Search is invaluable. The volume of instantly available content would be impossible to make sense of without it. How would we track news or answer questions or find beloved cat pictures?

But, it’s important to realize not all search is the same. There are different levels to search, each of which is important for different tasks.

Here at Evid Science, we tend to view search as an inverted triangle, where the deeper you go, the deeper the results become, but, the more specific the requirements are as well. That is, each layer in the search triangle answers a very different type of question.

Searching by Keywords (and a bit more)

At the top level of the triangle is “general” search. These search capabilities are the broadest, and they give the most coverage (generally). However, they will only point you to the results you want, rather than give you all of the answers directly.

For instance, to search for papers on a disease, you enter the name the disease and you get back all of the papers that mention it (and maybe a few that mention its synonyms). But then it’s up to you to read. So, you get a broad and comprehensive answer (“this is how many there are, and here is a list”), but the answers are still buried inside.

Pros: Broad, Fast, Familiar, General (e.g., search on medical papers can return cat pictures too)

Cons: Pointers to answers, rather than the answers themselves. It’s still up to you to do the onerous work of turning those pointers into the answers you want.

Examples: Google Scholar (which we’ve written about before), PubMed.

Searching with Word Meaning

The next level deeper in the triangle is one level deeper in terms of search capabilities. We call these Categorical searches. These search capabilities have a deeper understanding of terms (the semantics), and so these search capabilities understand how various words relate to one another. For instance, you can search for a type of disease and get presented a list of related topics, which are specific variants of that disease or related diseases. This type of search is particularly helpful because it lets you start more broadly than general search, but then drill down to what you care about based on the related suggestions. For instance, you might start by searching for “inflammatory bowel disease,” and the system will recommend “Crohn’s disease” as a related topic, which may be what you were actually looking for in the first place. So, it leverages both the semantics to answer broader questions, and helps you discover what you really wanted. They will also start to give you some of the answers, perhaps highlighting related phrases, etc. But the onus is still on you, as the searcher, to read the content and find your answers.

Pros: Understands search terms more deeply, so it can find you related content and suggest related search terms. This helps you find more content and discover content you might not have thought about ahead of time.

Cons: Still requires human to read the content

Examples: Semantic Scholar (a favorite, brought to you by the good folks at the Allen Institute for Artificial Intelligence!), IRIS.AI

Results, Rather than Search Results

At the pointy part of the triangle are cases where search is really about finding results. In contrast to broad questions, such as “what studies looked at budesonide,” these “answer engines” are purpose built for specific questions. You might ask – “How was the PCNA human antibody used” and BenchSci will give you answers, complete with figures; or you might ask, “What about the BRAF gene” or “What are the adverse event reports for this drug?” and Genomenon and Advera Health can provide those answers, respectively.

At the very tip of the pointy end is Evid Science, where you can ask very particular outcomes questions, such as, “How does budesonide compare in remission for Crohn’s disease?” and since our AI “reads” the literature it will surface the actual patient-level numeric results to answer your question.

In all of these pointy-end cases, the key difference is that we’ve moved from “reference search” to “results” – directly. These services give you deep answers, but to specific questions. It would be hard for a general search strategy to do this, since its focus is breadth and identifying the sources of information. Many of these services, Evid included, also support discovery of new information, since while you can search for answers, you could just as easily browse and explore the results. So, it’s like the categorical search, but one level deeper.

Pros: Rather than just references, you can get the actual results. This can dramatically speed up your “time-to-data” and unlocks new, data driven approaches to tasks such as understanding how all of the drugs compare within a disease landscape.

Cons: Uncovering deeper answers requires deeper technology – and anything new brings with it a learning curve. This learning curve is technical (new techniques needs to be perfected), but also behavioral. So while early adopters can find immediate benefit, there is a behavioral change that must be realized as well, for general adoption.

Examples: BenchSci, Genonmenon, Advera Health AnalyticsEvid Science

The bottom line is that the type of search depends on the type of question you are trying to answer. The key is to understand that not all search capabilities are the same – sometimes search is search is search, but sometimes it’s not.

If you would like to know more about Evid Science and how we turn the literature into results that can be surfaced and analyzed, please drop us a line!