Hearing beyond what users say
Sparrho 4.0: Keyword filtering is back
In my previous posts, I explained how Sparrho was not yet solving the 95% solution. We had created a portal that presented a cross-section of the latest publications in any selected fields, but many of our users were also spending lots of time checking the latest from their favourite journals. We had always planned overhaul our content sourcing tech to overcome some legacy code issues, but we realised at the same time that we could also provide a new ‘aggregator’-style view for those who needed it.
A straightforward recommendation experience
This new journal aggregator tool was rolled out on 21st March, whilst we set to work overhauling the tech behind keyword filtering and recommendations — we of course knew that our users don’t just want to filter by journals. In the process, we decided to use what we’d learnt over the past few years to streamline our channels and discovery process.
We are no longer asking for direct indication of what is relevant or irrelevant, but our AI is still building a picture of your research interests to improve the ‘search’ / ‘keyword filtering’ algorithms of your channel in the background.
Understanding our users’ motivations is crucial for our AI to make high-quality recommendations, and our research had shown that the reasons for marking items ‘relevant / irrelevant’ varied wildly.
Tapping into meaningful human curation
We also know that human-to-human recommendation is actually the best kind there is — so this is why we’ve introduced pinboards, shared areas to allow our users to recommend articles to each other.
For years now, we’ve heard how researchers find that staying up to date with the latest research is a serious problem. However, it’s important to hear beyond what your users are telling you. ‘Staying up to date’ is itself a treatment to an underlying issue — we needed to work out why researchers were looking for new research in the first place.
Imagine we are instead an online shopping platform — a customer ‘Sam’ recently bought black socks. Do we, recommend more black socks? Or blue socks? Or leather shoes? In this example Sam is going to a wedding — leather shoes are what he needs. Only by understanding Sam’s motivations will we be able to make a truly high quality recommendation.
Motivation informs recommendations
Our user experience (UX) research allowed us to crystalise those reasons for ‘staying up to date’ into 2 key themes:
- to get new ideas; and
- know what their friends and enemies are up to.
It seems simplistic but nearly all research discovery can be boiled down to these 2 statements. The first requires a broader, less strict view into the latest content, while the second requires a much more restrictive one.
This means that there is no one size fits all solution and we needed to reassess how we could help our users get the information they needed, bearing in mind the different reasons they were looking for it.
Helping you find inspiration
The current keyword filtering algorithm in your channels, like old Sparrho (3.0), is focussed on reason 1 — helping you get new ideas.
There are different reasons one might ‘pin’ an article but this is on the whole a positive action, and we know we should show more content like the pinned item. We currently don’t have the opposite negative action and leave this reductive filtering control entirely down to the user.
We found that on the whole, when one is looking for new ideas they are happier to see more articles, even if some are slightly less relevant, and would be annoyed if things that were relevant were not shown. This is why we’ve chosen to focus our current channel algorithms on showing more, rather than excluding more. There is more going on than a simple ‘OR’ query, here your keywords are taken as a set that together define a topic, and we’ll go into more detail about what’s going on in future blog posts.
Now there is plenty of room to both increase control and understanding of what is relevant/irrelevant to a user or channel but I hope this helps explain the choices we’ve made so far.
Refining stringent searches
This is just the beginning — now we have substantially improved our tech we can move quickly and will work to continually improve the relevance of your channels based on what we learn from your reading habits. Having just journal filters set provides a restricted view on our content but we have plans to extend the control you have over what you want to see in your channels further. Watch this space!
