Description
Q1. (Note, for this question you need to form a group of three to four people, any
smaller or larger is not acceptable.. Form a group as soon as possible. If you
do not have a group you will not be able to complete this question.)
You are working at a hot tech company, and your R&D department has just
developed a new brain-computer interface. The device reads how people are
feeling and is small enough to be built into a laptop. It can distinguish between
four states: happy, sad, frustrated and bored. Unfortunately, it doesnt work
perfectly. They give you a confusion matrix of how well it would work.
predicted happy predicted sad predicted frustrated predicted bored
actually happy 80% 5% 5% 10%
actually sad 5% 60% 30% 5%
actually frustrated 5% 25% 65% 5%
actually bored 2% 2% 1% 95%
a.(group) Design a product that leverages this new brain-computer interface.
In 1-2 paragraphs, describe your target audience and your project and a high
level idea of how it works.
b.(individual) Each group member must come up with two storyboards that
describe the experience of using your product, each storyboard should have one
paragraph explaining it. Your product must be complex enough to have enough
meaningful storyboards.
For each storyboard, pick a primary task to be done with your product – make it
a task complex enough to fill up a storyboard. Each storyboard should require
5-8 panels, so in total you will have 15-24 panels to turn in. Each storyboard
should fit on two 8.5”×11” sheets of paper and be drawn with a thick pen like a
Sharpie. Using a thick pen limits the amount of detail that you can add, forcing
you to only draw the most important elements of scenario, user, and interface
that communicate your ideas. Clarity, communicativeness, and innovation are
1
more important than aesthetics here.
Your storyboard doesn’t need to include the minutia of your interface unless it’s
important to the task and what is novel about your interface (e.g., dont show
people clicking on File → Open). Your storyboard should have enough detail
(e.g., dont say Bill was unhappy then he used our awesome tool and now he is
happy). You will be graded based on the criteria shown in table 1. Here are
some guides to understanding storyboards:
http://www.usabilitybok.org/storyboard
http://hci.stanford.edu/courses/cs147/2009/assignments/storyboard_
notes.pdf
c.(group) Evaluate the designs of your colleagues using peer analysis techniques
discussed in class. Have team members come up with questions about your storyboard and your product. For your interface pick 5 questions from each team
member and write 1-2 sentence answers. As a team decide on a joint design,
and repeat the process in a) as a group this time presenting three storyboards
for the group.
d.(group) Your R&D department tells you that if you can collect data from
the user, it would increase the performance of the system so that it is nearly
flawless. You need one minute of brain signal for each state. How would your
teams design change? What are the tradeoffs? Add a storyboard and write a
couple of paragraphs describing the change, if any.
2
Criteria Guiding Questions Bare Minimum Satisfactory Above and Beyond
Do your storyboards Task is vague, Storyboards Task is unique
clearly communicate or ill-specified. communicate and addresses
a user’s real Storyboards do an authentic a real need.
Task choice problem or need? not demonstrate need and the Storyboards clearly
(20%) Convince us that the need for task effectively. convince reader of
this problem needs such a task. the task’s authenticity.
to be solved!
Do your storyboards Little variation Storyboards show Storyboards demonstrate
communicate multiple among each significant deep thought
significantly different storyboard of variation in about multiple
aspects of your either interface interface or design alternatives,
Design product? Do or scenario. scenario. Utility of designs
alternatives you demonstrate Designs do Designs solve is shown
(40%) how your idea not convincingly problem to clearly & elegantly.
solves the user’s accomplish the a degree.
problem or desire? task at hand.
Generate as many
as you can
and show us!
Are your design Storyboards Storyboards Illustrates ideas
ideas communicated poorly communicate intelligently, focusing
clearly? Are the communicate design ideas on important
important aspects design ideas. effectively, using scenarios and
of your interface Lacks key a solid mix interface elements,
illustrated? Do elements necessary of illustrations Relies less
Clarity your storyboards to establish and words to on labels for
(40%) give a decent scenario and focus on explanation.
understanding of design solution. key elements
how your interface of story.
words? We
are not looking
for artistry,
just good
communication!
Table 1: Criteria for product design
3
Q2. This is a simulated social network dataset of 10000 users (think Facebook or
Google Plus). There are 10 files in total. You can find them in CourseWorks.
These files provide snapshots of the network from Monday (2013-10-01) to Sunday (2013-10-07) in 7 csv files respectively. Each file contains the number of
users visiting the social network site, the number of posts, time spent on website, and new friends made on that day (if users didn’t visit the site that day,
then there is no record in the corresponding file). There is a csv file about users
profile. More details can be found in “README.txt”. There is also a big csv
file (a 10000×10000 matrix) representing all the friend links between the 10000
users. If users i and user j are friends, then the (i,j) entry will be 1, otherwise
it’s 0. Note that the matrix is symmetric.
(a). Join the 7 csv files on the users actions. Note that most users don’t visit
the site everyday. You are supposed to calculate the total number of visits, the
total number of posts, the total time spent, and the total number of new friends
made during this week for each user. You should also extract the total number
of friends for each user from the friendship csv file. Add all these statistics into
the users’ profile dataset. The profile should be finally be a data frame of 10000
rows and 15 columns.
(b). Perform a basic exploratory data analysis to better under understand
this social network. Here are some questions you should answer/guidelines for
exploration:
• What’s the distribution of the number of friends?
• What’s the relation between the number of friends and age?
• Does age affect whether users move away from their hometown to live
another city?
• Characterize the relation between age and relationship status.
• Draw scatter plots between the number of total actions, visits, posts and
time spent.
• How does the sign up date influence the number of new friends they made?
• Is there any relation between age and users’ actions?
• Do earlier users have more friends?
(Some reference to help with plotting in R: http://flowingdata.com/category/
tutorials/). Keep a log of your exploration and turn it in as part of your assignment.
(c). Think of the social network is a graph where each user is a node and a
friendship is a edge between nodes. An interesting character of a graph is the
average degree. Let G be the graph, N be the number of nodes, and di be
4
the number of edges that node i has. The average degree of the graph can be
computed as follows:
η(G) = 1
N
X
N
i=1
di
For a graph of 10000 you might be able to compute average degree. But for real
social networks the number of users (i.e., N) is more like 109
instead of 104
.
It is then computationally infeasible to get an exact answer. However, we can
sample. Let’s consider two sampling methods to estimate average degree. In
both methods, we begin with a uniformly random sample without replacement,
let’s call this subgraph G˜.
Method I: for each sampled user, we observe all the edges directly linked to that
user.
Method 2: an edge is observed if and only if both nodes of that edge have been
sampled.
Based on the subgraph G˜ we sampled, calculate the estimates using the same
formula above. Take the sample size as 3000, repeat this process for 10000 trials.
Draw the distributions of the two estimates (using histograms). Also calculate
the true average degree. How do the two estimates compare to the true average
degree? Which sampling method is better? Why? Can you come up with an
easy adjustment to correct the worse one?
(d). Based on your analysis in (b),
1. build a model to predict the number of total actions. Begin with a simple
model and build up to make more complex models by adding additional
predictors, interactions between predictors and transforming your predictors. Use cross-validation to evaluate the models.
2. segment your users based first on reasonable heuristics and then using
k-means. Try varying k=2,3,4,5. What is the optimal k? Why? Find a
way to characterize your clusters using summary statistics and/or visualizations.
3. Once you’ve settled on your user segments, count the number of edges
within each cluster of users and between each pair of clusters. Now imagine
you didn’t have access to the dataset that contains user profile information,
is it possible to detect these same clusters of users using only the friendship
matrix?
Q3.(individual, no teamwork) Look at the blog post on Data Products in the
5
Wild: http://columbiadatascience.com/2013/09/29/data-products-in-the-wild/.
In the comments section, respond to the question at the end of the blog post,
or respond to the responses of any of your classmates. Write 1-2 paragraphs.
Q4.(individual, no teamwork) Getting started on the Kaggle competition:
• Go to our inClass Kaggle competition and download the data.
• Submit your first individual entry
• Document your process and thinking.
• Tell us what user name you are using
