John Dempsey
Development of useful social discovery algorithms and tools
Contact Information
Programmer-writer
Office Information Bridge Framework
Biography
After promoting the emerging Windows platform with ISV’s, I
left Microsoft in the early 90’s to join the development team that produced
RealPlayer client. In 2004, I developed LJMindMap, online social network
visualization software operating on the user graph of LiveJournal.com. The
project has deployed 50,000 user-centered representations of community space. I’m
currently a programmer-writer for Office Information Bridge Framework (LOBi).
Position Paper
As developer of successful community structure visualization
software in the largest centralized blog network, my current objectives center
around development of useful social discovery algorithms and tools. This
includes the theoretical challenges around algorithm development, as well as
technical challenges of advanced data acquisition, browser integration
techniques, and application of richer emerging platforms such as .NET
Framework. Much enthusiasm still surrounds “pretty pictures” of social network
mapping. A researcher I follow closely said of the phenomenon: “I see ample cases
of visualisation addiction: excitement is present, but no one really knows what’s
so good about the fix.” My challenge is to move from a product of transfixing
but largely entertainment value to a functional social discovery toolset. Toward
this goal, I’m testing “edge suggestion” algorithms, advanced data collection
techniques (server log analysis, response pattern analysis, change-over-time
techniques), as well as smart integration of community structure intelligence
into the “blog” experience.
My implementation may be considered novel for its operation
as a viral marketing meme. The product is its own advertisement, and since the
subject is an individual’s immediate social structure, the distribution model
usually lets new people glimpse the value and motivates participation. I have
avoided features and formats that can’t be distributed on the most basic
browser running on the most basic hardware platform, while also trying to
maximize value within the basic browser feature set.
My experience exposing 50,000 individuals to visual
representations of their personal online social structure has brought these
conclusions: (1) Curiosity and demand for this kind of product is high, even
for entertainment value alone and lacking any additional function. (2) While
some people have an elaborate intuitive awareness of what they are seeing, some
have no context to understand what they might be seeing, even when content
defines pronounced trends. This suggests either a variation in appropriateness
of algorithm suitability for any given individual, and/or a variation in
individual aptitudes for conceptualizing community structure, even in the
context of elaborate networking activity.
Additionally, I have these observations: (1) Some social
networks are remarkably large, with massive edge densities, but I not found a
simple explanation for this, (2) Simple connectivity graphs often fail to
represent social structure as the subject may understand it, depending on
appropriateness of algorithm used. Blending multiple datasets and algorithms
can provide much more nuanced and perceptive visualization tools. (3) Blending
modalities (color, motion/time, position, font size/iconic representation) can
overcome a desire or expectation for pat, literal interpretations, and synergize
patterns between multiple datasets without algorithmically identifying them
explicitly. (4) Even a small contribution of additional user-provided graph
information (self-description of “my online clique” for example) can
substantially augment algorithmic analysis and inference quality.
Back to Social Computing Symposium 2005
|
