Embedded Files
Above: Library and Information Services Metadata Librarian Mary Rose, from the Southern Illinois University Edwardsville Lovejoy Library, explains metadata.
What is it?
Metadata recognition refers to the automatic identification of subcategories of informational fields such as reference metadata that are attached to an item record Reference metadata refers to any characteristic identifying information about an item, usually necessary to search for the item, such as title, author, subject headings, publisher, publication date, or any other sub-fields of the item record (Yang et. al., 2021).
Templates
One method of using machine learning techniques to identify metadata for a specific library resource is through template-based approaches. Templates need to be created through code which tell the software system exactly which pieces of metadata to identify, and to connect across formats to another software system, format, or output (Yang et. al., 2021). The template-based approach is heavily used for citation extraction technologies in the library setting. Metadata librarians can choose which identifiers are extracted such as title or author, which creates a string of attached metadata connecting the resource’s citation to a large body of resources that the machine has learned from or is continually learning from as more items are added to the collection, and more users continuously create metadata (Yang et. al., 2021).
Current and Future Uses
Current and future uses of metadata recognition technologies are being utilized in major media companies, trained and maintained through streaming services (Tilton et. al., 2020). Metadata can be used through media formats to recognize the locations in television and movies where filming took place, automatically identify actors in media resources, and to connect related items for ease of searching (Tilton et. al., 2020). Academic libraries, public libraries, and special libraries who house resources specific to the film industry would benefit from this use of metadata recognition technology to assist in searching, identification, and increased accuracy of information of items in their collections.
Use Case: The Map and Imagery Laboratory (MIL) at the University of California Santa Barbara
Geographic information retrieval (GIR) is a highly complex type of machine learning which is not strictly based on key search terms in order to recognize objects to retrieve them. This type of metadata recognition relies on some level of predictive, fuzzy metadata, approximate locations, and forms the basis for “intelligent” technologies being currently researched (Chen et. al., 1997). Libraries would benefit from the addition and utilization of this technology when searching historical images and maps which may feature out of date location names. The user would not have to type an exact, pre-registered search term to locate a resource, instead, the software would use the metadata attached to the item to predict what the user is searching for, offering suggestions which feature metadata that is closely similar to that of the item they are searching for by term (Yang et. al., 2021). GIR is a semantic and neural network-based technology that connects items of various formats, including multi-media items, in a concept-based learning algorithm that spreads, allowing the machine to perform its own reasoning to find predicted resources that the user requests (Yang et. al., 2021). Libraries that have image databases that are geo-referenced, such as at the Map and Imagery Laboratory (MIL) at the University of California Santa Barbara, are able to find patterns that occur within aerial and satellite images to automatically identify the location in the image, and then to index it. Such patterns have been used to identify and index images of various locations around Southern California by the MIL lab at UCSB (Yang et. al., 2021).
References
Chen, H., Smith, T. R., Larsgaard, M. L., Hill, L. L., & Ramsey, M. (1997). A geographic knowledge representation system for multimedia geospatial retrieval and analysis. International Journal on Digital Libraries, 1(2), 132-152. doi:10.1007/s007990050010
Tilton, L., Alexander, E., Malcynsky, L., & Zhou, H. (2020). The role of metadata in american studies. Polish Journal for American Studies : Yearbook of the Polish Association for American Studies, 14, 149-272.
Yang, T., Hsieh, Y., Liu, S., Chang, Y., & Hsu, W. (2021). A flexible template generation and matching method with applications for publication reference metadata extraction. Journal of the American Society for Information Science and Technology, 72(1), 32-45. doi:10.1002/asi.24391
Page updated
Report abuse