At lunch today, Marc Smith (Microsoft Research Community Technologies Group) gave a very interesting presentation on the work his department is doing in researching the use of online communities. Citing Usenet he pointed out that anarchy does not equal chaos. In fact, you actually see order without authority.
He also pointed out that online communities lack many of the visual signifiers that we take for granted in the real world. When we walk into a crowded room, we immediately get a sense of how many people are there, what sex and age the occupants are. In the virtual world, we get none of this.
In a collective commons (ie a virtual un-moderated community) reputation systems are vitally important. Marc talked about Explicit and Implicit reputation systems.
Explicit systems make statements about behaviours and relationships ie. the systems used by Ebay, Amazon & Slashdot. Problems with such systems are that the ratings are misleading – 20 positive comments can be worth as much as 20,000. They are also slow to react.
Implicit systems base reputation on observed behaviors ie. Google, Amazon, Blogdex, Technorati. Google being the best example – a page’s reputation is improved by the number of pages linking to it. There is no sense of why it’s being linked to. Saying ‘this is the worst page in the world’ is the same as saying ‘this is the best page in the world’ with Google’s reputation system – any link improves the page’s rating. Hence, implicit systems are full of ambiguity – observable behaviour is not always an endorsement. Marc suggested that an 80/20 system would be good – 80% implicit/20% explicit.
Marc also suggested that in terms of implicit recommendations, that ‘Email this page’ and ‘Print this page’ were very good signifiers of user approval.
The most valuable community members are those who post to few groups, regularly. Analysis done on usenet stats (and searchable on Microsoft’s Netscan site) reveals that around 2% of contributors post daily. Out of many million users, that’s still a sizeable community helping out others and giving their time for free.
Netscan essentially gathers meta data about communities which can be used to assess their ‘health.’ For instance, are they active and growing, with regular, trustworthy contributors?
Using this data as a source, Marc showed a very interesting example of Tree Maps (a space-filling approach to the visualization of hierarchical information structures) created by Ben Shneiderman, to make visual, usage patterns over time.
(Ben has also written some very interesting papers on Information Visualisation.)
Such data visualisation for instance made it easy for Marc to show how usenet groups in China have increased with incredible speed over the last 18 months.
We also saw other visualisation techniques (bubble and piano roll) that enabled a user to be classified by assessing their postings over time. Spammers, Regular Posters, Those who only show up with questions could all be identified from a visual representation of the statistical data.
As Marc pointed out, these ‘measurement tools actually become evaluation systems’. As a casual visitor to an online community, access to this data allows me to (at a glance) evaluate the trustworthiness of an individuals posts based on their posting history. ‘Usenet Views,’ the tool used to create these visuals should be made public soon.
Marc demonstrated work by Gina Venolia on thread visualisation, or ‘the social life of a conversation’ From these visualisations, UI components have been created that allow better tracking of threads in an interface than the conventional ‘Explorer’ metaphor. These UI components are freely available from microsoft for non-commercial use.
Finally, Marc talked about AURA (The Advanced User Resource Annotation system (A.U.R.A.) – designed to provide the ability to access and author annotations on objects and places using machine readable tags). Put simply, it turns the world into a webpage. Currently, barcode readers are fitted to handheld devices, enabling a shopper to scan an item in a store, and immediately read usenet (or equivalent) posts, reviews and recommendations. Useful examples were given in the case of identifying foods with GM ingredients, or product recalls etc. In the extreme, giving users access to information such as ‘this will kill you’ because it has been discovered to contain peanuts.
The interface not only allows users to scan items, but to post comments about them, and make them public if they wish. Essentially allowing any consumer to connect with an online community about any object that can be identified.