As an information professional I feel duty bound to continually improve the service I deliver and as a manager I feel it is my responsibility to drive change in my services instead of waiting for change to happen to us. Feeling buoyed by our success in launching our browser extension Lean Library in 2019 we began to explore implementing some form of Artificial Intelligence (AI) into our service.
When researching AI options we backed away from a customer facing search tool as this technology currently lacks the sophistication to handle more than two search terms. Also within the service we already offer our users a variety of access points into the evidence base: HDAS, Discovery tool, our browser extension. So, we had to ask ourselves whether adding another search tool would benefit our users or overwhelm them?
We began to think more deeply about how AI could benefit our service. The primary focus of our team is the creation of evidence reviews: a rapid synthesised literature search available to anyone in the organisation. They are increasingly popular and while our Trust has doubled in size over the last few years the Evidence Service staff numbers have remained static. This growing tension between demand and supply led us to explore whether there was anyway AI could help us in carrying out these searches; this led us to Iris.ai.
We are the first NHS organisation to use Iris, a “young” AI with a primary function of Chemistry R&D although throughout the pandemic it has been used for COVID-19 research. The software is currently made up of two elements called Explore and Focus, Focus mode is essentially a way of refining your search results so below I will focus on the Explore mode; the search function of Iris.
We bought Iris in an off-the-shelf format; it has read and continues to read all Open Access papers (there is a more expensive option which allows it to read all your online holdings). In reading papers Iris can understand keywords, concepts, context and relationships which it can then map against all the other papers it has read. This theoretically changes the nature of searching as the AI will be able to identify relevant papers that might not contain the keywords used in a more traditional search.
The first thing to note when using Iris is that it uses natural language processing (NLP). Essentially, the software wants you to type your question in a normal, fluent format. This is a seismic change for librarians used to honing search questions to the bare number of keywords; Iris wants you to enter between 300 to 500 words. When inputting your question Iris is identifying keywords and context that it will match in the information it has read. A library user isn’t going to deliver their search question in this format so the librarian either needs a strong understanding of the subject area or a dialogue with the user to get the context that IRIS needs to function. On submitting a search question IRIS will create a fingerprint of your results comprised of concepts it has identified.
Fig 1. Iris concept map
At this stage you can download all results or click into concept cells and see the papers Iris has identified for you, clicking into a paper gives you the option for Iris to search for related papers. In this image the 76% is the relevancy score Iris has attached to a specific paper.
Fig 2. Sample result
At this point you can begin to remove unhelpful terms, promote more helpful terms and apply limits such as date or relevancy percentage. Applying any filter creates a modified concept map.
Fig 3 search limiters
Fig 4 hierarchy screen
We negotiated our deal with Iris at the start of 2020 with a start date of 1st April; so as with many things our use has been affected by COVID-19 as our team priorities shifted and our opportunity for collaborative learning decreased. In this current state the software is a very specialised tool, in no way intended for your library user or student and even for information professionals it presents a steep learning curve asking us to reformulate questions in a way that might feel unnatural to us. We use Iris concurrently with HDAS; so in this sense it is not saving us research time however it does add depth to our searches finding relevant papers that are not returned through standard methods.
As I mentioned earlier Iris is still “young” at version 6.0 and will continue to develop and grow, with an exciting future already outlined. Essentially we are asking not what Iris can do for us today but what we may do together in the future. Moreover investing in Iris, and our other technologies has not only directly benefitted the service but has also help change the perception and profile of the evidence service in the wider organisation, in this sense embodying change, progress and technology can never be a wasted investment.
Evidence Services Lead
Mersey Care NHS Foundation Trust