I’m just back from a meeting of KConnect, an EU funded consortium bringing together academia, SMEs and care providers with a view to improving health care. This is the 3rd meeting in two months and thankfully this one was in London (the others being in Luxembourg and Vienna).
There was lots of exciting developments and the following apply to Trip:
Answer engine. The answer engine is a system designed to infer questions from a users search terms and pull through the best answer (as well as the usual search results). We have a proof of concept model, which is great. However, there are a few issues with it. A few conversations with consortia members and it looks like that will be fixed fairly easily. With the prototype I estimate we got between 50-60% right and with the new tweaks, when they’re rolled should see that rise to over 80%.
The answer engine relies on being able to match search queries to articles. That needs us to accurately reduce document titles down to medical concepts. We should now be able to do this more accurately, more quickly and it’ll give us lots more flexibility with how we display the results.
Search log analysis. I typically refer to this as clickstream data, but search log analysis appears the favoured term. A new feature that we’ll be rolling out relates to search refinement and this will manifest itself in two ways:
- During search. We currently have a drop-down list of suggestions, supplied via PubMed. However, it’s very ‘dry’ and never feels particularly ‘human’. By using real-life data from Trip we’re confident that our system will be more real for our users.
- Post-search. Users typically only search using 1-2 terms. This system can be used to suggest additional terms to bring the user closer to the results they need.
Below is a list of query suggestions based on a search for hypertension:
- 976 hypertension pulmonary
- 715 hypertension arterial
- 694 hypertension treatment
- 572 hypertension pregnancy
- 439 hypertension diabetes
- 401 hypertension stroke
- 248 hypertension exercise
So, if you’ve search for hypertension we might show the above and say, something like ‘are you interested in hypertension pulmonary, hypertension arterial etc. A user clicking on a search phrase will see the search automatically undertaken.
This should be rolled out shortly and we’ll test it to ensure it actually improves things!
Algorithm changes. Already planned from my visit to Vienna but further refinement of the specification which will see us trying multiple tweaks – in isolation and in combination – to try to ensure an optimal new search algorithm.
There are a few other bits and bobs that significantly affect Trip’s future development – but these are outside of KConnect and therefore can wait for another post. But, rest assured, these changes are all ambitious and doable!