By Hong Zhou, Senior Product Manager, Research & Development
Atypon’s overarching vision has always been the development of technologies that help researchers know more, do more, and achieve more. Artificial intelligence, or AI, is one path to those goals, and is the focus of much of our research and development (R&D) efforts.
What is AI?
AI is a broad set of technologies that enables a machine’s computational capabilities to “think” and “learn” like humans. There are many different types of AI, some of which are described in this blog post, and each of which solves different problems.
How is Atypon applying AI?
Atypon is currently focused on the branches of AI that make it possible to enhance search and discovery mechanisms. This kind of data-driven, machine-based decision making can personalize search and make content more easily and quickly discoverable, yielding a more productive and convenient research experience.
Critical to the success of AI-driven discovery is understanding a user’s intentions even before they begin a search. Here are a few examples of how Atypon is applying AI to do so.
Improving the accuracy of search
Entity extraction technologies recognize valuable information such as key concepts and phrases and then extract them from content automatically.
In healthcare, for example, PICO (which stands for the problem/patient, intervention, comparison, and outcome content in an article) helps clinicians answer healthcare-related questions via search. But nearly 70% of clinical questions return incorrect results or go unanswered: either the clinical information is unstructured or there is no way to distinguish which papers contain the answers to these questions. The existing workaround involves manually extracting the PICO sentences from papers—an error-prone, time-consuming process.
Atypon is employing entity extraction to automate identifying, extracting, and storing an article’s information in PICO categories in order to reduce time and costs even while improving accuracy.
Compare figures easily
Entity extraction technology also enables researchers to compare and interpret all of the figures associated with a single experiment—without requiring editors to segment compound figures manually. It even extracts caption information and assigns it to the corresponding figure. The newly automated process will not only reduce the cost of segmentation but also extract more valuable information so that researchers can search and compare their results more accurately and easily.
Create and implement topic tags faster
Entity recognition technology automates the classification and organization of information extracted from content and uses tags to improve its discoverability. The concept of tags is not new, and tags underlie many important Literatum website features, such as the creation of topic pages and content bundles. But tagging documents and maintaining the necessary taxonomies is painstaking work. Literatum’s existing auto-tagger automates this process but requires a large set of tagged examples to “learn” from. Entity recognition obviates this “training” set, making it much easier for publishers to implement and maintain tags.
Search complex figures accurately
The figures found in biomedical literature are an important part of research, education, and clinical decision making, and there are many different types of such figures—ultrasound images, 3D images, charts, graphs, and so on. The multitude of disparate figure types and lack of corresponding metadata makes retrieving and synthesizing the information in figures challenging at best. Atypon is developing a solution based on entity recognition that will automatically classify figures and pair them with the corresponding captions, or with the text in the article that describes them.
Finally, new search enhancement tactics will personalize discovery by understanding each user’s intentions in order to return the most relevant results. This enables standard keyword searches to deliver personalized search results that are ranked by relevance to each user. So different users performing the same search will not necessarily receive the same results.
Atypon’s R&D has always been driven in part by the needs of—and input from—our clients. The results of our R&D are continually integrated into Literatum. The AI-driven features discussed above will be implemented over the course of 2018.
And look for our new Research & Development web page, which will allow our customers to explore demos of our latest work and provide feedback early in the development process.