I'll take the opportunity to post about my recently started (2 weeks now) project on "design dependent search" (idea 16).
The idea I'm trying to leverage is that typing long search queries on mobile devices is super awkward, but the touch screens make clicking much easier and more fun. My site displays similar search results geographically near each other on the screen, allowing you to make your searches more focused just by clicking on certain areas. Right now, it only works on blogs with RSS feeds due to my limited internet crawling resources. I hope to add media (picture,movie, and maybe music) search by the end of the summer. I don't see this necessarily working as a Google replacement though.
Interesting. I had a play with it and I think there is some potential there. Its somewhat similar to http://www.chunews.com/ which I wrote some time ago in that you can follow the links between similar articles and paged. I like your implementation though.
How are you working out whats related? Links or some deeper textual analysis?
Thanks for the heads up on chunews. I was having some trouble finding similar projects to judge where I should focus development. How many users did you get?
As for linkedness, I'm going to do a detailed blog post about it sometime, but here it is in short:
Use the HITS algorithm to get hub and authority scores for each domain. Pages are related only if websites with higher scores link to both of them. This makes spam much less of a problem (I think!), and will make for some interesting search algorithms that I haven't quite implemented yet.
I was more my attempt at recreating TailRank when it closed down (since I assumed that it couldn't be that difficult to write). It just seems to overlap what you have there quite a bit.
Not many users. It tends to get a lot of SEO traffic though (usually breaking news stories funnel through it pretty quickly and you see a large spike) with anywhere from 50 to 500 uniques a day.
Ah interesting. Chunews does statistical word ranking on each article it pulls down to group similar stories together. From memory it uses links as well but in reality since it checks exact links that never affects this.
One of these days I may iterate on it but for the moment it shows me whats happening in the world at any point which is fine for me.
I find it quite hard to quickly visually scan the results. And I'm not getting very good results for the queries I tried. Are there any specific type of queries this is good for?
Considering it's a search engine, you should probably not hide the search box behind an extra click.
I called it a search engine, but it is not supposed to be a google replacement and do general queries. It only finds blogs similar to ones you're already interested in. At this point it's more of a fancy RSS reader with a twist.
I've been wondering whether or not it's best to hide the search box. I decided to hide it because I don't really want people to use it. As you've pointed out, it's not really that good. I'd rather people find blogs similar to ones already on the screen by clicking. That's where the strength of my display comes in, I think.
I wanted to experiment in new search interfaces, but I thought that taking on google directly would simply be too hard. That's why I didn't jump right in to general purpose search.
The idea I'm trying to leverage is that typing long search queries on mobile devices is super awkward, but the touch screens make clicking much easier and more fun. My site displays similar search results geographically near each other on the screen, allowing you to make your searches more focused just by clicking on certain areas. Right now, it only works on blogs with RSS feeds due to my limited internet crawling resources. I hope to add media (picture,movie, and maybe music) search by the end of the summer. I don't see this necessarily working as a Google replacement though.
You can check out the prototype at http://blogumbus.com and visit the developer blog at http://dev.blogumbus.com.