How Does Google Understand Passages vs. Subtopics

Tomek Rudzki asked if there were any intersections between the neural net understanding of Subtopics and Passages. Subtopics is a reference to Google’s ability to show a diversity of answers for broad search queries. This is how Google’s blog post explained subtopics: “We’ve applied neural nets to understand subtopics around an interest, which helps deliver a greater diversity of content when you search for something broad. As an example, if you search for “home exercise equipment,” we can now understand relevant subtopics, such as budget equipment, premium picks, or small space ideas, and show a wider range of content for you on the search results page. “ Splitt responded that these were two different things. Subtopics were a way of understanding queries, a way to give users “ways of zooming in to the content.” He said Subtopics is a way of understanding things and Passages is a ranking thing. Cindy Krum brought up how the official Google announcement talked about BERT in the context of Passages. Splitt indicated that BERT is a way of understanding content and that of course, before the Passages ranking can happen, Google needs to understand the content. He stressed that these are related but different things. Splitt looked up in the air and took a breath as he tried to work out in his mind how to explain this seemingly complex relationship. Splitt answered: “The problem here is there’s like lots of smaller systems talking to each other, so everything is blurry. There is no clear separation between the things. But for me, how I look at it, is when we talk about Passages as a thing you see in search results ranking, that requires that we understand the document, of course. We can’t say like… this Passage is about this thing if we don’t understand that. But then there’s this other understanding which is more like, how does this relate to things and how does this get disambiguated from other similar topics and other similar entities? Extracting these conce