For the past while, we've been developing advances in technology that have allowed us to analyse and incorporate important data from reviews on other websites. From a base of less than 2,000 native reviews, we just launched yesterday a massive increase to effectively 49,000 reviews. What this means, from your perspective as a user of this site, is that you now have a ton of information from which to draw from, in your search for the right supplements to match your goals.
By using natural language processing techniques, we can now analyse the texts of reviews and determine with reasonable accuracy, which effects the users experienced from their use of the supplements. This includes beneficial effects, such as fat loss, strength, or taste, as well as side effects, such as headaches, crash, and nausea. So far, out of the 42 effects we have on this site, we've covered over half of them to include in the extraction process, and out of the 39 side effects we list, we've included almost half of them. We chose these effects on the basis of the popularity of the searches for them over the past few years on this site and its predecessor, as well as having knowledge of the fact that people are most interested in the side effects of a supplement, as much as its main purported effects.
All of this new data makes the Supplement Explorer orders of magnitude more useful than it was before. You can find, with a higher degree of confidence, supplements that fit in with your goals, and see how many people rated them for those goals. Since we now have a large number of reviews, we have also changed the way the Explorer tool sorts the results. The sorting formula that is used now takes into account the number of ratings as well as the average rating, so the results at the top are more natural and intuitive.
Some of the processing techniques that are currently applied are based on machine learning. Because the current state of research in this area falls short of achieving 100% accuracy, some of the results are incorrect. The good thing is that we've got a widget for every analysed review, that allows a logged-in user to help provide data to the system that will be used in its next training run, where the accuracy should improve. Here is what the widget looks like for the beneficial effects (goals):
This is a very exciting new avenue of growth for Supplementarium, and we're still in the very early stages. We'll soon be able to expand to over 100,000 reviews, and from then on, we'll be analysing and linking to reviews from additional sites. This website no longer consists of just a couple of thousand of reviews in its native format, but is becoming a dynamic review aggregator. Of course, we are still welcoming reviews entered directly into this site, since in our own format, the system doesn't need to try to understand and analyse what you as the reviewer meant, and the quality is generally higher.
We hope that you find this site overall much more useful now, and can spread the word about it to your friends and acquaintances.
