Our original SmartSearch worked very well. A user selected a couple of documents and we’d suggested closely related articles, based on clickstream data. This worked really well, when we had clickstream data, often we didn’t. So, that was a problem.
Our new system will work on a similar principle but will include citation data and semantic similarity data – which articles cite which other articles and which articles are semantically similar to other articles? This will allow for a very rich network of connected articles; in turn allowing users to easily find other articles of interest. Think about this, instead of scrolling through loads of articles the system should overcome much of that. Find 1 or 2 articles that interest you – our system finds the rest.
Testing has been really positive.
Initially, this will be just for Covid-19 data but when the new site is released (?September) we’ll look to incorporate it in to the main Trip.
Much of this work was inspired by speeding up evidence reviews. An important second component of this work is to allow rapid quantitative synthesis. This is being planned and we’re currently exploring funding opportunities to get this off the ground. The intention with this is to create SR quality reviews all done within a handful of hours. Longer than our fully-automated system but much nuanced and could be an important step to full automation.