What is frassle?
Frassle is a new kind of information system that leverages the unique tastes and opinions of many individuals to help you manage information overload. Frassle builds on important ideas from weblogs, aggregators, social networks, and search engines.
The Internet, Your Way
The dream behind frassle is a fully personalized view of the internet. Frassle doesn't make intelligent decisions for you, but it uses information you provide to learn, over time, what's important to you. It helps you to organize things you find on the net. Best of all, frassle doesn't need lots of care and feeding—if you write a blog or keep bookmarks in your web browser, you're already producing all the information frassle needs.
A Personalized Directory
When you visit Amazon.com, you see a store that has been adapted to your interests based on what you've purchased and rated in the past. Frassle generalizes systems like amazon and collaborative spam filtering to address content from all over the web, but also scales to more refined organizational systems—taxonomies you build any way you like. This means that frassle can help you locate material you'd call "funny" just as easily as it can help you ignore "spam" or narrow a search based on relevance to the "2004 Presidential Election".
Frassle allows you to:
- make posts (title, link, and commentary) to a blog
- discuss blog posts by making a comment
- any comment you make shows up on your own weblog. Frassle links the comment to the post that it replies to, and displays it along with the original post as well as in your weblog. That way, people who visit your blog can see what you're writing regardless of where you wrote it.
- build and maintain a category hierarchy
- create categories
- move and rename categories
- browse and search your posts by category
- categorize other people's posts in your directory, so you can find them later and share them with your blog's readers
- subscribe to other feeds using its RSS Aggregator
How does it work?
Many services, such as Yahoo! and The Open Directory Project, have built internet directories. But because their directories have to please everyone, they are generic, incomplete, and rarely useful. One reason for these problems is that it's very difficult to build meaningful consensus about what certain categories mean. Therefore, these directories tend to categorize well-known and well-understood objects, which are easy to find anyway, while utterly ignoring anything unique and cutting-edge.
Frassle doesn't bother trying to get groups of people to agree. In frassle, the important relationships are between pairs of people; and even then, we only expect partial agreement. What we may or may not agree about is the meaning of a category.
Determining the extent to which we agree is easy. Because every document on the internet has a URL, we can simply compare lists of URLs. If our lists for a pair of categories look alike, we have similar ideas of what those categories mean.
Consider for example this article from the New York Times. I would categorize it under "library science" and "search". If you categorize it under "librarians" and "reference", frassle can learn a correlation between my categories "library science" and "search" and your categories "librarians" and "reference". It might record information like this:
|the relevance of||your category||to my category||is|
Based on this information, the next time you post something under librarians, frassle can suggest it for my library science category with 35% confidence. If I accept that suggestion, the correlation increases—frassle learns that your idea of librarians is pretty similar to my idea of library science.
That's the basic power of frassle—it discovers agreement between people, and helps each of them utilize that agreement. By sharing our categorization information this way, we can each contribute to the other's directory, without any prior planning.
Frassle is currently being actively developed. It is a software application you can use over the web at frassle.rura.org. While the current alpha version is not complete and has many bugs, it does serve to demonstrate the major concepts. Shimon Rura is using it as his blog and continually enhancing it, and welcomes your suggestions to email@example.com.
Frassle is a labor of love. I'm building it because I enjoy the challenge and get a kick out of seeing other people use my software.
Acknowledgments: I'd like to thank Josh Ain for assisting with programming frassle, and Jessica Baumgart, Lisa Williams, Dean Landsman, and Critt Jarvis for offering useful feedback. I'd also like to thank all the users of frassle for helping me conduct my experiment. :)
Far in the Future: Scaling Up with Social Networks
Now, let's explore a features I envision for frassle in the future. I'm not sure if this is even feasible, but it's cool.
The Problem: The internet is a big place. Even if I'm hooked up to a few hundred other people, together we can only read and categorize a tiny sliver of the new information coming out every second.
In traditional human societies, ideas and pathogens spread in a web-like fashion, carrying their material first between friends and coworkers and eventually to many people completely unknown to the originator. Can we replicate this process in frassle?
Suppose for example that I have a category link with Alice. I like Alice's taste in humor, which means that her Humor category is highly correlated to my funny category. When Alice finds something new and decides it's humorous, frassle shows me that it's probably funny.
Now suppose Alice's friend Bob is also a funny guy. When Alice uses frassle, it shows stuff from his hilarious category as highly correlated to her Humor category. But I don't know anything about Bob, so unless Alice takes some particular item and categorizes it in her own Humor category, I won't see it. But in fact, if my taste in humor agrees with Alice's, and Alice's agrees with Bob's, then mine will agree with Bob's. Right?
This transitivity of agreement seems promising. I can't singlehandedly read much of what's on the internet, but if my friends and their friends all chip in, the pool of contributors is greatly increased. If we can compose category relationships between mutltiple people, each extra step links each taxonomy with dozens of others. Rather than just a single person or a group collaboratively developing internet directories, we have a whole social network.
This idea is conceptually quite simple but it remains to be seen whether it is truly useful, and whether it can be implemented efficiently.