Collaborative Recommenders Everywhere
MOTIVATION
People love to get recommendations. The world is too big not to use a little guidance, especially from those you trust. Moreover, people also love to recommend things to others, especially if you have a strong opinion. It's expressing yourself to a receptive audience.
Collaborative recommender systems enable this natural activity, but put some structure around it to wring more value out of the data people express. Sites like MovieLens allow people to express their opinions about things as ratings (say from 1 to 5), then use those opinions to get things recommended that they don't know about yet. One way to recommend is by a sort of 'automated word of mouth': I like "The Matrix", you like "The Matrix", you like "Reservoir Dogs" and I haven't seen it, so recommend it to me. That basic concept is extended using computers to sum across many users and many items. Other sites, like Amazon, allow users to create lists of related things (Listmania!), then display related lists whenever it seems appropriate.
Currently, community sites with collaborative recommender systems are uncommon. They are only present at web sites that have unusual expertise or large amounts of money. Even those with expertise but little money eventually do not have the resources or inclination to maintain their recommenders. The surviving recommenders serve a narrow set of content, mostly those things which can be bought.
Collaborative recommenders should be as common as web servers. Just as many different people and organizations have content they publish for a variety of reasons from ideological to commercial, so should they be able to allow a member of their community to easily pour through that content looking for the right stuff. Currently, free software such as Apache can publish the content, a free service such as Google allows keyword search of that content, but there exists no free software or service to allow collaborative recommendation of that content.
If collaborative recommenders were everywhere, people or organizations who were expert in either content (domain experts) or community (community leaders) could provide the energy to maintain content (items) suitable for recommendation, as well as the energy to build a community around the content. The community could then provide both content and opinions to help others find the right stuff.
In such a world, many recommender-powered communities would be lame, just as many web sites are lame. However, some would be successful, and around content much broader than the present day. Imagine recommending
- restaurants
- recipes
- charities
- software
- university classes
- soda (or pop!)
- beer
- cosmetics
- vacation spots
- hiking trails
- and so on ...
Even currently successful recommender-served content such as books, movies, or music might benefit from a different, perhaps non-commercially oriented, community site with a collaborative recommender. Even multiple sites centered around the same content could be useful if the communities are themselves different, and have different 'mainstream' opinions for their community.
STRATEGIES
Suppose that you believe that collaborative recommenders should be everywhere. What is the best way to accomplish the goal? One possibility is to find existing popular community software, and inject a recommender into it. For example, WikiLens.
There are other strategies as well, which some folk in GroupLens are thinking about. For example, what if you knew
- URL => book mapping
- What people thought of a URL (either explicitly, or by linking)
Could that be used to get recommendations?
Mining blogs is a possible source of information. See for example http://allconsuming.net. With some structure, especially a way to identify things that is broader than 'Amazon id', could that produce recommenders everywhere?