I've worked on academic deep learning and summarization for years.
Summarization is a foundational technology for the information age and a remedy for the attention economy.
Here's a🧵 for how we think and apply summarization at @YouSearchEngine
Why focus on summarization in search?
Time is finite and valuable.
You should be able to search less and do more and get things done quickly, not get sucked into engagement loops, fed a bunch of annoying ads, and junky SEO’d content.
Yet summarization is one of the hardest problems for AI and NLP.
One reason has to do with personalization — a good summary for one person might not be useful to another.
Summaries need to be as unbiased as possible.
There are many applications of summarization in search ->
Search for The Foo Fighters and you can see how complex summarization can be.
A good summary isn't just a short generated text, but instead shows all the different dimensions (quick facts, music, videos, social media) --> you.com/search?q=the%2…
Now search for “best Thai restaurant near me.”
You probably want sth visual and not a large body of text. You'd want to see the restaurant name, rating, review, if they have delivery or pickup, photos of dishes, and the location and contact info. you.com/search?q=best+…
How about searching for a recipe for banana bread.
You probably don’t want to sift through a 1,000-word SEO’d blog with a life story.
You want to see a preview of ingredients + instructions eg. via our side panels you.com/search?q=banan… (original link of course stays on top)
Or with product reviews, like “best headphones,” you probably want to see a summary of the reviews in one place to make the best decision.
Quickly look at pros, cons and specs in one place: you.com/search?q=best+…
Summarization can also boost productivity.
If you're a developer searching for code snippets, like “send SMS in Twilio,” you'll notice copy/paste buttons next to the snippets. A good summarizer knows code snippets are important.
Search less, code more. you.com/search?q=send+…
Summarization can be generative sometimes. So you can execute the intent communicated in your search query.
Eg. for a query like “CSS flexbox” - a good summary might be actual code that's generated using a large neural net language model
Those are just a few examples of applying summarization to search to help you do more. We have many more exciting summarization features to debut soon.
Follow me @RichardSocher and @YouSearchEngine as we build the future of search.