How we rank
The AI magic that makes Truely tick
Learn about the nuts and bolts of our ranking mechanics
We gather reviews from reliable sources
First, we compile a list of products that belong to the same category. We then scour the web — using platform-specific web scrapers — to find as many reviews of these products as we can.
We repeat this process regularly, of course, to keep our data fresh.
We filter out suspicious reviews
A rule-based algorithm analyzes and evaluates every single sentence from the scraped reviews.
It then checks these against texts from reviews we know are reliable, and ruthlessly removes the non-conforming ones.
We cluster and summarise reviews
This is where BART enters. BART is the name of a state-of-the-art machine learning model that allows us to group reviews by topic and context. Here we determine so-called ‘review polarity’ — whether a review is generally ‘good’ or ‘bad.’
We score products
This is where RoBERTa comes in. RoBERTa is an algorithm that allows us to assign scores to positive and negative customer experiences around similar topics. We aggregate and average these scores to generate a final score per product or service.
We rank products on leaderboards
The final step is evaluating and comparing the highest-scoring services versus their time in market. We do this so that newer services have a fair chance to rank high on our leaderboards. Only if people love them, of course.
Grow your business with Truely Free Tools
Get an overview of users’ opinion around the internet and track your performance. Manage your product information and deals.