We’ve recently begun to make use of readers' click data to help personalize the recommendations they receive across the TrendMD network. The results so far are pretty exciting and offer a glimpse into how personalized recommendations can help readers discover content that’s actually interesting to them.
How TrendMD Uses Personalized Recommendations to Enhance Content Discovery
I’ll illustrate the effects of personalization with a recommended set of articles on Gut, a BMJ journal currently using the TrendMD widget.
Here is what the general audience saw approximately 3 weeks ago (widget below abstract):
In this example, the widget output was based solely on semantic relatedness. Three weeks ago, the widget’s average click-through rate (CTR) was 1.4%. (CTR = the number of times a click is made on links presented by the widget divided by the total impressions (the number of times links were seen by users)
The Impact of Collaborative Filtering and Semantic Relatedness on CTR
Fast forward to today, this is what the general audience now sees:
Notice that the recommended set of articles differs from what was seen 3 weeks ago. The widget’s output is now based on both semantic relatedness and collaborative filtering (i.e. people who read A also clicked on B). The current average CTR on the generalized audience widget is 2.1%.
Increasing Reader Engagement with TrendMD’s Personalized Widget
Now, here is what I currently see:
Again, notice that the recommended article here differs from what the general audience currently sees. The output of the widget here is based on a) semantic relatedness, b) collaborative filtering (i.e. people who read article A also clicked on article B), and c) personalization (based on click data collected by TrendMD — what I have clicked on in the past).
The best part is that I often read about literature related to the microbiome, and the recommendations presented by the widget reflect this. The current average CTR on the personalized audience widget (i.e. for the tens of thousands of users TrendMD has click data for) is 3.9%.
The 3 different CTRs illustrate the degree to which more data from different inputs can make recommendations more interesting and engaging to readers. Even more importantly, the larger the set of articles, the more impact collaborative filtering (i.e., people who read article A also clicked on article B) and personalization have on optimizing recommended link placements and increasing the CTR compared to semantic relatedness alone.
In addition to testing personalization, we’ve also been experimenting with adding ‘serendipity’ to the mix. We’ll be posting another set of results over the next little while. Stay tuned!
