Frequently Asked Questions
Content relevancy is a measurement of how relevant a piece of content is to a particular keyword, beyond simply determining if that keyword occurs in the content. An article on NASA is not particularly relevant to the keyword "glasses", even though the article might mention that word in describing part of a satellite.
How Does nTopic Work?
nTopic takes keywords you provide, builds a "perfectly relevant" document, and then compares your content to that document. It then compares your relevancy against the probability that any random 2 documents from our cohort will have greater relevancy to one another than the ideal document to yours. A score of 99% means that 99/100 times your document would be more relevant to the ideal keyword document than any two random documents would be relevant to one another.
What is a Good Score?
You should aim to score 99% or greater whenever possible. This indicates strong confidence that your document is relevant.
Do Search Engines Use Topical Relevancy?
We have both correlative evidence to suggest topic modeling is used by search engines and causative evidence to suggest that applying content recommendations from nTopic can improve rankings and traffic. Does this mean that search engines definitely use LDA or other topic modeling? Absolutely not. We do not pretend to know Google's algorithm. However, we do know that improving your content relevancy using the nTopic measurement does tend to increase rankings.
What is the Difference between the Free and Paid API?
The free API allows you to get unlimited nTopic scores using a random sample of the content you provide. While statistically accurate, this isn't a complete analysis of your content and thus is not as precise. Small changes to your text might not yield changes to your resulting score. The paid API analyzes 100% of your content and also makes recommendations for keywords you should add and remove in order to increase your content relevancy.
Unlike most data-APIs available on the web, nTopic must make unique calculations on-the-fly every time you call it. Unfortunately, it is not possible to speed up the process much beyond 6 seconds.