Today we have upgraded the API for our NLP analysis algorithm so that it works significantly faster and picks up more topics. The API is the part of our system that integrates the web interface with our Knowledge Base. When you update your associations or topics, this creates a ripple in the space-time continuum (OK… not really… but it does change your link graph, which we need to recalculate). The most noticeable effect will be the little “working” timer on the top left of your dashboard, but there is a little more to this.
Improvements to our Understanding of Topics
Part of the upgrade has been a better understanding of topics in particular verticles. As our customers use the system, it creates flags for us that help us understand where we need to improve. Periodically, we can then update the system to better understand topics cited on web pages.
inLinks will generally always find more topics in content that Google’s NLP API. To be fair, that is partly because if you make a call and the Google API finds something we don’t, our algorithm learns, but in general, we almost always find significantly more relevant topics in a good item of content. In this comparison of the BBC’s News page, we found 147 topics compared to Google’s 60. This is a part of our secret sauce but shows just how important schema and inLinks will be in 2020 in helping Google better understand the content.
Occasionally this may mean we pick up topics that are not so relevant and in that case, we try to learn from this as well.
Dig into the Topics
Last week I also found I was unable to dig into Topics. If you go to the Graph tab, now, you can easily drill into all the topics that we have found.
This is helpful, because it is not always clear what a topic is about just from its name.
Silo Creation Bug Fixed
Sometimes you have a group of pages, rather than a single page, about a given topic. In this case, you may wish to create a “silo”. You still need to decide on a headline page, but the system can then create a hierarchy for the content within the silo. Before Christmas, creating silos did not work, but this was also fixed recently. Silos are an advanced feature, so we will create a separate blog post about Siloing content in due course.