A few days ago I read an interesting blog by Josh Porter at bokardo.com, on Facebook and Circles of Relationships. In it, he discusses how trust affects the relevance of the pieces of information we receive - honing in on the idea that information we receive from people we know has more inherent relevance than information we receive from strangers. I think “trust” is probably not the best word to describe what is being judged here - let’s face it, you probably trust a writer from the New York Times more than you trust your stoner roommate for certain things, especially factual information. But context, however, is a word that works well for describing what we need to judge the information we receive, especially for qualitative information, such as opinions.
Context
When we receive information from people we know, we have a lot more context for framing that information, than we do from someone we don’t know. I can temper any opinions I receive from people I know with my knowledge of their perspective, which is obviously harder to do when you don’t know the person sharing the opinion - see my post about user reviews on Yelp. At the same time, just because you know someone, doesn’t mean you share the same tastes in everything, in fact it simply means you are more aware of the ways in which you are similar and dissimilar from that person.
Josh’s post contained Ben Schneiderman’s Circles of Relationships, which illustrates a way to think about the groups of people in our network, relative to ourselves. Alex Mather’s answer to this, as it specifically relates to social networks, highlights an important segment, “People like us,” as well as showing how family and businesses have no place in the inner circles.
I think what is interesting here is not in the similarities between these two graphics, but in the differences - and the ultimate fact that each person’s social network is unique. The people we seek information from depends on what we know about them and what kind of information we’re looking for.
My Recommendation System
My way of thinking about the people in my network would look (and behave) something like this: (Click on “Movies,” “Clothing,” or “Concerts”)
With regard to receiving recommendations from people I know, the fact is that I have friends, family, colleagues, etc., who are nothing like me when it comes to certain things. It doesn’t mean that I don’t still want to be connected to them, and learn about what they’re interested in, but it does mean that I should be able to control how they’re likes and dislikes affects the recommendations I receive. I think the ideal recommendation system would be able to take into account my own understanding of how similar or dissimilar I am from the people in my network. A system like this, which is able to predict what I would like based on the likes and dislikes of people I know, might be called “collaborative micro-filtering.”
From Wikipedia: Collaborative filtering systems usually take two steps:
1. Look for users who share the same rating patterns with the active user (the user whom the prediction is for).
2. Use the ratings from those like-minded users found in step 1 to calculate a prediction for the active user
Collaborative micro-filtering would take four steps:
1. Identify people I know
2. Allow me to judge my similarity to each person, for a variety of topics
3. Make recommendations based on first and second order relationships between me and the people I know
4. Continually refine this similarity index for each person I know, based on our rating patterns over time
A specific example:
1. I know Jim, Joel and Christi
2. When it comes to movies, Jim and I always agree, Joel and I sometimes agree, and Christi and I never agree
3. Jim really liked Babel, *and* Christi really hated Babel, therefore, I will probably like Babel
4. Through continual data collection, maybe it turns out I’m more similar to Joel than I initially thought
A recommendation system like this would give me opinions of people I know, but within the context of how similar or dissimilar our tastes are. But this system could still factor the opinions of people who I don’t know in a relevant way. Perhaps if Jim and Christi had not yet rated Babel, I could still receive the recommendation, because users similar to Jim had rated it highly and users similar to Christi had rated it poorly. Additionally, it would still let me see clothing recommendations from Christi and people like Christi, because even though she sucks at movie recommendations*, I think her fashion sense is great.
*No feelings were harmed in the making of this post
May 29th, 2007 at 1:30 pm
Great visualization, Sarah! It shows how complicated these relationships are, and suggests how contextual they are and dependent on the individual. In some way we all know how complicated our own lives are, but I think we sometimes wrongly assume that others don’t share the same fate.
I was writing a follow-up to my piece, as several folks had the same concern you did…and then I found your post visualizing the issue very nicely. :)
May 29th, 2007 at 2:13 pm
sarah, awesome post and blog for that matter. my reactions:
1. i totally agree that “social networks” mean very different things to different people. further I think a single person sees different social networking sites as different ways to interact with their network and community. I know I interact very differently with my network and the community on Yelp, Myspace, the blogosphere, and even Friendster.
2. I think collaborative micro-filtering is a an awesome idea and Facebook, with its transparent API, has the best chance to facilitate it. I see lots of value in integrating Yelp, Netflix, iTunes, etc into Facebook and having a 3rd party put together a micro-filtering interface. I’m more in your age range (rather than in college) so I don’t use facebook that much (yet?) but it seems to be the only place where all of the pieces could come together.
May 29th, 2007 at 2:33 pm
Thanks Josh! Your Facebook post inspired me.
May 29th, 2007 at 2:33 pm
Alex - I see a lot of potential in Facebook, too, because of it’s focus on micro-communities - the Facebook API will only strengthen its ability to really mirror and enhance the way people connect in real life.
The big question I have now, though, is will it become the uber-network that everyone goes to? If you are able to access your Yelp, MySpace, etc., communities using Facebook, do you need to visit those sites anymore? Or is the idea of an uber-network just ridiculous, because like you said, you interact with each network differently?
May 29th, 2007 at 9:39 pm
I am totally on board with your idea. Your simple flash application touches the surface on the complexities of our online communities.
May 29th, 2007 at 9:47 pm
[…] Categorizing our communities 30 05 2007 Sarah Cooper posted a great response to the following series of posts regarding “Circles of Relationshipts”. The summary of which can be found here and here. […]
May 30th, 2007 at 7:42 am
Great concept and I love the visualization. For me, the problem with an ubersite is managing the range of identities we use in different contexts - I feel that it’s not so much that we interact differently on different sites as express ourselves as quite different people. Or will we simply have to become comfortable with an online expression of self which is much more comprehensive than our offline expression?
May 30th, 2007 at 7:55 pm
[…] Joshua Porter made a fantastic post yesterday that expanded upon his post the previous week, discussing circles of relationships and how each circle has a special significance to us as individuals. His follow-up post includes Ben Shneiderman’s original circles of relationships diagram, as well as a fantastic variation provided by Alex Mather and a beautifully dynamic Flash relationship example by Sarah Cooper. All of these examples try to categorize and map our trust relationships with others, but Sarah’s seems to document it best by showing how it changes with context. […]
May 30th, 2007 at 11:46 pm
Great post…very interesting. The application we’re preparing to launch…is similar in some aspects. Interesting take on everything…I’m definitely a reader of your blog now. Look forward to future posts.
Ryan
May 31st, 2007 at 1:59 pm
Michael - I have to say that the problem with most social networking sites is that they force you to express a public identity, and as such, you’re always going to have to make a lot of choices about how you wish to be perceived. It *is* like that in the offline world too… but its more organic and there’s much more room for grey area when it comes to expressing who you are.
June 3rd, 2007 at 11:46 pm
[…] Then Sarah Cooper published a small flash module that summed up the issue… well, in a flash. Go visit the live module and play with it – but essentially it demonstrates that the same people in your network can command different levels of closeness and trust depending on the topic. Click on a different topic, and the relative trust positions of the players all change. […]
June 6th, 2007 at 8:13 am
Thank you for creating a great visual for the dynamic relationships that inform decision. I’d like to add another dimension. Based on insight from my younger sister gained decades ago, one might value the opinion in another “context.” The conversation with my sister went like this: “Why did you ask Angela which dress she would choose to wear to the dance? She has no fashion sense.” “Exactly,” replied my sister.
June 8th, 2007 at 4:26 pm
[…] Micro-Filtering is a neat idea […]
June 8th, 2007 at 5:59 pm
[…] sarahcpr » Blog Archive » Collaborative Micro-filtering ..the ideal recommendation system would be able to take into account my own understanding of how similar or dissimilar I am from the people in my network…cp[and]… might be called “collaborative micro-filtering.” (tags: collaboration community socialnetworks collaborativemicrofiltering) […]
June 9th, 2007 at 1:46 pm
That’s an excellent insight into how online networks can better reflect our offline relationships. Great visualization too! I think you’ve added an extra step into how this could work though. Ideally the goal isn’t to identify who’s taste in movies you most agree with, but for the system to do it for you. Instead of this:
1. Identify people I know
2. Allow me to judge my similarity to each person, for a variety of topics
3. Make recommendations based on first and second order relationships between me and the people I know
4. Continually refine this similarity index for each person I know, based on our rating patterns over time
Why not do it like this:
1. I identify people I know.
2. The service gathers data from our reviews and other consumer behavior.
3. The service lets me choose a category, then displays the people with taste similar to mine, and can display recommendations based on those people’s choices.
3a(bonus power feature). The service highlights key points of agreement and disagreement in each category.
June 11th, 2007 at 10:18 am
[…] Interesting thoughts by Sarah Cooper on what she calls collaborative micro-filtering: in other words, who I turn to for information about a specific issue depends on how much I perceive that this person and I share (or not) opinions about X. She suggests that Facebook could take advantage of this to create some interesting tools. I think that expertise would also have a huge impact on my perceived trust in someone’s opinion. If Matt says that knife X is the best carving knife available, then I’m going to look favourably on mark X, just because I know that Matt spends a lot of time cooking and has taken classes in the subject. It’s not even necessary for me to share the same opinions on the end product with the expert: we could disagree on what would constitute an ideal meal, but that doesn’t mean I wouldn’t trust his opinion on the tools needed to create that meal. […]
June 11th, 2007 at 11:22 pm
[…] Another element to the recommendation filter may be built around trust…this is a meme going round started by Bokardo, and refined by Sarah Cooper…came across this via Green Chameleon. […]
June 12th, 2007 at 12:26 am
Jay - Your idea makes total sense, and it is essentially what NetFlix has done with their Friends application. I especially like how they use little quizzes to point out movies that we’ve disagreed on (he hated it, but you loved it!) Ideally, the service would be smart enough to calculate my similarity to people I know based on our activity over time. But since I am such a fan of explicit personalization too, I think it would be nice to be able to “turn up” recommendations from certain friends, even if the system deems that our tastes are dissimilar. It’s weird, but for some things, I don’t even trust my own taste. And I have friends who I consider experts in certain things, whose refined tastes in wine or art I might aspire to have. So even though I do think the system should do the bulk of the work, I still like the idea of having some control in this case, too.
June 12th, 2007 at 4:22 pm
[…] Collaborative Microfiltering (tags: sarahcpr.com 2007 at_tecp collaboration blog diagram filtragem_colaborativa) […]
June 13th, 2007 at 4:42 pm
so, one of the most popular facebook apps right now is “top friends” (http://www.facebook.com/apps/application.php?id=2425101550&b) which in a way tries to solve this same problem - distinguish your “24 BFF” from the rest of the herd.
June 15th, 2007 at 6:19 pm
[…] sarahcpr: Collaborative Micro-filtering Det behövs tydligare kontroller för att sortera rekommendationer från våra sociala nätverk. (tags: community den_sociala_webben circle_of_relationships visualisering facebook myspace filter sarah_cooper) […]
June 28th, 2007 at 10:49 am
[…] Pero lo mejor es el flash de Sarah Cooper que de una manera muy sencilla explica como estos círculos de relación cambian en función de la actividad o del interés que nos mueva en ese momento. […]
August 1st, 2007 at 11:58 pm
[…] Sarah Cooper posted a great response to the following series of posts regarding “Circles of Relationship”. The summary of which can be found here and here. […]