« Definition: update | Main | The end of objectivity »

February 02, 2005

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341bfb6353ef00d834587ced69e2

Listed below are links to weblogs that reference Why Social Software Makes for Poor Recommendations:

» Why Social Software Makes for Poor Recommendations from del.icio.us WebCites
This post is off base. Social networks are not just friends networks but authority networks, knowledge networks, etc. You do not have to be friends to be in a network, just connected.... [Read More]

» Collaborative Filters and Japanese Society from selbr.de
One of the question of the emerging long tail economy: Who knows the coolest stuff? Answer 1: My friends (colleagues, co-workers, fellow experts). Answer 2: A scattered group of people, connected only by their coincidental common taste in one particula... [Read More]

» Spiderworks from Tenebris
Lulu storefront of the day for Friday, February 4, 2005: Spiderworks--"quality eBooks" from the likes of Danny Goodman, author of Danny Goodman's AppleScript Handbook (Mac OS X Edition). [Read More]

» Output du weekend sur del.ico.us from A Frog in the Valley
Je me demandais la semaine dernière si certaine personnes qui lisaient mon carnet n'étaient pas abonnés au fil web de mes signets partagés sur del.ico.us... Je me disait que je devrais à l'occasion poster ici aussi...... [Read More]

» MORE ON SOCIAL NETWORKS AND RECOMMENDATIONS from Knowledge Problem
Michael Giberson Chris Anderson at the Long Tail says there is a problem with using social networks to generate recommendations: The problem with social software as a recommendation network has its roots in the problem of social software itself. "Frien... [Read More]

» Why Social Software Makes for Poor Recommendations from UserDriven
"When it comes to recommendations, friendship is overrrated." (The Long Tail) [Read More]

» Re: Social Search from tribe.net: longtail.typepad.com
here's an interesting one on social-filtering and social-search from Chris An... [Read More]

» Recommendation Systems and Social Networking from A Copyfighter's Musings
Digital Music News reports on Pandora's deal with Friendster. [Read More]

» Startseite - Musicload from MP3 Download
Die Musikdownloadplattform von T-Online mit Songs aus allen Genres. Titel sind im geschtzten Windows... [Read More]

Comments

Christopher Carfi

I agree with the last paragraph (excepting perhaps the final sentence). Collaborative filtering is the killer app for navigating the LT.

On the other hand, I'm not sure I agree with the wholesale dismissal of "friends" as recommenders. In the same way that you so eloquently put that we all have orthogonal recommendation networks, we also have orthogonal friend networks, don't we? The group of friends with whom you might go to a concert could, quite easily, have zero overlap with the group of friends you see at the kids' soccer games on the weekends. But yet, they are all equal in their "friendness." Again, to paraphrase your statement, the only common linkage between these friend networks is you.

To your point, all friends will not provide equal recommendations in all domains. But, I would argue, some friends will provide superior recommendations in some domains. And in these cases, as discussed here, these long tail discoveries can really strengthen friendship and community linkages.

Stephen Johnston

Don't forget emotions. Certain cheesy rock ballards, weird-smelling shampoos or unreliable British sports cars may become magical if your partner dotes on them. And representing emotions digitally is even harder than just reporting facts about sales or processor speed. A tough one indeed!

Toby Padilla

Very insightful. I've definetly found this to be the case on Musicmobs. Up to this point profiles have been based strictly on musical preference and not on personal. The relationship between users is labled "favorite users" -> "fans". I like this much better than "friend".

I agree with you that the generic social networks like Friendster are pretty bad for recommendations (unless they are going to focus on making "friend" recommencations). I think you'll see many niche sites pick up the ball in more specifc domains.

Kevin Laws

More support for your observation: at Epinions, we initially thought that social networks would be the source of recommendations. We quickly changed that to using the social networks to weed out the poor recommendations (if you're trusted by nobody or trusted members rate the review poorly, you don't even make the first cut). We used other methods (expertise in a specific area as determined by the community as a whole, for example) to weight the actual ratings.

GuruChick

There is a collaborative-filtering research project at the University of Minnesota with a great tool called MovieLens. It gives you the best of both worlds. Its main feature is to do collaborative filtering on movie preferences, so you can get suggestions based on the opinions of the anonymous masses that share your tasts. But you can also have "buddies" and allow the program to figure in their specific tastes as well. It's interesting to see how different the results can be.

Tim Keller

You need to rank recommenders based on the number of others who trust them, calculated recursively ala Advogato (or PageRank).

I don't need to know anything about user X personally, just that a lot of others trust his opinions. And all I need to know about any of them is that a lot of people trust -their- opinions. It bootstraps itself.

Tim

Frank Ruscica

Given an efficient market, prices are the most informative means of ordering by quality.

The Long Tail will be supported primarily by advertisers and affiliate programs.

Ad rates and affiliate sales, then, are the prices of the LT.

So efficient search/navigation will require 1) availability of data about sites' ad rates/sales revenue, and 2) ad-bitrageurs, who will seek profits by buying and selling LT ad inventory, and, by so doing, will make price formation efficient in the transparent LT (i.e., given requirement #1 has been met).

All told, Open AdSense is the key...

Cui Bono? :-)

Oliver Morton

This is the obverse of the moral epiphany in "High Fidelity", in which our Hornby hero learns to value people for what they *are* like, not what they like.

Todd W.

I hesitate to say this, it sounding like "tagging will solve anything and everything", but perhaps a tagging system for categorizing your "friends" would solve the recommendation problem rather neatly by narrowing the domain in which a friend is useful?

Cole

I have to say/guess that this hypothesis doesn't hold true for probably half of the population in the US (especially with the younger crowd - young defined at under 27) because music, books, movies, and tv are all socio-economic indicators and filters for seeing whether or not a person is worth trying to be friends with (for the majority of people). I suspect that the crowd who takes the time to read a blog about the long tail is probably filled with early adopters who aren't just passive consumers, but the rest of the (young) world chooses their media tastes based on their existing group of friends. So if you make friends with people who watch "Friends" and listen to Matchbox20 and go to see Kirsten Dunst movies, then you will end up watching "Friends," listening to Matchbox20, and going to see Kirsten Dunst movies. Not many people take time to try out and internalize their personal media preferences any more; instead, they pick up what people around them are already using.

Take the movie Garden State for example. It did so well because its biggest selling point was that the soundtrack was chosen by the director who knew that he made a movie for people who are of the same media tastes as he is. He loves The Shins, Zero 7, Thievery Corporation, and Frou Frou so he knew that if people knew what music was associated with the movie, then the only people who would go see the movie would be people who would be sure to like the movie, which in turns makes more people go see the movie because it got only *great* buzz (I've never heard a bad review of it from someone who saw it... 55 yr. old movie critics aside). By knowing that people's media tastes are aligned with their friends' tastes, he picked up 98% of his possible market.

This is becoming truer and truer everyday, but I can easily see how it doesn't really apply to the "atomic" suburban family who has social interaction that isn't as fluid or spontaneous. When you start building your life around where you moved for your job or spouse, then you just have to be friends with the people you are stuck with.

Nancy McGough

I think that you would have a better understanding of social software if you use del.icio.us rather than Friendster as your touchstone. To me, social software does not inherently have anything to do with "friends." It simply has to do with a group of people using the same software, sharing some data, and making connections. And it can be good at filtering the long tail.

Nancy (who found this blog item through del.cio.us)

Sean

I agree with Nancy above...Replace "Friends" with "Contacts"...a contact can be a friend, a relative, or someone who is an authority on Sushi...i.e. social networks simply give you the ability to filter by taste. You can vote someone inside your favorite taste vertical, and make them more (or less) of an authority on that vertical. I like a sushi restaurant, I see who else does. Theoretically, I can trust them more to show me other sushi places...they certainly don't have to be a friend...?

Robert

Interesting post, but don't a large percentage of friendships form on the foundations of shared interest and taste?

mac

There's 2 types of friends recommendations the way I see it. One is based on what they like, and the other is based on what they know YOU like. Social networks like Friendster really only account for the first one, so it isn't very effective as you mentioned.
However, as mentioned above, there are other social networks more focused on shared tastes, rather than who you know, and provide something closer to the 2nd type of recommendation that is more useful.

James

Sometimes it works out. My top three 'musical neighbours' on audioscrobbler are my wife and two of my closest friends. Certainly I have plenty of other friends whose musical tastes I wouldn't want to use in filtering, but I would suspect that a fine-grained social network specification would be a very useful element for me in selecting new music. Social software _alone_ may not be useful, but as part of a package it definitely has a place.

Patrick Slavenburg

I think the problem is we try to categorize "real life". "real life" is hybrid, we have friends, contacts, relationships for a multitude of reasons, and these reasons change over time and thus change the relationship. Any type of format reminds me quickly of solving problems in my physics studies with very simplified equations (first order instead of say 10th order) and thus can give you an *idea* about what is going on but that is where it stops.

It seems a lot I see/read here is that: it gives you a vague understanding of the dynamics but doesn't really get to the point where on a detailed level you are able to get a good solution.

I think that problem cannot be solved and we may be much better off by either solving a small part of the total problem (focusing on one type of recommendation, in one field of expertise) or keep it "vague" and somehow let the human brain make the last few steps in a connection whenever someone is triggered by something.

In other words: create a pool of as many "triggers" as possible, which takes more time, but probably gives someone the best solution in the end.

Old Napster did a good job by being able to look into someone hard drive and check if you like what you see and make this person "important" to you.

Phil Wilson

Just to re-iterate what a number of other people here have said, a social network isn't necessarily just a set of people you've slavishly marked out as your friends, but people you're related to in a multitude of other ways; perhaps such as taste, lifestyle, expressed trust, and so on.

windy

I think that problem cannot be solved and we may be much better off by either solving a small part of the total problem (focusing on one type of recommendation, in one field of expertise) or keep it "vague" and somehow let the human brain make the last few steps in a connection whenever someone is triggered by something.

gifts for her

I feel surprise to read this article.I have recently used the new centre is a channel for the social computing community.It is the best softwer.I will do research on it.Thanks.

short answers

The truth is there are no shorts answers to this discussion. Social Software has its limitations, but the point is that we are sharing data.

The filtering and search can be done later. And better.

The comments to this entry are closed.

Tidbits

The Long Tail by Chris Anderson

Notes and sources for the book

FREE was available in all digital forms--ebook, web book, and audiobook--for free shortly after the hardcover was published on July 7th. The ebook and web book were free for a limited time and limited to certain geographic regions as determined by each national publisher; the unabridged MP3 audiobook (get zip file here) will remain free forever, available in all regions.

Order the hardcover now!