Data Science & Metadata Research
To be discoverable by today’s online users, traditional library data must be transformed. OCLC Research analyzes bibliographic data to derive new meaning, insights, and services for use by library and information seekers. This work includes special projects, data science research, engagement with metadata communities, publications and presentations, and the creation of illustrative experimental applications.
Presentations
![Fast and Discriminative Semantic Embedding](/content/dam/research/presentations/2019/052719-wang-fast-descriminative-semantic-embedding.pptx.thumb.1280.1280.png)
Fast and Discriminative Semantic Embedding
Gothenburg, Sweden
We present a novel, effective and efficient method for term and document embedding method. Our experiments show it outperforms state-of-the-art methods in terms of the STS benchmark and subject prediction when trained on the same datasets, while at the same time being computationally cheaper by orders of magnitude.
Topics: Semantic Embedding
![An Innovative Approach to Scalable Semantic Embedding](/content/dam/research/presentations/2019/051319-wang-an-innovative-approach-to-scalable-semantic-embedding.pptx.thumb.1280.1280.png)
An Innovative Approach to Scalable Semantic Embedding
Pittsburgh, Pennsylvania, USA
Semantic search, in addition to keyword based search, is a desirable feature for many digital library systems. Even in the largely structured library data world, there is still a lot of tacit information locked in the free-text fields. Embedding words and texts in compact, semantically meaningful vector spaces allows for computable semantic similarity/relatedness which would make search more intelligent.
Topics: Semantic Embedding
![Subject Prediction Using Semantic Embedding](/content/dam/research/presentations/2018/oclc-dans-2018.pdf.thumb.1280.1280.png)
Subject Prediction Using Semantic Embedding
The Hague (Netherlands)
Koopman and Wang describe semantic embedding and their work on the Ariadne random projection algorithm that attempts to predict the right mix of subject headings.
Topics: Semantic Embedding