4 ms·

I've thought about number two as well. How about a combination of caching all pages you read and how long you are spending on each page (and maybe if you scrol

I've thought about number two as well. How about a combination of caching all pages you read and how long you are spending on each page (and maybe if you scrolled as well)? The idea is to save enough data that you can have an algorithm that will automatically differentiate between stuff you closed within a couple of seconds and didn't really read and stuff that interested you and you actually read through. And the reason for doing this automatically is it seems very difficult to have the discipline take an action every time you see a page that you may want to save so that you can search through again (such as clicking a box for every such page). Whatever mechanism it is, it needs to account for stuff read through readers and over multiple computers/devices too. I don't want to miss a search result just because I was reading it on an iPhone or in my Google Reader. Or instead of caching locally, a search interface that would accept a long list of urls to look through.

ntoshev 17y ago
I have thought about it as well. The signals that could determine relevancy in your browser history are likely different form the ones that determine it for web search, and measuring engagement with a page is a good example (it would help web search as well, but it would be harder to measure). Other things you'd take into account are: * continuity in time - reading several pages on similar topics one after the other. * analysis of the subgraph of the web graph that you visited * you'd be more aggressive with query expansion because you work with much smaller corpus, e.g. you'd want to take hypernyms and hyponyms into account
- jimmybot 17y ago
  Ah yes, engagement is the word. Interesting. I'd imagine the query expansion should also work across the author dimension--include and rank matches from the same blog higher, even if according to my history, I haven't read that post.

ntoshev 17y ago

I have thought about it as well. The signals that could determine relevancy in your browser history are likely different form the ones that determine it for web search, and measuring engagement with a page is a good example (it would help web search as well, but it would be harder to measure). Other things you'd take into account are: * continuity in time - reading several pages on similar topics one after the other. * analysis of the subgraph of the web graph that you visited * you'd be more aggressive with query expansion because you work with much smaller corpus, e.g. you'd want to take hypernyms and hyponyms into account

jimmybot 17y ago
Ah yes, engagement is the word. Interesting. I'd imagine the query expansion should also work across the author dimension--include and rank matches from the same blog higher, even if according to my history, I haven't read that post.