open-source program, was utilized to improve the recognizability and accuracy of the OCR output. After running numerous experiments, we were able to achieve a good result.

Digital Assets

Unique identifiers link the metadata records and thousands of video files. Any discrepancy between the two would cause massive confusion and problems for researchers. Derivative files add a further layer of complexity. We created many derivative videos from the originals to put on streaming platforms like Minimum Viable Annotators to reduce loading time. It is important to note that these files are not for close analysis.

The multiple-entity metadata model is essential to solving this challenge (see Table 3).  By categorizing each video as an entity, one film can have multiple versions listed under the “related file” field. We used MediaInfo—an open-source metadata reading application—to gather each file’s name, size, resolution, color space, and duration. A Python script collected all this information and turned it into CSV files. This comprehensive metadata not only helps researchers compare different versions and decide which one they need but also makes it simpler for project managers to calculate the size of the entire collection.

Database

A new information management system was needed to accommodate the new schema and thousands of records. After extensive research, we chose Airtable as our online database; an offline copy was also saved and regularly updated for security. Airtable combines the user-friendly features of a spreadsheet with the functionality of a traditional relational database. Furthermore, it is not static—any changes made to the records will be immediately visible on the user end. 

MEP uses Airtable to provide researchers with multiple points of access. With password-protected links, researchers can view, sort, and search selected metadata of one collection quickly and easily. These views can be embedded on any website to enable multitasking. Moreover, researchers with access to the entire database can search across all the records in one go using Airtable’s advanced search tool. Additionally, Airtable also has an application programming interface (API) to facilitate integration with other tools as a backend, and integration with Scalar is currently in progress. 

To make Airtable the central hub for metadata organization, we developed an ingest-and-export workflow. Excel and Google spreadsheets are still the default format for delivering new metadata to MEP, but adding and verifying new input are done through Airtable tools. The tool is handy for updating existing records. The import tool can take any spreadsheets or XML files and add them to the database as long as the headers are consistent with the schema. We also created tutorials that guide users on routinely maintaining the system.

Conclusion

MEP is continuously striving to make metadata from archives more accessible for researchers. Looking forward, MEP will continue making improvements to the current workflow. For example, an automated process could provide more reliable upkeep of an offline database copy. The creative use of automation tools with Google spreadsheets and Airtable could also save considerable effort while helping the ever-growing collection stay updated without constant monitoring. With these ongoing improvements, we anticipate that MEP’s metadata organization will become more efficient so that more researchers can benefit from its offerings.

Shiyang Jiang is an archivist at China Film Archive. She was previously a media annotator for the Media Ecology Project at Dartmouth College, where she worked to improve the interoperability and standardization of metadata. Shiyang holds a master’s degree in moving image archiving and preservation from New York University and a BFA in theater, film, television, and literature from Xiamen University.

Title Image: Table 2. New Application Profile (Preliminary)

Metadata and the Media Ecology Project © 2024 by Shiyang Jiang is licensed under CC BY 4.0.

The Journal of e-Media Studies is published by the Dartmouth Library.