Social filtering for online forums

Yayhooray re-launched with new features and functions, and what looks like a rich environment for writing, browsing and discussion. As far as I know it’s the first forum built to use the buddy list as a form of content filtering: to increase the signal to noise ratio in the content.

Here’s a bit of Yayhooray history:

Built by skinnyCorp in 2001 as an experiment in online community. Along with o8 it soaked up some of the users from Dreamless, the ‘design forum’ that reached critical mass and became its own worst enemy at the end of 2000.

Originally it was built to manage itself through a levels system; allowing users to earn administration responsibilities (similar to implicit moderation systems employed by other forums like metafilter). It worked well at a small scale but led to cliques forming around the early adopter’s own social networks.

The levels system evolved into a points system, allowing anyone to award points to anyone, on a limited (one a day, one person a week) basis, similar to karma systems adopted at slashdot and kuro5hin. This briefly led to multiple account scams, and ended up in the ‘point orgy’ where ‘points were swapped rather than STDs’.

In the end, both systems were abused, subverted and widely discussed, often taking over from normal discussions and swamping the site with controversy. Many regulars left to other places, some seeing closed, invite only communities (like humhum) as the only option left for humane, creative discussion.

Yayhooray, in this latest version, is setting itself up to deal with these problems by globally filtering the content through a buddy system, rather than explicitly administering the content and user reputations. This applies to the entire site including the categorised discussions, blogging interface, links database, buddy lists and search.

The most obvious feature is a meter on the left hand side, which allows 4 different filtering settings:

  • you and only you
  • you and your buddies
  • you, your buddies, and their buddies
  • every user on Yay Hooray!

    This applies a filter to the entire site, including user lists and search, which took me a little by suprise. The site is effectively meshing off into small, interlinked communities of interest, based on individual social networks and collaborative filtering.

    In my case, buddies are mostly people that I have met, talked to, or seen invest time into making things: initiating photographic threads, dealing with social issues, administering creative collaborations, giving good design critique…

    Logging in now (using ‘you, your buddies, and their buddies’) I see a small subset of the overall forum, focused on these parts of the discussion. Given that the filter is so prominent and usable, it is also possible to jump out into the chaos of the full site.

    There is also a useful, if somewhat harsh, system that censors posts and links based on a list of people that you class as ‘enemies’! Being based on proper XHTML, CSS and DOM technologies means that censored posts are easily toggled on and off.

    On the downside there will most likely be confusion and clashes when different groups that don’t mesh with each other, but have completely different experiences of the place, come together in a single thread. There will also be more repetition, or double posts of content gets repeated amongst different groups that are out of sync by virtue of the filters.

    To fully appreciate this you need to invest time in it, and to build up a network of trusted buddies. YH can be hyperactive and annoying, it must be difficult for a new user to become engaged. The filters are perhaps most useful for long-time users looking for relief from ‘worst enemy’ problems.

    Because it has become an adaptive social platform, and has the potential to be subverted and shaped into many different kinds of system, I will reserve judgement for now, and make a new report soon.

Information architecture books

Information Architecture for the World Wide Web

Louis Rosenfeld, Peter Morville. Now in its second edition, undoubtedly the best introduction to IA.
amazon.co.uk / amazon.com

Information Architects

Richard Saul Wurman.
amazon.co.uk / amazon.com

Practical Information Architecture

Eric L. Reiss.
amazon.co.uk / amazon.com

Information Architecture: Blueprints for the Web

Christina Wodtke.
amazon.co.uk / amazon.com

Mapping Web Sites

Paul Kahn, Krzysztof Lenk. A suprisingly badly designed book, but packed full of very inspirational information architecture examples, from isometric site-maps to flow charts and wireframes.
amazon.co.uk / amazon.com

Dynamics in Document Design

Karen A. Schriver.
amazon.co.uk / amazon.com

Content Critical

Gerry McGovern
amazon.co.uk / amazon.com

The Web Content Style Guide

Gerry McGovern
amazon.co.uk / amazon.com

Web Navigation

Jennifer Fleming.
amazon.co.uk / amazon.com

Sorting Things Out

Geoffrey C. Bowker, Susan Leigh Star.
amazon.co.uk / amazon.com

The Intellectual Foundation of Information Organization

Elaine Svenonius
amazon.co.uk / amazon.com

Laws of the Web

Bernardo A. Huberman.
amazon.co.uk / amazon.com

Infosense: Turning Information into Knowledge

Keith Devlin.
amazon.co.uk / amazon.com

The Social Life of Information

John Seely Brown, Paul Duguid.
amazon.co.uk / amazon.com

Mobile interaction design case study

Pollen Mobile develops location-based services for the consumer and business markets. Mamjam is their first product: a location based, social entertainment service based on Short Messaging Service (SMS) messages. It enables people in the same venue to chat with each other by sending text messages from their mobile phones. Mamjam interface: user match screen 1 Mamjam interface: user match screen 2

Brief

Pollen approached us with a very broad intention to use SMS to drive social interaction and entertainment in new ways.

We initially developed three quirky ideas based on playground games, internet chat, and community storytelling that we presented as the basis for discovering business goals and user-needs.

After our initial brainstorm, we initiated a more rigorous user-centred, interaction design process that is detailed in this case-study.

Handsets & Networks

We found several pivotal issues we needed to resolve: SMS has extremely limited functions; with few opportunities to create rich, engaging, extended interactions.

Handsets

Mobile phone handsets provide no navigation between multiple messages, no indication of user status or location, and have no practical means of viewing session history. Users are accustomed to using SMS for quick functional communication, and extended contact with friends. They certainly do not rely on messages for any kind of complex interaction.

Every transaction between user and server on a mobile phone is a sessionless operation. Each message contains only the time it was sent, the number it was sent from and the content of the message [1].

Unlike http systems, the server cannot rely on location and session information being stored in the message address. This is complex from a user experience perspective because people are used to responses exhibited by systems that do carry session information and behave quite differently [2].

Networks

SMS messages are managed by the networks with cells, each cell carries messages particular to that region. Cells are notoriously unreliable, and we found that it was common for messages to hang in the system for over ten minutes. This presented some serious problems. Satisfying communications rely on a high level of continuity, and the timing between messages is a critical indicator of the emotional state of your chatting partner.

Mamjam’s service is location based: users are in contact with other users in the same area. However the existing (second generation) handsets cannot determine location, and although locations are triangulated by the network, this information is not publicly available. The location thus had to be manually provided by the user; in a way that then could be usefully interpreted by the server.

Researching and developing a reliable language for users to identify their location became central to the interaction design problem.

Many competing SMS services are currently internet-based: requiring a signup for services from a web site, rather than directly from handsets.

Modes

A system like this could conceivably be built without the use of modes [3]. From the users perspective a modeless system could be overly complex and exhausting: every message must somehow include exact commands and instructions for the server. But a modeless system is very attractive from a technical perspective: the server is more likely to correctly interpret instructions.

Process

Requirements

We consulted with Pollen and selected SMS users to draw up several personas and scenarios. This included contextual enquiry, business goals and user-requirements gathering. We identified the following requirements:

  • Users must be able to join the service immediately, not just from a website prior to use.
  • The service should accommodate both new and returning users.
  • Users are likely to be exposed to the service through all sorts of channels, and therefore signing up should accommodate all points of entry.
  • The structure should be designed to accommodate expansion of the service.
  • The basic structure of the handshake should carry to other SMS systems Pollen may choose to develop.

    First Iteration

    The initial interaction architecture outlines our first intentions for the system. (For legal reasons we can’t include the full size diagram.)

    mamjam interaction design version 1

    The system works in a similar way to internet based chat rooms, connecting users who are ‘online’ at the same time, with the extra dimension that they are in the same physical place. Mamjam supports private, one-to-one communications only: users can’t shout to groups or broadcast messages. Once a user has found a chatting partner the system simply directs the text traffic between them until one party decides to pursue some one else, or signs off.

    This structure required users to enter a lot of information about themselves before they could initiate contact with one another. We felt this was valuable in order to reduce the interaction load while chatting. This also resulted from a (perhaps misguided) adherence to the ‘internet chat room’ model.

    This system was implemented on Pollen’s test servers, and we organised user-testing sessions. This revealed several problems:

    The sign up process was off putting. Users motivation for this product is for entertainment and social contact: they weren’t happy to tolerate a lengthy sign-up process. This architecture required four messages for a new user to sign up. In some cases the user would be spending the equivalent of a 10 minute voice-call before they had connected with someone to chat. It was clear that the service needed to offer a quick method of signing up, perhaps at the expense of more advanced features.

    In trying to optimise the system for both new and advanced users; signing up for the first time required a different interaction process from signing up for a second time. There were also several different methods of identifying your location to anticipate every possible user-interaction. There were thus four or five possible entry points into the system. This caused more modal problems than anybody anticipated; the SMS server had to process language and match patterns in an almost infinite realm of possibilities.

    Second Iteration

    It became clear that the three biggest problems for the social interaction process were:

  • Aligning the systems perception of user-context with actual user-context.
  • Ensuring users have an accurate perception of the system state.
  • Maintaining a rich connection between users, allowing them to interpret and react to one another accurately.

    This discrepancy between user perception and system perception can be referred to as ‘slippage’. Slippage is most problematic during the initial handshake when the user is most insecure about their request and about the system itself.

    Text messages to and from SMS servers rarely arrive as punctually as they seem to in normal use. This meant it was possible for one of two users, both having agreed to start chatting, to reject the other on the basis that they had failed to reply to their confirmation. In fact the rejected user had replied with confirmation, but their message had been delayed. The message would then arrive with the first user who had since moved to a new part of the interaction process. Their reply could potentially interrupt another process or get lost in the system, confusing and infuriating both users. Serious slippage!

    We also found, as predicted, that users did not read back through their old messages. Some phones have a very limited capacity for storing messages and no phone facilitates simple navigation of previous messages, so the current message was the only one through which we could usefully rely upon for users to react to.

    mamjam interaction design version 2

    The second interaction architecture was developed with the problems described above in mind. Some changes have been made to the system since, mostly around modal issues, and the commands through which users communicate with the server. Although there are still issues regarding slippage, the second iteration makes this much less of a problem. The system is basically modeless, except for the first transaction. All users (new and existing) enter the system in the same way, new users are chatting within two messages, and existing users are potentially chatting after their first message.

    For an overview of the commands and interactions possible with the system look at the Mamjam How To and Advanced Features.

    In Use

    Mamjam is now fully operational, spinning off other services based on the basic interaction architectures we designed for the initial chat service.

    Extended Services

    In a recent, typical promotion, at the Mood Bar in Carlisle, Mamjam sent a message to people who had Mamjamed there, offering them a discounted drink if they showed their mobile at the bar. The conversion rate from message sent to offer redeemed was 30%.

    Building relationships, Community and Storytelling

    Having heard that a large number of people were texting their ex-partners late at night; under the influence, Mamjam sent a message asking for their own dating disasters. 13% of people responded with their own story by SMS; 50% of those responding within the first hour.

    These users were not given incentives like promotional offers, the call to action was not a simple generic mechanic like reply YES or NO; it was much more involved. Users were required to read and understand the message received, then conceive and craft a response to fit into 160 characters. Yet the response was high and the quality of response excellent.

    Stimulating usage

    By reminding BT users of a free messaging offer, the objectives are to stimulate Mamjaming outside of the locations in which they first Mamjamed.

    Message: Spice up your text life for FREE! Mamjam is still FREE to receive for BT users. To chat now just reply with mamjam and your location eg MAMJAM LONDON.

    7% of the database of BT users read the message, and then decided to log on to Mamjam. Between them on that day they sent 3,400 chat messages.

    Some usage statistics

  • First time Mamjam users begin chatting by sending only 2 SMS messages.
  • Users are matched with someone within 120 seconds of logging onto the service for the first time.
  • The average Mamjam user sends and receives 24 SMS messages per session.
  • The top 10% of users send 60 SMS per month and generate an additional 72 outbound messages. Generating an additional 18 for the network operators.
  • The top 50% of users send 20 SMS per month and generate an additional 24 outbound messages, generating 6.30 of revenue for the network operator.
  • Repeat usage: 30% of daily users are repeat users.

    Conclusions

    We think that the best solution to this particular service has been found, given the limitations detailed above.

    There are obvious and not-so-obvious limitations to SMS communications. The most notable limitations are the handsets continuing to rely on short messaging rather than a more advanced chat service, and the network operators inability to develop services and platforms outside of their own internal structures.

    This research and product development has generated a lot of further ideas for asynchronous communication structures, and communication solutions for packet switched networks for mobile devices.

    Footnotes

    [1] Some phones support greater functionality than others, Mamjam needed to support a broad demographic so only the most bottom-line functionality was available to us.

    When sending a message from a server it can be set to “Flash” mode, causing the message to open in the users phone immediately. Some cells also support a “broadcast to cell” function, whereby a single message can be sent to all phones within that cell. This function is expensive and only available to phones on a given network. back

    [2] Information transferred with HTTP is also sessionless, but browsers and servers are afforded with functionality to help them overcome modal issues, like cookies, history bars and links for example. There are other interface restrictions to consider regarding the manipulation of text like the absence of cutting and pasting for example. back

    [3] The most comprehensive discussion of modes I have come across is in The Humane Interface by Jef Raskin , pp37. back

    References and Links

    At the time of writing the Mamjam numbers are 82888 (BT/Vodafone) or 07970 158 158 (all other networks). Just send any text message to sign up and test it for yourself.

  • Mamjam website
  • Pollen Mobile
  • Mamjam reviewed at The Guardian
  • Jef Raskin ,”Modes 3-2” The Humane Interface , 2000, pp37.

    Professional Credits

    Interaction design

    Jack Schulze Adi Nachman Timo Arnall

    Technical Architecture

    John Gillespie

    Information Design

    Jack Schulze