Google’s latest algorithm and predictive video search.

BERT is the latest, important, upgrade to Google’s search algorithm. It is bringing machine learning algorithm’s understanding of content and context much closer to ours with the result that searches will be better, closer to what we are actually asking for and crucially, much more dependant on good content, both written and video.

BERT brings the ability to look forward and backwards within a search query, including at those joining words, to place the words in context much better and importantly to understand what is the key focus of the specific query.

BERT is already live on searches in the US and in the UK shortly, but VideoBERT appears much more of a future project. The objective is to model text and video and learn their relationships.

VideoBERT extends the algorithm in a different way. It operates as an AI based system, learning key characteristics of activities. The benefits will be to provide captions, classify actions and even predict video to show in future to better answer a search query.

The VideoBERT model was prepared using over 1 million YouTube clips of cooking, gardening and vehicle repair, both audio and video. As an example, now given a few frames showing cake making ingredients, VideoBERT can work out that this may be to do with a cupcake or brownie dependent on the ingredients and put forward appropriate recipes, captions or clips.

As it develops and becomes part of daily search, the need to optimise marketing video properly will only increase as the huge volume of content on the web will compete more and more for attention.

And our tip, if Google is testing extensively on food preparation, gardening and vehicle maintenance, and that is a business area you’re in, it’s a pretty good bet those are areas where starting to provide optimised video on and driving visitors to your site has to be a good idea.