The Haystack project out of MIT was mentioned on Slashdot today. It’s a new kind of information manager that seems to put all your information, contacts, notes, email, etc, in one place, with new interfaces for viewing it, including multiple overlapping category hierarchies. The Slashdot article mentions the use of AI techniques for indexing, but I can’t find much of it on their website. Still, the idea of a smart PIM is an idea that I’ve had in the back of my mind for a while. It seems like there’s a lot of room for current AI techniques to get leverage in this area.
When I first thought of this idea, I called it
Personal Google. Basically, you could put any kind of data you want into it. Email, calendar, contacts, web bookmarks, browser cache, RSS feeds, PDF and PostScript papers, maybe even images (though they’re a little harder to deal with), and it provides intelligent free-form searching capability, plus a lot of intelligent browsing. It would provide multiple interfaces to the data, through a modular architecture. Here are some ideas for interfaces:
- All the standard PIM interfaces.
- Basic keyword search.
find items like thison any item.
- A hierarchical taxonomy of items, created automatically, through something like COBWEB, maybe augmented by semi-supervised clustering methods, where you can specify that some items do or don’t belong in the same cluster, like Sugato is working on.
- Indexing or lookup via factored
conceptsthat can be combined in an item, via latent semantic indexing or latent Dirichlet allocation.
- Search via topographical navigation between document clusters (or topic clusters) using a self-organizing map. Could be cool for images.I think I’d want it to run as a server, so I could get to it from anywhere, and add bookmarks, papers, etc to the database from anywhere. I’d use it as a web homepage. It’s possible that Haystack does some of these things. I haven’t had a chance to download it yet, because they don’t have a Mac version. I’ll get the linux version at some point.