Meaning = Data + Structure: Inferring structure from user behavior November 19, 2007
Posted by jeremyliew in attention, data, semantic web, structure, user generated content.trackback
A little while ago I started a series about the structured web where I claimed that Meaning = Data + Structure. I followed up with a couple of posts on ways that structure can be added to user generated content, through user generated structure, and through inferring structure from domain knowledge. The third way that structure can be inferred is from user behavior, otherwise known as attention. As Wikipedia notes:
Attention economics is an approach to the management of information that treats human attention as a scarce commodity, and applies economic theory to solve various information management problems.
Alex Iskold has a good overview of the attention economy elsewhere at ReadWriteWeb.
By watching user behavior, by inferring intent and importance from the gestures and detritus of actions taken for other purposes, you can sometimes also infer structure about unstructured data. Google does this with its PageRank algorithm, Del.icio.us uses individual bookmarking to build a structured directory to the web, and Xobni maps social networks through analysis of your emailing patterns. Behavioral targeted advertising is based on the assumption that users display their interests through the websites they visit.
Using implicit data to infer structure requires making some assumptions about what each behavior means, but it can be a useful supplement to the other two methods of inferring data. As with inferring structure from domain knowledge, it requires a well defined ontology so that people and things can be mapped against it
Would love to hear more examples of using attention data to infer structure.
Lightspeed Venture Partners Blog 
[…] 1. User generated structure 2. Inferring structure from knowledge of the domain 3. Inferring structure from user behavior. […]
MyBlogLog uses repeat visits to a site as an indicator of fandom with regard to that site. Repeat visitors are made community members. It seems to work make readers happy to receive the “Welcome to so-and-so’s community email” very well.
1) Nytimes most emailed article
2) Related books/concepts etc… (customers who bought this also bought)
I would love this kind of functionality on my phone. Monitor who I call weekly, monitor who I call monthly, measure how long I’m on the calls, detect anomalies in my calling patterns and then provide a recommended calling list for when I’m driving from Sand Hill back up to the city. Maybe as Android opens up the mobile phone space this data will become more accessible. Right now the only way I’ve considered doing it is by scraping the call logs on my billing statement.
Hi Jeremy,
Sorry I missed you at Defrag. As you’ll see from my follow up on the conference, inferring meaning from attention streams was a large part of Defrag’s discussion – http://www.readwriteweb.com/archives/defrag_five_themes.php
Regarding other examples — as I mention in my R/WW post, at mSpoke we have developed some unique IP that allows our recommendation engine to take multiple attention streams and combines them together to infer a user’s aggregated intentions. We actually show the user these representations of their intentions (we call memes). A user can then adjust / remove any of these memes or add new ones we may not yet have discovered.
I agree the Data + Structure is an important approach. I also believe our system’s flexibility to interpret transparently multiple attention streams is really the next step in this evolution. Attention from one site is good in some cases, but insufficient in many more!
The easiest place to try our engine out is FeedHub (www.feedhub.com) a product targeted at heavy RSS users. In this case, the attention streams are both a user’s interaction with their personalized feed content as well as any other attention streams they choose to add. Currently we support delicous tags, digg votes, blog & link blog posts. However, support for more attention streams are coming!
Also, it’s worth pointing out our engine is also leveraged by a number of OEM partners, so this paradigm already has extended beyond intense Feed Readers.
Hopefully, we’ll connect on my next trip to the Bay Area to chat more in person.
– Sean
Last.FM uses listening patterns to structure their band focused radio stations.
I would actually love to see any convincing data shared from the industry about how effective behavioral targeting is or isn’t, or more to the point – how predictive past website visitation patterns are at a category level with respect to subsequent purchases of products in that category.
Then I’d like to see data around how _addressable_ that intent is – namely advertising data around clicks or conversions or something else that shows that people will actually respond based on the implicitly gathered information…
http://blog.wajsbrem.com/images/wf/index.htm
Ranking websites and sets of URLs by their number of bookmarks across multiple social bookmarking sites. Prototype only.
I am not sure user behavior = attention – at least not as a universal rule or statement. Attention is a loaded term and the idea of the attention economy seems a way to justify value and business models that might not be all that. Just because someone looks at something for any length of time doesn’t mean it actually has any business value.
That being said, I think essentially what we’re trying to ultimately determine is a user’s intent. There seems to be three ways that folks are trying to do it: contextually, psychographically, or demographicallly.
Psychographical cues – ie. a user’s interests – can be self-reported – ie. interests put listed on a Facebook profile or implicitly calculated ie. behavioral. From my experience the more generic the behavior (ie. visited a URL) the less accurate the implicit intent (based on a set of actionable goals presented to the user based on that implicit data) and the more specific the behavior – purchased a product, listened to a song – the greater potential for accurate estimation of a user’s intent. Last.FM uses music listening behavior to estimate music intentions, MyBuys and Loomia use purchase behavior to make purchase recommendations, MyBlogLog uses repeat visits to determine affinity for websites – all reasonably tailored & focused behaviors (and perhaps even a bit contextual)
It’s an interesting study in math and algorithms for sure. Conversely Lookery (where I work) focuses on self-reported demographic information to base targeting decisions on and Facebook is focused on self-reported interest data. Attention or Behaviors sometimes don’t do they job and its best to go on what a user just says about themselves. Who wants to guess when you don’t have to?
[…] Mining implicit data on behavior to create this structure is actually a better indicator of the real strength of relationships. Xobni does this via email; do any readers know of any third party systems that do this for social networks? […]
[…] Data + Structure, basé sur une structure créée par l’individu ; le domaine du savoir ainsi que le comportement de l’utilisateur, qui se focalisent eux sur le problème de la production de sens à partir du […]