Content Recommendations with Medit
At Medit, we want to help you discover the content that matters to you quickly. This is no trivial task. Every day, we add new medical content from thousands of sources to our database – journal articles, blog posts, videos, news articles and podcasts. It is clear that nobody has the time to read all of those!
Just like our content, the interests and preferences of our users are very diverse and can change over time as well. People are complicated and it’s hard to know what each user likes to read at any given time.
So, how are we cutting through the noise to get our users the content that is relevant to them?
Building a ‘recommendation engine’
At the heart of Medit is our content recommendation engine. This engine uses various data and machine learning techniques to filter and rank new content for each individual user, taking into account the user’s medical specialty, profession and followed interests, as well as the different features of each content item. These features might be the recency of an article, and whether it is tagged for topics you are interested in. We also use natural language processing to correlate an article’s actual content to its popularity so that the content recommended isn’t just new and relevant to you, but also something your peers have been engaging with.
Ranking is also impacted by journal impact factors or website page rank as well as the collective user behaviour, e.g. a blog post that is viewed by a lot of users will increase in rank. Furthermore, we also take into account the detected preferences of users that are similar to you…after all, the content that GPs may like to read might be completely different from the type of content neurologists prefer.
The content recommender is used in Medit’s ‘Explore’ section. It is also used to create your daily ‘Just for You’ list on your Medit home screen.
Filtering through the noise
In addition to providing smart recommendations, we also provide filter ‘widgets’ on the Medit Home screen. These showcase trending content suggested for you from different angles, e.g. you can see the content that has received the most user engagement (‘Trending on Medit’) during the last day or week, irrespective of your usual preferences.
These widgets are currently generated once a day.
Adding a human touch
As much as we love our recommendation algorithms, we also regularly ‘hand-pick’ recommendations for you. Our editorial team is constantly reviewing publications, websites and social media. When we come across content that we think is really worth reading, but might not be trending yet, we share it with you.
These posts are usually marked as ‘Recommended Reads’ in your Medit Home screen.
Constant testing & adjusting
We constantly retrain and tweak our recommender models using new user and content data. Recommendations will get better the more we can learn from past behaviour patterns.
We also deploy different recommender models alongside each other in our app so that we can monitor and compare their performance. That way we can see which recommendations our users like best. Of course, it’s never done – test, tweak, rinse & repeat!
We hope that this blend of recommendation and filtering solutions is useful and helps you better explore the latest medical news, cases, and research. It will get better with every use.